/* this is gross and I hate it but it works to fix 
   Safari’s layout of the table’s top headers */
@supports (font: -apple-system-body)

Gross, and horrible, but 🙏

Source: meyerweb.com

Originally published on Remy Sharp's b:log

It was like that from the very first web site - the logo was designed in early 2009 in 12 variations (which you can see from years 2009, 2010 and 2011 before they were really redesigned).

Before I (inevitably) forget, it made sense for me to write up some of the things I learnt along the way now that the 2026 web site is live.

I think Chris Mahon's did an excellent job with his translation of my mood board into Figma pages.

The spec this year was a fanzine style, which is a very tactile paper based medium, which doesn't always translate well to digital. The actual visual impetuous came from the (very fun) Running Man remake, the credits in particular:

The CSS

It's very likely a lot of this (clip-path for instance) have been around for ages. But my client work has been almost exclusively on API backend design for the last 4 years or so, which means FFConf is usually the only time I get to work in the front end to a specification.

Vertical text

I had been wrestling with transform rotates and origins, and had the text sliding all over the place.

Ana Tudor came to my rescue and took away all the pain in a simple single line: writing-mode: vertical-lr. An absolute slam dunk for getting text along the side.

Ana's other tips

• inset: 1 can be used as a shortcut instead of top/left/right/bottom: 0 when stretching across the page.
• I don't need to do transition: rotate(180deg), rotate (and others) are first class now, so rotate: 180deg works (less bytes too).

Polygons

I'd not used polygons in CSS yet (didn't have the use case), but due to the cutout nature of the speaker photos, and the proximity to the talk titles, I quickly realised (after failing with margins) this was the way to go.

• shape-outside floating the element and having the text shape around the element, in particular the cutout of the speaker photos
• clip-path letting me create the sharp angle shape on the footer
• Firefox interactive polygon devtools

I would primarily develop in Chrome Devtools directly using workspaces, but Firefox offers interactive polygon editing which was immensely useful.

Contrast issues

Due to the effect of the "paper", I didn't want to make all the text partly opaque, so instead I lay the paper (which is mostly transparent) over the entire page (with pointer-events: none).

This did mean that there was some grey flecks of colour over black text which definitely caused contrast issues.

In skimming the rendering options in Chrome's devtools I found the "Prefers Contrast" (I'm not sure if I knew about it already or not).

Using @media (prefers-contrast: more) allowed me to remove the paper effect and keep a striking black and white design.

But what if the user wanted control? I added a labelled checkbox and found that instead of using the @media query I could use @container style(--high-contrast: 1). All the modification styles are nested inside of the @container query.

Then the --high-contrast is toggled using:

:root {
  --high-contrast: 0;
}

@media (prefers-contrast: more) {
  :root {
    --high-contrast: 1;
  }
}

body:has(#high_contrast:checked) {
  --high-contrast: 1;
}

Bonus bits

I hadn't realised I could style the ::placeholder pseudo element (not sure why I thought I couldn't).

Lastly, a lovely win for me, I discovered nth-child(odd of <selector>) 🤯. I was trying to set the speaker blocks with alternating position of their photo (which I'd select with odd and even), but there was an extra element between each speaker.

In reviewing the MDN page I discover the of syntax, and it let me target exactly the way I need using: &:nth-child(odd of .speaker-card).

Originally published on Remy Sharp's b:log

<div id="meter" data-props-for="range">
  <input type="range" min="0" max="100" value="42">
</div>

With CSS like:

@container style(--live-value: 100) {
  .gauge__num { color: var(--max-tint); }
  .gauge__flag::after { content: 'max'; }
}

Lots of useful categories too.

Source: prop-for-that.netlify.app

Originally published on Remy Sharp's b:log

I wanted to pull a few quotes:

The Internet Watch Foundation reported that child sexual extortion cases in the UK rose 72% in a single year — criminals tricking young people into sending nude images, then blackmailing them. In 97% of those cases, the victims are boys.

Sacha also spots this:

Buried in the coverage is the acknowledgement that all adults will need to verify their age if they want to take or view nude images on their own devices. [...] You’ll have to prove you’re an adult to own and use a phone.

On one hand, to own a phone requires (since the availability of mobile phones) proof of being an adult, because to get a contract of any sort you had to have a bank account. This obviously changes with "burner" phones (a £10 job from ASDA) and a PAYG sim.

On the other hand, who am I proving I'm an adult to - which walks a weird line of privacy invasion.

Quoted within Sacha's article is Signal's response, and exactly what I was thinking when I saw the announcement too: isn't this at odds with the UK wanting to take some kind of lead in international AI?

Child safety looks like well-funded education, robust social services, and meaningful guardrails on the very AI technologies and platforms the current government is eagerly courting.

An excellent post well worth the review, regardless of parent status, it's going to affect us all.

Source: newsletter.sachajudd.com

Originally published on Remy Sharp's b:log

Admittedly, it's imported from my prior blog and the first time I posted on on this domain was in September, but the post in June counts! But then, my domain was registered a full year before on 13 June 2005 (which neither myself, nor the Internet Archive can remember what was on it!).

I could probably pull up the best hits based on traffic (though my Google analytics were stripped years ago, so I'm not sure how I'd work it out). But I'm not sure the highest traffic really represents the hits to me.

Instead, I'm going to reflect on the last two decades of posts and comment on how what I wrote about has changed.

In the first couple of years of my blog, I was posting regularly throughout the month. Even publishing twice a day.

The posts were a mix of movies I had seen and my excitement in sharing jQuery with non-devs (ie. designers, who back in the mid-2000s didn't have the same exposure to front end code as they do now...I think).

I'm not if it was that the mid-2000s the front end for logic wasn't considered serious enough or... something else, but I always feel like the atmosphere has changed a lot over that time. React has somehow let the trolls out from under the rocks and there's a constant tension around the front end.

Still, my posts in those first 5 years wanted to share technical solutions as simply as possible to newbies and non-devs to show there was a way that didn't require starting your JavaScript journey in 1994.

A personal place

In 2010 Julie and I lost Tia to stillbirth. After a full disconnect from the internet, I tentatively re-entered using my blog as the way to share what had happened and to share our thanks to the overwhelming love we had had from both friends and the community.

Since then I've always used my blog to reflect on another year passing without our girl on, or near, her birth date.

From 2008, I would also write a round up of my year. Something that's celebrated throughout the web blogging community each year around New year (not my blog post! All of yours! Silly thing).

This would also serve as an opportunity to share how our family has grown. Over the years a lot of us (you and I my dear reader) have gotten to chat in person, so it's also fun to hear comments on how our kids have grown. They're a hugely important part of my life, so those comments make me happy.

Ch-ch-changes

Just like a house you've no intention of ever leaving, my blog has had various renovations.

I added my reading records, making an effort to (attempt to) review the books.

I added a short lived Things I Learnt. And a newsletter section, which lasted around a year. And a Things I Fixed (as my experience in hardware grew).

The links section still lives, and probably inspired by Jeremy's links section.

I also added an attic (really, it's a loft, but it sounded more impressive as an "attic") of designs and my ethos page, which is like lifting the lid on my brain and seeing the ideas I try to live by.

In the last 5 years the momentum has proven difficult. The pandemic lockdown cut off a lot of my community connection (through events in particular), and as the years rolled on I've found less and less interest in the web.

Obviously there's AI playing into this. I've written a little about it, and have strong opinions (not all obvious) and I've enjoyed making a lot of personal, offline, projects.

But I don't feel like I've anything to add to the discourse on the web. Maybe I will again one day. I'd love to have the excitement I used to have, but these days, it's just not there.

So my blogging slowed down. It's evident in my archive, and you can see the topics drift away from anything practical and as I got older it's more idea and opinion based. Lol. Like this post!

Your contributions

From day one, I had comments on my site. Those posts that attracted more visitors tended to draw more discussion, which, I mostly approved of.

I had ported the comments from WordPress to Disqus, ejected Disqus when I found they were adding Facebook tracking and landed on commento.io.

At some point during this year (2026), Commento just dropped off the web. First it was hanging requests (which is how I spotted it) and now it's just a 404 holding page. Along with dropping off the web, all the comments you've all left over the years went with it.

A constant side effect of the web: entropy. The lesson there: own your stack if you want to last.

Stick around for some more?

Honestly, the idea of blogging for another 20 years really doesn't sound realistic to me at all! I'll be in my late 60s and frankly, I find it hard to see that far.

That said, in the 20 years behind me, I've published 674 blog posts (excluding this one), 252 links and 188 book reviews. Annoyingly I've got 75 blog posts in draft, some dating back to 2019 (I really need to tidy these up).

My blog is also entirely all my own code - for better or worse.

I'll keep publishing here. I do love my little corner of the web. A mix of musings and code and goodness knows what. Thanks for reading and hanging out online with me for all these long days 💞

Originally published on Remy Sharp's b:log

It's not, it's masonry (which, why not display: masonry 🤷

There's a also a decent (albeit stiff) video on how it works.

TL;DR: mostly you can swap display: grid for display: grid-lanes (but Jeremy goes further to actually add the CSS support check first).

Source: adactio.com

Originally published on Remy Sharp's b:log

In our household we've already had this restriction in place (at a network/DNS level) so it doesn't impact our kids experiences at all, but I do know from talking to the kids that the do wish their friends had it too. I would add that I'm in no doubt that kids will find work arounds, but the larger effect of this, I think, is important.

A few quotes that interest me from the GOV.UK news post:

The ban will therefore include platforms like Snapchat, TikTok, YouTube, Instagram, Facebook and X. We do not intend for messaging services like WhatsApp and Signal to be included in the social media ban

I'm assuming that "like" refers to the user-to-user definition, and I'd expect or hope to see pretty clear definitions (though the government doesn't particularly have a track record on clear definitions).

The government will also be looking in more detail at overnight curfews and breaks in infinite scrolling for under-18-year-olds and will set out more detail in July.

In particular the breaks in infinite scrolling - though I've been personally arguing that it could be banned altogether - the only function it serves from a UX perspective is "stickiness" aka - addition.

I'm sceptical about the implementation rules and I'm particularly worried about how the government goes about setting the definitions of what products will need to do.

Source: www.gov.uk

Originally published on Remy Sharp's b:log

I think what I find interesting here is after the jQuery mobile themes and especially Bootstrap you come to just visually adapt the most common style.

What this author has done is asked for the design systems of existing software to be adopted into the code.

What frustrates me a little bit is that I actually like the "Original" style, which is what AI is pumping out at the moment, as what I see all the time, which just says that I'm not very original…

Source: envs.net

Originally published on Remy Sharp's b:log

The UK government's ultimatum to Apple and Google is a blueprint for mass surveillance. A warning on why "nothing to hide" won't protect your digital freedom.

...

We're being steered toward a future where "safety" is just a euphemism for "unrestricted access."

If I had read this a few years ago, I'd assume it was the plot for some dystopian novel, but no, it's reality today under (what was supposed to be) the progressive government - not even the Tories.

As per the author, I don't expect any of the Big Tech to do anything about it if it means it affects their share of the gold.

I guess I'll be looking around for alternatives soon enough too (and then for the kids, and the extended family). Yay.

Source: the.unknown-universe.co.uk

Originally published on Remy Sharp's b:log

This is the extension of a previous post, Molly Guard in reverse which introduces the molly guard with:

…the little plastic safety cover you have to move out of the way before you press some button of significance.

Anecdotally, this is named after Molly, an engineer’s daughter who was invited to a datacenter and promptly pressed a big red button, as one would.

Then she did it again later the same day.

Love it.

But what's more to love about this particular post, is that Marcin found the actual Molly with her dad!

Source: unsung.aresluna.org

Originally published on Remy Sharp's b:log

The "AI overview" is its own content, not just a list of search results.

This is really interesting and I would love to see this set the tone for other countries (::cough UK::).

If enough of that wrong content defames companies or individuals, it could become a serious legal problem not just for Google but for other providers of similar services like ChatGPT, Claude, or Perplexity.

Big Tech helped themselves to our content. The content bloggers put out on the web was always intended to be consumed freely, but it we didn't agree that you (Big Tech) could use it for massive profits without the least bit of credit (let alone royalties for the content being used).

The big AI companies have harvested and thrown the web into their models without moderation causing all kinds of real-world harm, so this looks to be a positive step to forcing them to take some real responsibility.

Source: the-decoder.com

Originally published on Remy Sharp's b:log

The term deepfake traditionally refers to photos, audio, and video, but when it comes to discussions of consciousness, we need to regard text as a deepfake medium as well.

This does remind me of the story of ELIZA and Joseph Weizenbaum, which people quickly attributed intelligence to the bot. With 6 months of ELIZA's debut (in 1966), he warned loudly and publicly about the perils of anthropomorphism. He continued this campaign until his death in 2008. Yet, here we are again.

Source: www.theatlantic.com

Originally published on Remy Sharp's b:log

Source: dbushell.com

Originally published on Remy Sharp's b:log

The open rate is usually around 40% (though I know some email systems synthetically do this), but our click rates are also usually pretty good.

This week however, open rate was 3% and click rate was just 8 clicks. Something was definitely…afoot.

Spam scores

I had a report from a friend of the event that last week's newsletter landed in their spam folder. From this, I decided to integrate spamassassin to try to score the content (which it almost always comes back clean).

As belt and braces, I also paid for some credit for Mail Tester which gives me an email address that I can send to and it'll run all manner of tests, including stuff outside of the actual markup.

I had a score of 9.8/10 - so I dipped into what could be improved, and suggested "check DMARC policy state".

DMARC

The page suggested:

Not being that versed in email security set up, and like many of us, having to wear many hats, I just copied what I needed and trotted along adding this record.

What I didn't appreciate was that this change translates to:

If the email didn't come from Left Logic, just reject it.

The problem being is that the email doesn't come from Left Logic, it comes from AWS because I use SES for sending the newsletters. So even though our newsletter product said they were all sent. AWS also said they were all sent. Only a handful of people had the email.

The only reason I figured it out was because I hadn't received the email to my personal email.

The temporary fix was to add this to the _dmarc.leftlogic.com as a TXT:

"v=DMARC1; p=none; rua=mailto:dmarc@leftlogic.com"

I do need to sort out proper SPF records and DKIM and "Easy DMARC" (whatever that is), but next time around I might be a little more careful with what I copy and paste…

(though, who am I kidding!)

Originally published on Remy Sharp's b:log

The development this iteration of the game, the one in 2026, took me about a week and I haven't written a single line of code. Here's how things have changed.

TL;DR

I get it, this is a stupidly long blog post. There's no real value buried inside of it, it's pretty much entirely a story of how technology has progressed over 30 years (not really a spoiler) and my journey to evolve a game that I found in 1996.

The latest edition of this game was coded entirely by Claude Code with one strict rule: keep inside of the 32K boundary limit. There was a huge saving that Claude never suggested (1bpp font tiles instead of 2bpp) which bought me 3K - which was enough to add a full tutorial in the game along with additional small features.

I also tried to vibe code another game using the same process only to fail - the key take away for me was that my experience and knowledge was still the guiding light. With the other game, it failed because my knowledge was shallow, so the output was shallow. Sort of obvious on paper, but a useful exercise for me.

Oh, and if you scroll to the end, there's a link to the game download page.

1996

At some point during my days during Sixth Form (college to some, basically aged 16-18), a close friend had (some) machine (that neither of us can remember) and they let me play a game on it.

I remember really enjoying the game. It was technically a very simple puzzle game: a bunch of balls that you would tap the matching colour groups of balls and they would collapse down into the corner. The goal being to clear the screen.

I never caught the name of the game but my dad had a Psion and remembered that in the instruction manual (long gone are those times) it mentioned you could write your own software.

I'd been making little Visual Basic programs for friends for a few years by then, so I decided that since I had access to the internet (of sorts) through bulletin boards I would try and build my version of game so I could play it all the time!

When I finally managed to download and printed out the documentation (I remember it just kept printing and printing and printing), the very first is what hit the hardest.

SDK. This was foreshadowing. I had no idea what an SDK was. Nor any of the other mass of acronyms. And software development being gate-kept by the usual nerds meant there wasn't going to be any help with the "basic stuff".

(In truth is might have been any kind of acronym, I just remember being bewildered).

Google doesn't exist yet so I tried thumbing through the documentation and spent probably weeks just going back and forth just looking for any clue that would help me get a footing so I'd have something to start with.

In the end, I got absolutely nowhere. I park the idea and put it down to lack of knowledge, and lack of experience. This was not for someone like me. This was for people who had the inside knowledge. Plus, at that age, I had plenty of other (better) distractions.

2002

In 1999 I started my work placement for a year as part of my Computer Information Systems Design sandwich course degree (a mouthful!). In the end, I'd never return to finish my degree and stayed on for a decade.

But back to '99 and the start of the new century. The company was a web agency with the boss and two students (myself being one). Along with client work, the agency also had it's own side projects, one of which was Eurocool - a website dedicated to software for the Palm Pilot (they even made BBC news).

So I eventually bought myself one and through Eurocool I started notice that some of the games and some of the utilities came with source code and the source code was not huge and by this point I had learnt a lot more programming.

By that time I had tinkered with some C, worked with Java. I was writing JavaScript (JScript and VBScript). I was writing Perl and I could start to understand the functions of the source code and I went about making my first utility for the Palm pilot for me.

I started with a couple of smaller bits of software for the Palm Pilot, before I realised I could (and should) make a game

As it turns out that I'm not a very good graphic designer, so when it came to drawing circles for the Palm Pilot it was…very, very bad. The best I could do was draw squares and colour them in. So this is where the name Marbles Squared comes from ("it's a feature, not a bug!").

It took me until the mid 2020 so realise that the original game from the 90s was actually SameGame that I had unwittingly cloned (though only the game play mechanic, the Wiki page details the scoring system which I was never aware of).

One of the key mechanics of my version of the game was that the the random function would have a seed, and that seed would make the grids being played repeatable and shareable.

My game even turned up in American Scientific (though the original article has been long disappeared sadly). After a few years, the game had been downloaded over 1/4 million times.

More importantly though, the coding part took me (probably) 3 months to go from scratch to the first version of a workable game, mostly coding late into the night because I had a full-time job (running in start-up days of getting home at 8pm and later). I'd continue to iterate on it for a further 6 months and the last release was in 2003. If you want to try the game on a (pretty faithful) web port of the Palm Pilot, I'm hosting the Palm version on the Marbles website.

2008

By the mid-late 2000s, I had move away from London and I joined the web scene down in Brighton. Coincidentally the iPhone had launched in 2007.

Eventually I caved and got myself an iPhone and Steve Jobs original vision was for developers to use the web and the power of the browser (sidebar: that wasn't really his vision, he just didn't want 3rd party apps).

So that's what I did. I took my knowledge of Marbles Squared and the only code that was directly ported was the random function so that I could continue supporting the seed mechanic.

I also add a timer bar across the bottom of the screen (to let me play with CSS transitions), but functional it did nothing, so watching friends play the game and panicking about the timer running out was a real delight!

Developing the core of the mobile version was a matter weeks (if that) because I was breathing JavaScript daily, and that I was so familiar with how the game would play.

As with all these projects, it really wasn't the core gameplay that took the time, it was the UX around the game. The damn button in particular with the "new" border-image (vendor prefixed to kingdom come!) probably proved to be the trickiest parts.

A it probably took me a few weeks to finish the mobile version.

2020

In 2020, the ZX Spectrum Next Kickstarter project was delivered (after quite a few long years of waiting) and with the pandemic lockdown in full swing, it was time to tinker.

The Next came with an upgraded (and much more capable) NextBASIC (plus the crystal boost to 28Mhz made a big difference). So I set about making myself a game.

I wrote a clone of Oh Mummy that I used to play on the my friends' Commodore C64 but after that was done I turn my hand to recreating my Marbles game.

The challenge here was that even though the Next was faster than an original ZX Spectrum, NextBASIC still wasn't fast enough to do the grid traversing to workout available moves - or certainly not in a way that felt responsive.

So this meant turning my hand to Z80 assembly. I'd tinkered with the language but never had a real project to solve with it. This was perfect for dipping my toe into. Plus, there was 40+ years of experience on the web to learn from.

The bulk of the game would still be in NextBASIC but it would call out to a library that I wrote in assembly for hard tasks.

That project probably took me over 3 months because it was a brand new language. It was also much harder to debug. The development cycle is a lot longer compared to something like JavaScript.

But as I reach the end of that piece of work, I realised that I wanted to add a high score table.

The Spectrum Next (most times) included a Wi-Fi module, but it didn't have any native HTP protocol support so was another side quest.

I built an HTTP client in assembly for the spectrum and one of my more prouder moments that my .http is now part of the Spectrum Next distro, part of the default library of tools.

This would let players see a global leaderboard on the spectrum and share their own high scores directly from the retro hardware.

The http part was probably another three or more months, but absolutely worth it.

2026

In the intervening years AI has exploded as we're all well aware.

In the last few months I have taken to using LLMs to create small personal projects that serve a single purpose, a tool to calculate the distance between a and b, but also to add how long it would take me to walk or run and so on. Or maybe it would be a small song game that I would share with my friends.

About a month ago, I decided that I would take my experience of vibe coding and see if I could build the Game Boy game that I've been wanting the Game Boy version of Marles Squared (ironically I discovered, literally today, that Samegame was available on the Game Boy in 1997 - though it's very different from the game I wanted to enjoy).

I was able to describe in detail exactly what I wanted and I was able to iterate over and over and over to get all the details kind of fine-tuned. It certainly wasn't just a one shot build me a Game Boy game.

As with all the previous iterations, the real work was in the UI.

I did however, set myself the constraint that the game had to fit into a 32K rom (this means it works on a cart that doesn't have an MBC - memory bank controller - chip - which is what allowed games to grow up to 8MB large).

The LLM (Claude Code in my case) was also vital to really quickly turning around tooling that helped me in the development process. Tools to slice my graphics into tiles or to visualise tile data inline to code, or to convert fonts (for variable width fonts which I didn't use in the end due to byte size limits).

I was also able to take an existing rust implementation of a Game Boy emulator and layered on an MCP server so that that Claude Code could test, screenshot and probe the memory of the game to see if it was doing what I needed to do.

But the whole process was me just talking back and forth to Claude code in particular to get this game built and the vast majority of it took about a week.

After a couple of rounds of testing, I did notice that the couple of people who tried the game didn't quite know what the goal was. A "how to play" section was needed in the game, but given the constraints of 32K, lumping in a load of text copy wasn't going to work.

One of the biggest wins I had in optimising the codebase was not from Claude. I could see all the text tiles (the tiles that make up the font) were two colours. Whereas a Game Boy tile stores 4 bits of colour in a "pixel", I could see I was wasting a lot of space. GBDK even includes a native function set_bkg_1bpp_data that will make use of 1bpp tile data. From this single change (I had to update my tools to generate 1bpp) I was able to unlock over 3K. This space was then used for the tutorial and a number of other small changes (including increasing the pool of "easy game seeds").

So in the 30 years that's passed I've gone from taking weeks and weeks looking at documentation and getting nowhere to taking a week to build a fully working piece of software that suits all of my requirements.

But the reality isn't as straightforward to say you can just magic up anything you want. My advantage I have is that I've got decades worth of experience in software development. I also have an intimate understanding as to how this game is supposed to work.

As an experiment I tried to vibe code another game. Our family was playing 100 Jumps and I thought it'd be a good candidate for a Game Boy game.

The source code is online, and on the surface it seemed pretty straightforward to me to re-use the techniques that I was using to develop Marble Squared.

Except what happened is I kept iterating on something that was awful. The physics were wrong. The movement was wrong. The graphics were bad. No matter how many times I tried to iterate and even reset, it just didn't come out feeling like a decent experience.

I've come to realise that the reason for this, is because of where my experience lies, and in particular my intimate understanding of how my own game works. This knowledge allowed me to carefully plan out the development, to know where all the pitfalls were ahead of time, which I simply didn't have for 100 Jumps.

I think we're starting to see in the tech industry that the pitch that AI is here to replace our jobs isn't entirely true. In fact, a lot of us have known for decades that the real skill isn't in producing code, but actually understanding functionality and translating it to the real world.

If you made it this far, thank you for reading. If you simply skipped from the top to find the download link, I appreciate you too 😄

You can download the game over on itch, and if you fancy your chances, you can submit your high score via the Marbles website.

Originally published on Remy Sharp's b:log

It took me until around 20% of the book to get on with it properly. I was struggling to even want to pick up the book. It's well written, but the subject, end of life, was so unappealing that I found myself wanting to avoid reading the book.

At some point around 1/5 of the way in, this feeling settled and I was swept along by the narrative that the main character, Bo, gives us.

I think reading this there's going to be different aspects that hit different people hardest. I'm not a dog person, so the thread about Sixten, his dog, though sad (the dog doesn't die!) it didn't gut punch me.

I did however relate hard to the inner monologue not linking up to what's actually said. Something I suspect most people can relate to - which I suspect is the reason this book is well recommended.

The book is told almost exclusively from the thoughts of Bo, the elderly 89 year old man (which took me quite a few pages to actually work out). From his thoughts we travel back and forth in his memory which worked really well. I didn't feel lost at all, maybe because Bo's memories felt like they flowed naturally back and forth.

The story really does feel like it's being faithful to the human flaws many of us share.

It also, for me, did an excellent job of building up my relationship with Bo and gently brought me to tears right at the end in a way that made me message my kids to tell them I love them.

Originally published on Remy Sharp's b:log

Originally published on Remy Sharp's b:log

For little pop out menus it's perfect. Then I also started using it for modal boxes (yes, you might be thinking ahead of me: dialog, still, let's find out what happened).

The popover property was really easy for this, and the ::backdrop selector was perfect for adding a little backdrop blur to tell my user "this is the thing you're dealing with".

I especially love that all the UX around dismissing is handled for me. I can hit the escape key, or click outside the popover and it's dismissed.

Except, I kept noticing that when I dismissed my popover it would sometimes vanish then immediately pop back to life…

On using the wrong tool

It was through a chat with Bramus during a BTConf lunch that it was pointed out that I should be using a dialog for a modal.

I think the reason I didn't think of this earlier is down to the age old problem of naming. I have a mental model that a modal is a different thing from a dialog box. Specifically a dialog lets me set it aside whilst I carry on with what I'm doing.

For example, in many native desktop apps, a preference dialog can be opened, but I can still interact with the software.

Whereas a modal is "you need to do this and we'll lock out the app until you do". Although dismissing the modal is also valid as an action.

Anyway, that was my thinking as to why I hadn't used a dialog for modal actions.

I also didn't know I could use the ::backdrop selector with dialog - mostly because I'd heard of it in the context of the popover.

So what's the anti-pattern?

A popover doesn't trap focus (which is the whole point of a dialog). But I wasn't looking at keyboard focus, I was looking at mouse focus.

When I used the popover for a modal (looking) element, and used a blur on the backdrop selector, I didn't realise that clicking on the backdrop would let the click carry through to whatever was underneath it.

So visually the UX is saying the modal is the thing that you need to do, and clicking anywhere else will dismiss it. Technically, I could have a solid colour hiding everything under the modal and you'd have no idea what you were clicking through to.

Let's look at the two buttons below. The "click jack" one will show an alert box. If you click the "modal" button, you'll get my modal, and under the solid backdrop, I'm going to position the "click jack" button across the entire screen, so when you click away to dismiss the modal, you should get the alert.

Besides the JavaScript click handler for the click jack alert, this is all done with CSS:

/* put a solid colour over the page */
#demo-popover::backdrop {
  background-color: hotpink;
  /*
    note there's a bug that I've filed in FF that
    doesn't like solid backgrounds.
    https://bugzilla.mozilla.org/2038260
  */
}

/* expand the "bad" button under the page */
#demo-popover:popover-open + #demo-clickjack {
  position: fixed;
  top: 0;
  left: 0;
  width: 100%;
  height: 100%;
}

This isn't really a demo of click jacking (I've not fully thought though whether it's a real world issue or not), but it shows how I had managed to click to dismiss my modal and then accidentally hit a "live" target in my blurred background.

Now with a dialog

In this example, when the ::backdrop is active, it does still have a giant "click jack" button under it (you'll need to inspect the DOM to validate this), but the backdrop also traps the clicks and it doesn't go through.

That said, you can swap this same code to use popovertarget and the same problem occurs (because it's not modal, and won't trap the click).

Side note: oddly, if you happen to click on the button that opened the popover, it doesn't re-open the popover. You can see this in the MDN demo of dialog with popover.

I think ::backdrop might be the problem

For me to fall so quickly into this pattern of using a popover with the ::backdrop, and not realise I was taking the wrong approach feels like it might be something others will fall into quickly too - especially if you work in an environment where generated code isn't scrutinised.

I could even see developers trying to use pointer-events to prevent clicking through the ::backdrop whilst using popover which would really be committing to the wrong approach.

Another question is: what is the ::backdrop pseudo element for, if not obscuring the background? Therefore, if the background is obscured, then surely we're looking at a modal element (because the background is now visually out of focus)?

Which all leads me to wonder if popover should even have support for ::backdrop.

Originally published on Remy Sharp's b:log

Thanks to Big Brain Keith Cirkel for sharing this.

As (hopefully) you know, there's the $ and $$ functions in devtools. For querySelector and querySelectorAll respectively.

Well there's also $$$ to query through the shadow DOM. This means the query will cut through and into the nested DOM inside of the shadow DOM making it much easier to navigate and search for elements when debugging or styling.

It's worth also adding that, just like $ and $$, this third function also takes a second argument for context, so it means this is valid (where $0 is the currently selected DOM node in the devtools inspector):

$$$('ha-card', $0)

Begs the question: what's going to be lurking inside of $$$$ in 5 years time!

Originally published on Remy Sharp's b:log

Aliento

For some reason macos, for quite a number of versions has made it really hard to dismiss notifications (heaven forbid you repeatedly flash a raspberry pi pico). The hot target is horrible and there's many notifications that don't bundle together (the unmounted drive springs to mind).

Aliento give me a single key binding that swiped all notifications away. A daily use for me.

Aliento

Muzzle

Speaking of notifications, if you've ever given or seen a talk or video call when a distracting notification appears then this the app to solve it.

It's a toggle that muzzles your notifications. Once turned off, they'll all pile in - and you can review or use Aliento.

Muzzle

Shottr

I'm sure most of us have our goto screenshot tool. I've been using Shottr for quite a few years now. The base functionality is good, but what it adds is a quick way to draw, point, annotate and even OCR screenshots.

One extra little feature I like is that I can then drag the screenshot to where I want to share and then the file is never saved (my screenshot folder is 'uuuge already!).

Shottr

VoiceInk

As big as my phone is, I find myself dictating more and more (somewhere last year the voice recognition had a massive step change improvement).

Lately I'd been looking for a suitable voice to text on my desktop and laptop but ideally with the processing done on machine (rather than needing to wait for the roundtrip to some random server).

I use VoiceInk multiple times a day. Both for push-to-talk and for much longer sentences.

There's options to download different models, but I've found the Apple native STT to be suitable and usually the quickest (even though one machine is on Somona and the other is running Tahoe).

VoiceInk

KeyboardCleanTool

It helps you do the thing it says. Though it doesn't clean your keyboard, it does trap all key presses so you're not randomly posting emails containing keyboard mashing.

KeyboardCleanTool

Karabiner-Elements

This one has been installed on my machine(s) for years. It's one of those apps that sit in the background, do the work, and make things just a little bit more to your liking.

Primarily it's a key binding app. In my case, I wanted to swap my ctrl and alt keys, but only when using a specific keyboard. This does the job.

It also allows me to add much more complicated bindings (which historically have been tricky, but LLMs are pretty good at generating these if you need them too).

Karabiner-Elements

Better mouse

I've got a Logi MX (something) mouse. It's very fancy, and has more buttons than camo trousers have pockets.

Earlier this year, Logitech pushed a server change that somehow (I really don't understand how or why) borked the Logi mice/mouses(?!).

With that I immediately sourced an alternative that didn't bind my physical device to a cloud server 🤦

BetterMouse is that for me. Tonnes of configuration, specifically I can bind the buttons to perform custom tasks depending on the focused application.

What also stood out for me, the UI for finding the right button (that I wanted to bind) was much easier than the existing Logitech software.

BetterMouse

Sim Daltonism

I can't remember quite when I first installed this app, but it's got to be at least a decade.

It's a window that sits above all else, and let's me quickly cycle through different types of colour blindness.

I know that some of the browser dev tools included this, but I found a standalone app the simplest (since it can also see beyond the browser, such as mock ups in Photoshop or Sketch and the likes).

It's an app they I always reach for once the colour system is in place for the website I'm working on.

Sim Daltonism

Tahoe Electron Detector

Let me say this first: do not, if possible, upgrade to Tahoe. It's a total mess.

Still, if you unwittingly upgraded like I did, you might find that software becomes somewhat sluggish. For me I noticed hover effects in Firefox was firing log after I had left that target.

The issue is related (from what I understand) to some private libraries being used by Electron apps. A full reboot would resolve the issue until it raised its ugly head again.

So Tahoe Electron Detector checks your installed apps, and tells you which haven't been upgraded away from this particular bug.

Once I know what the list looked like it was easy for me to stop using those apps (Dropbox and Discord at the time, and I could use the web alternatives), and the sluggish problem went away.

Tahoe Electron Detector

Bartender

This is probably one of the most popular apps I'm listing, so I suspect many of you have already heard of it. Hopefully it helps a few of you though.

Bartender let's me pick and choose which menubar icons are visible, and which are tucked away. There's also a rule system that lets you show particular menubar icons depending on conditions (such as battery level or WiFi connection).

There are a few issues that I have with the latest edition (but are primarily due to macos rather than Bartender itself): there's a couple of system menubar icons you can't hide (like the notification centre), and I've found that some menubar apps don't open and stay open when Bartender is running (NordVPN being one of those apps).

Bartender

There's many other apps I use, but they're much more common place. There's also apps that are very cool (notist) that I just can't make into a habit.

On that note, I'd love to read your favourites that might be hidden gems.

Originally published on Remy Sharp's b:log

I really enjoyed reading this book. The story floats like a butterfly from one character to the next and it works without losing the reader.

The story, I think, follows the Flynn family and though they're painted as somewhat dysfunctional, they're all spirited in their own journey.

There's a line towards the end that reflects on the mother's character not having given up on her art career, but having chosen to be a mother. The book definitely left me with this sense: that we're okay.

There is a strong religious thread in the book (not being religious myself) but it wasn't an agenda being pushed, but rather something for some of the characters to hang on to.

Easily enjoyable. I loved the characters and would definitely recommend.

Originally published on Remy Sharp's b:log

I bought and read this book after seeing Caroline Lucas speaking with Zack Polanski during a live session of Bold Politics. It's a book that tries to reclaim "Englishness" from the weaponising that the Far Right have been using it for.

Overall, it's an interesting read, lots of history, both good and bad, lots of literature and how stories and poetry has been used over the centuries to act as a force for political change for good.

Solidarity and standing up for what's right is the theme the ran through for me.

From a technical perspective, I didn't enjoy that nearly every single page also came with a long paragraph of footnotes - which normally I skip (because it goes off to the end of the book and back, but these were inline on the same page), and completely broke the flow of what I was reading. I wish that these had simply been part of the written content rather than regular sidebars.

Then as I read, I felt like perhaps I wasn't quite the audience to fully appreciate the book. There's certainly a good chunk that is accessible, but it often felt like if I worked in politics this content would make more sense.

I did come away with 23 highlights and wanted to highlight more, partly to remember what I had read, partly as a "wow, I did not know".

I think if you enjoy your history and commentary and literature, then you'll enjoy this book. I struggle to read non-fiction in any reasonable amount of time, so I often find by the time I get to the end of a book, I forget the important things from the start!

Originally published on Remy Sharp's b:log

My Android phone, on our network, on the same SSID, speed tests at 1.1Gbps. My laptop gets around 160Mbps and my desktop pulls in around 80Mbps.

That's quite a difference, and even though I don't need the gigabit speeds - sometimes I can tell when the desktop is being slow (web site be getting big).

What was weird is that the network speeds get much faster if I add something like NordVPN to my connection…which doesn't make sense…

TL;DR - AWDL

It's a "magic" non standard enhancement (Apple Wireless Direct Link) to WiFi that Apple use that can seriously hamper your network connectivity. It's hard to kill, so the "simpler" approach is to ensure the Wifi channels you use are 44 (for 5Ghz in Europe, or 36 outside).

Slightly longer

My router, for it's 5Ghz channel, uses 160Mhz using (something called) DFS (apparently nothing to do with sofas…sorry, British joke): Dynamic Frequency Selection.

This seems to throw Apple a curve ball when it comes to their AWDL (honestly, screw Apple for having more tech that we can't control and ends up getting in the way).

AWDL is part of the system that's (from what I understood) responsible for things like AirDrop (which I don't use). Of course, turning it off doesn't really work - it's an easter process: it keeps respawning. It sort of stayed disabled on my laptop running Tahoe, but on my desktop running Somona it wouldn't stay down regardless of what I did.

Even though this problem is exclusively an Apple problem, the only way I could resurrect my network connectivity (and even though I caught it at 60Mbps, I'd seen it much, much slower) was by switching the 5Ghz to 80Mhz (all the hertz) and selecting 44 as my channel.

Night and day difference:

Ultimately my plan is to set the Wifi booster/mesh/thingy to have a Mac specific 5Ghz connection point, which is then hard wired through our house into the router, then I can have 5Ghz at 160Mhz band (so nice for all other devices, like our TV) and - hopefully - the pesky Apple devices can pull data at a decent speed too.

Originally published on Remy Sharp's b:log

The short version is: if you get the chance to attend these events or even anything similar, I'd highly recommend that you grab that ticket and let the event wash over you.

For those that didn't attend (or maybe you did and wanted to read my perspective) then here's my thoughts on the separate events and then the round up of my own experience.

State of the Browser

London based and fronted by the lovable Dave Letorey, always has a strong accessibility story - both from a practical ("how to make the web accessible") but also universal access - in that all are welcome and an open web is championed through the talks and the heart of the event.

I tried to take a few notes for myself during the day, but this quickly fell to the wayside by midday (I find if I'm taking notes I'm slowly losing track of the talk, so it was more important at the time to stay in the moment).

The day did start with Bramus Van Damme with an excellent dive into CSS anchor position, an exciting new feature of CSS that's landing throughout browsers that means we can move a lot of complex JavaScript into the bin. The CSS spec itself it's wholly simple, and there's a blind spot specifically around recreating call out arrows (usually done using rotated weird border magic).

I do question the ease of testing - though I think a talk on how to test CSS (with launching a browser and diffing the screenshot) would be extremely valuable.

Another talk by Fiona Safari on how they learnt to work with their introvertness(?) left me feeling very "seen", though Stuart did raise the valid point that this felt a little like the Barnum effect - where you're able to recognise yourself in most things (we're human after all, and patterns and self interest is programmed deep in us).

This quote really meant a lot to me, and I think about kids who don't push themselves to the front of the class - this is one to remember:

Zach Leatherman's talk looking at how a common picture compare component would and could be developed from the day the first img element was released to today. Especially drawing attention to the "dead zone" where code has landed in the browser yet the interactivity isn't at all ready.

The time to interactivity is a metric I think (I hope), we're thinking of all the time, but to see it articulated and in the context of a relatively contained context really helped - especially considering that webbies have to go back to work and argue the case.

Then Jason Williams gave a wonderful context to how Temporal came about and the sheer scale of the work.

I've spent my fair share of bemoaning the JavaScript date object (since '99, so yeah, I've been at the sharp end of Date suckage!), so learning about the mammoth effort that went into bringing a completely new date structure was a real treat.

I've been using Temporal for a good few years via a polyfill, but to learn there's so much more and it's arriving cross browser (and server-side) is valuable.

Chad Gowler's accessibility talk was absolutely superb and I do hope they have the opportunity to give the talk again at more events - so many people can learn from these experiences.

Likewise with Mike Hall's story of battling for every single byte for performance in a time when it really, really mattered (and in some senses, still does). Mike's always a pleasure to listen to (both as a speaker but also his podcast).

I'll be there for next year's State of the Browser

Web Day Out

My general feeling from (the first) Web Day Out was that there was a strong sense of practical takeaway from the talks.

I'm hoping the slides will be available (if they're not already) because there's lots of little details in many of the talks that I definitely felt like I could apply today.

All of the talks from Web Day Out were ace quality, reflecting the wonderful speaker line up.

Jemima Abu very much set the starting tone for the day. A taste of new and exciting CSS features (touching on some of the items from Bramus at State of the Browser), and speaking to the solid idea that there's technology that's moved out of the JavaScript cycle deeper into the platform, either available in CSS or event HTML itself (such as popup). Giving me even more reason to eject bytes from my code.

The original Lady of CSS, Rachel Andrew talked about what Baseline means to development and talked about the very practical nature of support.

One incredibly interesting item that I've not heard others mention: Rachel points out that given the nature of the data collected on Baseline support, it's now possible to accurately predict browser support by a given amount of time. Specifically, if a feature moves to "newly available" today, it will land in "widely available" in 30 months. This might sound like a long time, but Rachel makes the excellent point that developers working on a year-long+ project can use features that may not have wide support from the outset, but they'll be able to pick features based on expected support at the time they go live.

Rachel did also talk about how the Baseline project, in an early form, did consider whether to include polyfills as part of the support timelines, but the search for a "gold standard" of polyfill was…elusive. For my part, personally, I wouldn't want or expect to see polyfills being stamped with approval by browsers. As Rachel said herself, the requirements for polyfills (and what's considered Baseline) is entirely dependant on your own projects and products.

Next was Aleth Gueguen who shared their story of development and testing directly from the trenches - figuratively and literally. In the post-social a good number of people shared how much they appreciated the insights saying how they'd love to hear more real-life stories, as often it's the real development and challenges overcome that make an interesting story (echo'ing Mike Hall's story from State of the Browser).

Then Harry Roberts took the stage to talk outside of his usual stomping ground. That's to say, instead of performance and measurements he instead spoke to the impact of not building for the web. He shared practical examples of where (anonymised) clients had brought Harry in to work on their performance issues whilst they grappled with framework upgrades. It's a strange web-world we live in when the typical path development path is to reach for Next/Nuxt/Fux/Whux/evar because everyone is does and laden the users with bloat.

Harry had many excellent points during his talk, one that comes to mind as I write today was how optimising for 2nd page load with a coach load of JavaScript, when the bounce rate is so high is premature optimisation (pause for feigned shock) - which ultimately costs negatively.

This wasn't Harry's normal wheelhouse of talk - by his own admission - but an excellent one, and I hope he gets to repeat it for others.

Post lunch, was Manuel Matuzovič sharing his newly gained knowledge from questioning if what he knew 3 years ago still stands, and modifying and improving the tech he was using - squarely in CSS land. He used his own pet project oli.css as a place to experiment (I strongly approve - always find a safe place to play), and in particular shared his really interesting take on a CSS "reset" (not a reset per se, because browsers have managed to converge on a solid starting point, but a great starter CSS) and how he layers and reuses throughout the project.

Keeping in this space, which Jeremy made clear he had curated the talks and order into these excellent pairings, Richard Rutter talked us through what had changed in (web) font typography in the last 9 years since he published The Web Typography book.

I'll be honest, I really didn't think the web type scene was either interesting or even moving. But by gosh there was a LOT. Tonnes of interesting tweaks that we can be making here and today, lots of features that were in Baseline, others that were coming. Of all the talks, Richard's had the most "oooh - I really need to pinch this for my website" slides. I'm hoping they'll be available (and I remember to find it) because it really was a treasure trove.

In the final straight was Jake Archibald taking a zoomed in view on the customisable select element that is landing (or landed). A (typically) hilarious talk where he went all the way back to 1993 and went through all the events that needed to take place before developers (finally) laid their hands on a stylable select - whilst also showing us how to style the select today.

Highly enjoyable and difficult to ignore the contents of his slides! 😂

Finally Lola Odelola talked about their experience approaching and implementing a missing browser feature - a fallout from their talk at State of the Browser in 2024 asking the question: should there be a prefers-alt-text CSS feature?. Very much from the land of how-the-sausage-is-made and makes me twitch to want to have a crack myself (lol - if I ever had time).

What I came away with

Look, if you've been to even a handful you'll already know it's the people between the talks that makes it all worthwhile. It's always lovely when there's talks that give me a little shove towards tinkering with code, doubly so if it's a friend up on stage (new or old) giving the talk.

For the first time in a long while I attended the post-event parties (I tend to miss out on Brighton or London event parties because home calls). In doing so I had some really, really lovely conversations with different people (though landing myself chats finishing at 3am and 1am respectively 🤦).

As I sit in these events, I'm thinking about what I want from FFConf. A conundrum I face is that as I increasingly follow politics, the more I want to share what I learn about the world.

At some point during the last of the talks I realised actually what this web community is, is a microcosm of exactly the kind of progressive community that I identify with. A community that cares about representation, dignity for others, equality and sharing.

I've always known I was happier in (what I called) the hippy nation of webbies, but it makes me realise actually what we're doing in the corner of the world can be applied to how we conduct ourselves in the broader sense - and actually the politics I follow these days are very much aligned with the politics that I care about inside of the web community.

Originally published on Remy Sharp's b:log

I wasn't sure what I was getting in to reading Rivers of London (recommended multiple times on socials), especially as I kept finding the teens/young adult book cover for it (reminding me of the multiple covers the Harry Potter books had) - but I decided to jump in.

I quickly got sucked into the book and the narrative style that Aaronovitch used. For some reason I had Millie Bobby Brown as the voice which caused me a little confusion when I learn that the protagonist and narrator is called Peter Grant (I really didn't read the blurb) - but decided that she could still run the inner monologue.

The Rivers of London is a very London and British book (which I always especially enjoy because I can vividly see the places, hear the accents and usually feel the weather) that's a mix of modern copper plod story with fantastical (and certainly for me: original) characters from gods to vampires to ghosts and wizards.

There's a good balance of funny and witting against the (some times) gruesome and frankly pretty terribly bad things that make for the primary investigation for the protagonist, Grant, to work through.

Really fun stuff - I can definitely see myself returning for more in this series.

Originally published on Remy Sharp's b:log

In a project I'll often work on and move around different branches throughout the day, and as the years wear on it's rather dulled my memory - that's to say, I quickly forget what branch I was working on!

git recent

I created a git command line alias that helps my failing memory. What it does is list all the local branches, oldest to newest with the relative time and the name:

[alias]
  recent = !git for-each-ref --sort=committerdate --format='%(committerdate:relative) %(refname:short)' refs/heads/ | tail -10

It's a bit cumbersome with the !git part at the start, but it's because I want to limit it the most recent 10 results and I want it with the newest at the bottom.

$ git recent
3 weeks ago fix/ai-gen-docs
2 weeks ago feat/not-available-not-404
2 weeks ago fix/bulk-download-timeout
8 days ago fix/real-null-in-bulk
8 days ago fix/real-404
2 days ago feat/bulk-schema
2 days ago fix/transcript-endpoint-potential-null
2 days ago fix/sanitise-sql-input
21 hours ago main
20 hours ago feat/api-analytics

Memory win for me. Neato.

Originally published on Remy Sharp's b:log

Originally published on Remy Sharp's b:log

I accidentally read Dogs of War in 2018 - I never went looking for it and loved it. Equally, I can't remember when I found out about Bear Head but I knew I wanted to read it if it existed in the same universe. So finally I get around to it.

Overall well written, well drawn characters with distinctive voices in a universe that's really not that hard to conceive. I've read a few of Tchaikovsky's books now and as much as I enjoy them, for some reason it does take me quite a stretch to read the books (this one being just ~315 pages and took me a month) - not sure what's up with that.

The story is told from three protagonist's perspective: Honey, the bear from Dogs of War - once soldier, now an academic arguing for rights for others. Jimmy, the "grunt worker" placed on Mars to prepare it for the colonies that plan to settle and Carole (Springer…I want to say), the PA and "collared" assistant to the antagonist Warner S. Thompson.

Similarly with Dogs of War, Bear Head discusses self and identity (can a person be entirely digital?), freedom and slavery, and the untouchable super rich and their power.

There's some pretty horrible scenes around human slavery where people are compelled to perform, both involuntary physically but also compelled to think in certain ways (i.e. being submissive by design).

With a lot of what I'm reading lately, I can't help but associate it with today's real life events (in 2025-2026), and thinking of the corruption throughout politics we're seeing today - though I can see how this book being published in 2021 aligns fairly well with Trump's first regime from '17-21. Perhaps it's that backdrop that makes me a slow reader!

Originally published on Remy Sharp's b:log

Earlier today Jake asked me: what went wrong?

Fucking, everything.

TL;DR

I get it, this is a big wordy blog post. I got carried away and enjoyed telling the story.

The short version is: CloudFlare, probably best to upgrade everything to latest software (which I couldn't do), 520 status can actually be a mismatched TLS exchange between CloudFlare and the origin. Oh, and don't lean on LLMs too hard when the shit is really hitting the fan - try to take a big step back and make sure you take stock.

Otherwise, here's the too-long, do-read version -

On being in maintenance mode

For the last 5 ish/maybe quite a few years or more, JS Bin has run in a fairly automated maintenance mode. There's usually little flurries of wobbles that need my attention every 3/6 months. JS Bin is coming on to 18 years old, which is geriatric by web standards.

Typically it's dodgy content on the site that I need to put banning in place, sometimes it's take down requests that come through from Amazon (where JS Bin is hosted), sometimes it's a memory exception that takes me a little longer to recover from.

You can see from the last 11 years of status checks, there's been outages but nothing quite like what happened this time around (ignoring that big one on the far left…):

The larger hours when the server goes down (as I found out recently) is memory running out on the machine, and the machine will respond by pretty much collapsing. The result of which is that I can't connect to the machine via ssh to attempt restore it. It usually requires a forced reboot via the AWS console.

Then there's the odd occasion that even reboots from the AWS console, a literal "turn it on and off again", doesn't fly…

It wouldn't come back from a reboot

This latest outage just wouldn't come back from a reboot. I would trigger the reboot and then wait on the console attempting to connect over SSH so I could get some eyes on the situation but even then I couldn't get it.

JS Bin was rebooting and immediately locking up. So this tells me there's some external pressure on the machine that is not easing up.

The only option I have to hand is to shutdown the machine entirely for an hour or so, to let whatever is clawing at the door to go away for a bit.

When I eventually looked at the Cloud Watch logs, it was clear there was something absolutely smashing at the walls (and door, yes, I mix my metaphors). The amount of inbound network traffic is unprecedented for jsbin. From the chart below, it shows normal usage in the days prior, a trickle of network inbound traffic by comparison to what I was now seeing.

The dips that happen after the cliff edge of monitoring is the machine giving up and being unresponsive. From the chart, I can see that even the short outage wasn't enough to get this beast off jsbin's back.

Killing the appropriate process

When I finally managed to get into the machine, the first stop was syslog to find out what was responsible for the crash (or rather the symptom, not the cause, not yet).

I quickly found the garbage collection dump and stacktrace of node running out of memory. First thing's first: let's kill the offending process rather than having a full shit-the-bed approach.

# /etc/sysctl.conf and reload with `sudo sysctl -p`
vm.oom_kill_allocating_task=1

That would mean node (which was saturating all the memory) would be killed rather than anything the system could get it's hands on - which usually meant I couldn't ssh into the machine.

This change wouldn't stop things getting sluggish on the machine but it would mean that I could continue to diagnose even whilst the traffic was bombarding the server (albeit with a very slow terminal responding).

I could see the CPU usage was very, very high, and I could see node steadily increasing it's memory footprint (I use htop).

CPU usage of 100% is…okay because it means I'm using all the available process, but it leaves zero headroom and rather conspicuous that it wasn't dipping at any point.

Then chatgpt suggested upgrading node, which was weird because I'd never told Chatgpt what version of node I was running…

Side quest: node is really old

As with running in maintenance and things being in a status quo, I hadn't really touched node…at all.

As it happens, jsbin had been running on node 7 (not even "stable" node 8) for, I imagine, well over a decade.

To me, out of the blue, Chatgpt suggested I could upgrade node. This is a decent suggestion but at no point did I tell Chatgpt what version of node I was running.

On me querying, Chatgpt told me that I had told the AI which was a straight up lie.

After more prodding than I'd like, eventually it turned out that the version of node was in my terminal and at some point during the nginx tuning (in the next section), Chatgpt had helped itself to what was on the screen.

I've no idea if it has access to the scroll back history - I've got to assume it doesn't, but still pretty uncool.

I don't tend to use the "application" thing that Chatgpt has (I normally use the browser interface) but this definitely taught me a lesson: if there's sensitive data anywhere on the screen (ie. I had just cat an .env file) then it's entirely likely an LLM can see it too.

Anyway, I bumped from node 7 to node 22 and by some kind miracle it actually worked without any incident. It turned out that back in 2024 I had done some fairly significant work on the codebase so that I could run it on my local machine (which understandably didn't want to run node 7), so I had modernised the requisite parts.

Phew. At least the event loop is improved and going to be kinder on my CPU.

Except, it's still fucked and the adventure is far from over. Next I considered whether nginx (the proxy layer) could do with some optimisation.

Fine tuning on zip-all resources

Until now I've not shared what jsbin's main server runs on. It's a t2.micro AWS instance. Single CPU and 1GB of RAM. I'm often surprised it's managed so long on what I consider so little resources.

And yes, standing a beefier machine up (aka: throwing money at the problem) might help, but the time to reconfigure a new machine wasn't quick - I don't have an instant "build a new jsbin" script (remember, this server has been in maintenance for a long time and just happily running). It's also worth adding that although jsbin does has a "pro" offering, there's very little actual monetary resources. That's all to say: what can I do right now before creating a new machine (spoiler: I did make a new machine, and double-spoiler: it was terminated the following day as I didn't need it).

With the help of Chatgpt, Gemini and Claude (because somehow I have access to all of them, and I really don't have the full skills to know the ins and outs of nginx config) I looked at what could be tuned. I used multiple LLMs so that I could attempt to cross verify the advice (though I wasn't as diligent as I'd have liked).

The adjustments fell into these categories:

• workers spawned
• proxy timeout
• increase file descriptors
• keep alive timings
• remove http2 - to help with memory

Here's some of the actual config:

worker_connections 1024;
worker_processes auto;

keepalive_timeout 10;
keepalive_requests 100;

With the kicking I was getting from the inbound network traffic, this made little tangible difference. I'm sure it helps in the long run, but not when the machine is struggling with over 1000 requests a second and many more trying to squeeze in.

And then there was this:

Remy, have you considered CloudFlare?

I'd thought about add CloudFlare a few times in the past, but I was worried that the configuration adjustment and changes I'd need would be either complicated or would cause some other issue.

Except, nothing was working, so now was the time to get CloudFlare involved.

And so begin new problems…

Adding CloudFlare

Props where due, putting CloudFlare in front of JS Bin was relatively easy. CloudFlare detected most of the domains and where they pointed (the important ones at least) and it was a matter of swapping over the name servers from pointing to AWS's Route 53 to CloudFlare's own name servers.

Close to midnight on 29-Jan, I started to see jsbin.com load in the browser.

I thought there might be a few snags around, but quickly I got replies on github and via email that people were still (mostly) seeing errors.

In particular CloudFlare's 520 error page, which, as it turns out can be any number of issues.

Requests cutting through CloudFlare

Although I did have some success in loading, I could also see traffic still hammering through on the server.

With some help from our LLM overlords (i.e. I would have google'd it…"kagied it"?), I got the list of IP ranges that meant the traffic was from CloudFlare.

There was still a lot of traffic that didn't match those IP ranges. So the next step was to drop non-CloudFlare traffic.

This is where using the LLM cost me time via newly introduced problems (which wouldn't surface until the next day). I think I was too buried in the thick of things otherwise I might have caught it earlier (and not at the end of the process).

The first method to blocking non-CloudFlare was to use an nginx rule and variable, and to drop all traffic that wasn't tagged.

This would mean adding this line to each server {} block - this would say "If the request doesn't have the CloudFlare header, then don't let it through":

if ($http_cf_ray = "") {
    return 444;
}

Importantly there was more config directions that I followed (which I've tucked away, but you can read if you want), specifically set_real_ip_from 173.245.48.0/20; - this would trip me up later on.

/etc/nginx/cloudflare.conf

set_real_ip_from 173.245.48.0/20;
set_real_ip_from 103.21.244.0/22;
set_real_ip_from 103.22.200.0/22;
set_real_ip_from 103.31.4.0/22;
set_real_ip_from 141.101.64.0/18;
set_real_ip_from 108.162.192.0/18;
set_real_ip_from 190.93.240.0/20;
set_real_ip_from 188.114.96.0/20;
set_real_ip_from 197.234.240.0/22;
set_real_ip_from 198.41.128.0/17;
set_real_ip_from 162.158.0.0/15;
set_real_ip_from 104.16.0.0/13;
set_real_ip_from 104.24.0.0/14;
set_real_ip_from 172.64.0.0/13;
set_real_ip_from 131.0.72.0/22;

real_ip_header CF-Connecting-IP;
real_ip_recursive on;

http {
    include /etc/nginx/cloudflare.conf;
    ...
}

if ($http_cf_ray = "") {
    return 444;
}

This config was all well and good, but traffic was still coming in. This change would mean that nginx would have to process the traffic than ignoring (or "dropping") it entirely (#foreshadowing).

When I ran command below, it would list the connections, and I could still see non-CloudFlare requests being able to complete request (at this specific point in time, I was in a frantic cycle of "try a change - fail - worry a bit - ask LLM - repeat"):

ss -tan state established '( sport = :443 )'

The next change that I put in place was to drop the request if the IP didn't match the ones in the CloudFlare IP range (whereas before the change was looking for a CloudFlare specific header). Looking back as I write this, I'm not sure why the header method didn't work, but I was sure to add more petrol onto the fire:

geo $is_cloudflare {
    default 0;

    173.245.48.0/20    1;
    103.21.244.0/22    1;
    # etc
}

This flags if the remote address is a CloudFlare IP, set the $is_cloudflare=1. Then in my server blocks:

server {
    listen 443 ssl http2 default_server;
    if ($is_cloudflare = 0) { return 444; }
    if ($cf_valid = 0) { return 444; }

    # rest of config unchanged
}

What I didn't realise at this point, and took another 24 hours to figure out, is that the first CloudFlare nginx based change would say "if the IP is CloudFlare, then set the $remote_addr to the original request". Then this code says "if the IP is NOT CloudFlare don't let them through".

It's a mess. Now I've got reports from people on github telling me they're still seeing mostly 520 errors, and finally I realise I can replicate by using a VPN because for some reason it was still working for me.

There's two problems at this point:

1. Traffic still coming in that isn't through CloudFlare that's causing a drain on resources
2. Real users coming through CloudFlare are, mostly, not getting the site

Pretty much the worst of both worlds.

Finally, firewall rules spring to mind. Better late than never.

Dropping traffic

This was done on two fronts - just for belt-and-braces approach:

1. ufw (or as I knew it "iptables") rules one the server to DROP or ALLOW traffic based on IP range
2. AWS security policy to do the same.

Running the rules for ufw was relatively simple. A matter of allowing all the known IP ranges, then denying everything else:

$ ufw allow from 173.245.48.0/20 to any port 443
$ ufw allow from 103.21.244.0/22 to any port 443
$ # etc
$ ufw deny 443

Then repeated for port 80. I did get tripped up during testing when I wanted to allow my IP through, but that was solved with ufw status numbered and then ufw delete N.

Adding all the IP ranges on AWS was not so simple. The web UI just doesn't allow for large changes - it's very, very clunky.

Thankfully I could automate some of it from the command line and script the work. The annoying thing was that the AWS command line doesn't let me bulk modify (or at least I didn't find it) and each command takes a good few seconds to run and requires me to press enter after the response comes back saying it had worked.

This was the command:

for CIDR in \
   103.21.244.0/22 \
   103.22.200.0/22 \
   103.31.4.0/22 \
   # etc
do
   aws ec2 authorize-security-group-ingress --group-id $SG_ID --protocol tcp --port 80  --cidr $CIDR --region us-east-1
   aws ec2 authorize-security-group-ingress --group-id $SG_ID --protocol tcp --port 443 --cidr $CIDR --region us-east-1
done

With those in place, finally the server was breathing again. However, those pesky CloudFlare 520s were still preventing people from visiting jsbin.com (except me…somehow…).

CloudFlare's 520

I naively thought 520 was like a 503 (the entire server is failing to respond to CloudFlare) or a 504 (gateway timeout - usually when node doesn't come back to nginx or CloudFlare), but it's not. It's more like "CloudFlare made a request and the response is incompatible".

The only real clue I had was that port 80, plain http wasn't affected. A helpful comment from @robobuljan showed that it was only the https version causing issues:

$ curl jsbin.com             # (works!)
$ curl http://jsbin.com      # (works!)
$ curl https://jsbin.com     # "error code: 520"

Although this digging took most of the day, it was the part that the LLMs really couldn't help with (and I had thankfully set them aside whilst I chewed on the meat of this problem).

As I looked for any clues. In the CloudFlare SSL/TLS page a section called "Traffic Served Over TLS" showed that there was a split in the supported TLS versions (I didn't capture a screenshot, but these numbers are from their API):

• TLSv1: 36 requests
• TLSv1.1: 56 requests
• TLSv1.2: 1,922,523 requests
• TLSv1.3: 5,216,795 requests

That's a lot on TLSv1.3, but I wondered how old (or actually new) v1.3 was given how old my rickety machine was. So I went back to my nginx config and found this line, repeated against each server block that ran SSL:

ssl_protocols TLSv1 TLSv1.1 TLSv1.2;

A conspicuous lack of TLSv1.3, and I wondered what would happen if the user asked for 1.3, CloudFlare tottered off wanting 1.3 and then got back 1.2… or perhaps nothing at all.

I first tried to add TLSv1.3 to the list only for the nginx -t test to fail (the module wasn't installed and I couldn't get it without doing major upgrade work). So let's see if it can be turned off in CloudFlare.

The short answer is yes, but I struggled to find it.

The first place to confirm what's being support is the section under "Speed, Settings". But to completely turn it off, it's located in "SSL/TLS" then "Edge Certificates" and towards the end of the page.

Disabling this unlocked a lot more genuine traffic and we were cooking again. Almost, nearly. Again, for some, now the static assets weren't loading nor was the frame that actually live runs the authored code (the null.jsbin.com domain).

JS Bin was dropping some users

It took me a few more hours to solve this last step, and I'm still not entirely sure how it all hung together. But if you recall, earlier in my nginx config, I had told nginx that if the IP of the request was from CloudFlare that we should use set_real_ip_from.

This specific command was writing the original IP of the requesting user into the $remote_addr, which is what nginx was now using to drop requests with the return 444. Somehow this wasn't in the main server block (the part that returns the index page), but was somehow in the static.jsbin.com and null.jsbin.com.

This kind of confusion is the result of working late hours, and working in crisis mode. I should have known better and I'm great at dolling out advice but sometimes don't tend to heed it myself.

After I finally removed the janky checks, IP swapping and quite a bit of the cruft that I had introduced with my pal ChatGPT, this let the last of the traffic come through correctly.

JS Bin was back. Entirely.

The aftermath

Now that CloudFlare is in front of the server, life is surprisingly chill on this 1GB single CPU machine. Way more calm than a normal day of traffic. Look at that CPU usage!

I suspect if I hadn't leant on the LLMs so much during the scream-face-the-server-is-down moments, I might have caught the complexity that I was adding. Though equally I should have put CloudFlare in front of JS Bin years ago - and not at a time of crisis.

I definitely learnt a few gotchas, the TLS and 520 status codes are that for me.

The traffic has eased out from the AWS CloudWatch logs and I believe CloudFlare is now bouncing a lot of that away from me - and it seems like a lot of that is coming from Hong Kong (which I've set to have a JavaScript based check to get through CloudFlare):

Since I took that screenshot, Hong Kong has 10 million requests in the last 24 hours.

Even though I don't think I'll ever really know what caused the amount of traffic that caused everything to fall over, my gut suspects scrapers for AI and LLMs just slurping up as much of the web it can. The only evidence I have against that theory is that the traffic didn't come from a single IP.

Whereas, weirdly, I saw a single IP bot scraping this blog, accounting for over 3GB of data and over 325,000 requests in a matter of hours. Thankfully this was running on Netlify and is entirely static, and not running on node 7 🤦.

Originally published on Remy Sharp's b:log

Being an long-in-the-tooth old dog, I tend to learn and trick, and roll it out again and again typically without taking the time to find whether I still need the trick. Case and point, I learnt about the JavaScript performance trick of ~~1.4 === 1 to floor a value (and the same float | 0) but really these days it's not "faster" than doing it the legible way (i.e. Math.floor(1.4)).

Given I've had a bit of time away from the backend, here's an unorganised list of things I've found I can use, and thusly remove extra code that I no longer need.

The list

1. CSS: text-underline-offset - the distance I can set the text-underline-offset: 12px
2. CSS: gap - no more faffing with margins in flexbox.
3. CSS: nested media queries on selectors (and nesting general):

h1 {
  font-size: 1rem;
  @media (min-width: 600px) {
    font-size: 2rem;
  }
}

4. CSS: clamp(min, variable, max) - I used this extensively on FFConf 2025's css, the process of finding the right values is very much
5. CSS: content: open-quote can localised quotes and the q tag does this by default, via this neat insight from Stefan Judis
6. JS: catch without catching the variable, let's me get past the 'error' is defined but never used:

try {
  doTheDodgyStuff();
} catch {
  // nothing, it's fine
}

7. JS: pointer events have improved (though still from experience they're not 100% perfect) but are to replace the old double click & touch handlers nonsense as seen in the W3C spec
8. AVIF images are fully supported - I benefited from a train ride with Jake Archibald and in chatting I discovered AVIF are very well supported. This means easily getting 50% file size savings on JPEGs. I'm regularly running the avidenc command in directories:

ls *.jpg | xargs -P 8 -I {} avifenc -q 50 "{}" "{}".avif

So that's my little list.

Not even 10 things. I guess I've not learnt much yet, but even though I return to the server side of APIs in 2026, I'm sure I'll be kicking the tyres in the front again soon enough and adding to my measly list.

Originally published on Remy Sharp's b:log

Grabbed it for the Christmas period (slightly annoyed I finished it on Jan 2nd). It's decently written, apparently part of a series (this being book 6) but stands fine on it's own. There's a few references to something that happened before, but doesn't affect the story.

It does read like a Columbo or Murder She Wrote, where there's people killing other people like it's just another Wednesday. All ends are fully tied up without anything left to the reader by the end.

I did figure out the killer around 1/3rd of the way in - or at least I keep asking the protagonist "What about X?!".

Not terribly Christmasy, but the very last chapter was very sweet.

Originally published on Remy Sharp's b:log

Longest book: Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI - 496 pages

Shortest book: The Time Machine - 107 pages

Quickest read: 3 days - The Radleys by Matt Haig (341 pages)

Longest read: 2 months, 2 days - Butter by Asako Yuzuki (464 pages)

Diversity of authors:

• Women: 9
• Men: 5

Rated books

5 stars

• Minority Rule: Adventures in the Culture War - 319 pages

4 stars

• Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI - 496 pages
• In Bloom (Sweetpea, #2) - 432 pages
• Sweetpea (Sweetpea, #1) - 480 pages
• The Radleys - 341 pages
• The Man Who Died Twice (Thursday Murder Club, #2) - 422 pages
• The Echo Wife - 347 pages

Books by decade

1890

• 1895: The Time Machine by H.G. Wells

1960

• 1969: The Left Hand of Darkness by Ursula K. Le Guin

2010

• 2010: The Radleys by Matt Haig
• 2013: Spike Milligan: Man of Letters by Spike Milligan
• 2017: Butter by Asako Yuzuki
• 2017: Sweetpea (Sweetpea, #1) by C.J. Skuse
• 2018: In Bloom (Sweetpea, #2) by C.J. Skuse
• 2019: Reasons to Be Cheerful by Nina Stibbe

2020

• 2021: The Man Who Died Twice (Thursday Murder Club, #2) by Richard Osman
• 2021: The Echo Wife by Sarah Gailey
• 2022: The Satsuma Complex (Gary Thorn, #1) by Bob Mortimer
• 2025: Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI by Karen Hao
• 2025: Minority Rule: Adventures in the Culture War by Ash Sarkar
• 2025: Making a Killing (DI Fawley #7) by Cara Hunter

Originally published on Remy Sharp's b:log

Reflecting and comparing

I've just read through my opening to the end of 2024 and wow, it doesn't feel better (to me). Though I do appreciate two things that factor into this: 1. I'm a glass half empty fella, 2. the coverage (I follow) of events doesn't often highlight the positive things happening.

The world is on fire. Still.

With this particular phrase in mind, I saw The Holdovers last week for the first time, and this quote struck me, for a movie set in 1970:

the world doesn't make sense anymore. I mean, it's on fire. The rich don't give a shit. Poor kids are cannon fodder. Integrity is a punch line. Trust is just a name on a bank.

The response to which is:

…if that's all true, then now is when they most need someone like you.

I think that's why we need to still feel like it's burning so that we can find that fire and help as do what we/I can.

Two particular legal cases that I'm trying to follow (though struggling because I suspect these legal things are slow):

• Palatine Action having been proscribed, and the legal challenge, at time of writing, "Judgment is awaited".
• The ECHR "interim guidance" (quoted due to the lack-thereof) on "woman" and "sex" refer to the biological sex - and the resulting challenge showing that it's transphobic and legally wrong.

A few other highlights being Labour did not swoop in and do what I had hoped (i.e. do right by people) and instead has spent a lot of their time trying to out-Reform Reform. The Giant Cheese Puff across the pond is tearing apart democracy, literally and figuratively and the genocide continues even after a "ceasefire" and the far right continue to raise it's ugly head in the UK. Oh, an AI eating the world and people actually dying from the companies pushing engagement (CW: suicide).

So… with that floating around in the background, I have found it hard to muster the motivation to care about the web, about bits of code and about writing on my blog.

I will say that I've found a little light in the political sphere. I regularly watch Pod Save the UK on YouTube, and found Zack Polanski's campaign to lead the greens - and a lot of what I believed. I've joined the Greens specifically to cast a vote and not only did he win, but the Greens have seen a huge surge adding over 120,000 new members (me included).

In a similar vein to doom scrolling, I found myself watching Polanski's videos and interviews online to drop a pipette's worth of hope back into my system. This was a particular highlight for me earlier this year.

With that out of the way, what have I been up to?

Work

Work-work (or as most people know it: paid work) has really been two main items: working with Oak National and working on FFConf 2025.

I first started working for Oak in early 2021 and have worked with them for just over 3 years in total. Ranging from rebuilding their question and answer system, to designing and building their public facing content API (where you, Jo Public, can access any part of their curriculum).

I really like working with the people at Oak, and it's the first place I encountered the concept of working in squads (which I see as like mini movie production crews - ask me in person what I mean!).

My last contract ended in mid-August this year, but I'm looking forward to returning to work with them at the start of 2026.

Then onto FFConf, and with my Oak contract ending, allowed me much more time to focus on the event, planning and new software.

FFConf

This year's lead up to FFconf was very, very different to 2024's lead up.

In 2024, the stress of trying to sell tickets, trying to find sponsors, and feeling like we were spinning our tyres and getting no-where was bad. Very, very bad. Like "I want to throw this all in"-bad.

When Julie and I returned to FFConf's planning for 2025 we knew this wasn't the last one, but we also knew we needed to do something different at the very least for my own mental health (and so Julie wasn't worried about me the whole year).

The relatively small and simple change we made was that we decided we would send a newsletter every week. The same structure, but weekly on Thursday.

This seemingly obvious change, completely changed the experience for me. There was none of the stress and terror that I experienced in 2024.

Weirdly, tickets sold mostly similarly to 2024, but it felt completely different.

As for the day. It was wonderful. I had a smile throughout the day, and I felt full of love from the people who attended. The talks were amazing and inspiring and fun. I wrote about the day too.

Projects

As I mentioned earlier, I've gone through a lot of the year not entirely excited about web stuff. I want to be, but… I'm just not right now.

A side effect of that feeling is that I've not really put much out into the world. I know I don't have to, but I also know that it's my creative outlet, and it's a nice thing for me to do (for me).

AI has been a dominant feature of the tech industries in 2025 - unavoidable in a lot of cases - but I have been dabbling in lots of little personal tools and toys using (effectively) vibe coding.

I find myself using this vibe coding method to create personal items that make something really specific to me, like: a tracker to help me and Julie to remember who chose the movie last week and what the hell we watched, or, a command line tool to scan my network and try to work out where my esphome devices are, or, a percentage calculate (because I keep forgetting how it works), or, a tool to modify subtitle files to offset the timing, or, a stupid toy that let's me upload a photo, cut out a mouth, then it uses speech API to sing whatever you type in.

But these are personal toys. Prototypes. Things that are not meant for more than one user.

I'd like to list and screenshot some of these in the future to preserve them a little better.

Donations

Each year, since 2021, I've been trying to look at donating at least 10% of the business profits to charities.

I put a call out on socials for anyone's lived experience with any trans support charities and the replies grew and grew (as much as I wanted to give to them all, I simply couldn't).

When reviewing which charities to donate to, we always check their income and want to see that they're also using their funds.

• Choir with No Name - homelessness and marginalised groups
• Forward Facing - support for families with critical illness or complex disabilities
• The Clare Project - support for trans, non-binary and intersex adults

Separately, I also donated to these companies (which aren't charities) from my personal donations:

• Trans Kids Deserve Better
• TransActual
• The Good Law Project

I hope to continue this "tradition" while I can.

Personal

Having touched on it already in the opening of this post, a lot of my personal headspace is in politics right now.

Outside of that particular stress, I've got my Gameboy and retro hardware that I restore (though much less often these days).

I also regularly tinker with my Home Assistant set up (something I've been meaning to write a little more about - there's some fun and dumb automations I have set up). A few of my favourites are: when I go to the gym, the bathroom radiator starts heating up (whereby my undies start getting warm in lieu of my shower when I return), or perhaps the increasingly aggro announcement to remind me to empty the washing machine.

The family keep growing (physically, thankfully the numbers have levelled off now!). The pets (3 cats and 1 dog) are happy, and the kids (one teenager and one pre-teen) keep sprouting and amazing (and testing us) each day.

The best books I read this year were both non-fiction (and thusly took me ages to read), but I highly recommend both:

1. Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI
2. Minority Rule: Adventures in the Culture War

I'll see if I can rustle up my reading summary in the new year - though at 14 books, it's rather brief!

Next bits - the retrospective

Last year I said I wanted to:

• Increase my reading ✅ - in 2024 I read 13 books, this year I've read 14 (highly unlikely I'll finish the 15th in time!)
• Find the motivation to blog some more ❌ - I didn't find the motivation, though I did actually manage to blog more than 2024 (though still not my target of twice a month, every month)
• Hopefully some journalling will help 🤔 - I started strong, then slowed down to when I needed to put thoughts somewhere, to now… I can't remember the last time I wrote (and writing by pen is a slow process ofr me, my fingers can't keep up with the thoughts) - this one is a semi-fail
• Keep an eye on my health ✅ - this is a mixed bag really. The MS support here in Brighton is really good, so they regularly meet me to check things are okay (MRI scans, drugs, etc). I have managed to return to the gym on a 2/3 times a week schedule. I had some physio that actually helped with some of my finger tip nerve issues (from MS) and it's improved the sensitivity for the first time in probably a decade. My diet…isn't terrible, but it's a far cry from dry chicken every day - I just think age is a reality I can't just ignore. On the other hand, my hearing aids are superb and I love them.

But hell, like I read somewhere on the web in the last week: 2026 is probably going to be a shitshow too, so don't be too hard on yourself.

This year, I'd like it if I found some way to get myself to write some more. I do wonder what software development is going to look like for me (particularly after how little code I've written from the ground up in the last 3 months). I'd like to keep up the reading and gym'ing. I'd like to get to bed earlier, and perhaps be a bit kinder inside my own head (to myself primarily). I want to ask "can I help" a lot more (if only to Julie) - it seems like a good mantra. I want to have fun with some stupid home assistant things. I want to design and build more PCBs and electronics projects (and perhaps improve my understanding).

And I'd really like to go to the cinema more. I miss the cinema. Maybe that'll be my thing in 2026 🤔

If you made it this far, thanks. If you're Remy reading this from the future - I hope things are looking up.

I'm now looking forward to reading your year-in-review blog posts, and I'll see you in The Future / 2026 💜

Originally published on Remy Sharp's b:log

Just the kind of read I wanted after a few longer books. Not 100% even throughout the book - but I'll come to that.

The squirrel from the cover features a nice silly-ish part of the story where the main character will often run into a local squirrel and add his own voice in a back and forth between him and the squirrel. The sort of thing I'd expect from a Bob Mortimer book.

I'm not 100% sure where the book's tone is supposed to sit. Some parts are particularly dark - police corruption, police harassment and modern day slavery. Juxtaposed next to funny little things like socks, squeaky shoes, a cranky neighbour and dog who's constantly disappointed with our main character.

The pages did turn quickly though, and it was easy reading. Nothing particularly to shout about, but not bad either.

Originally published on Remy Sharp's b:log

I really don't know why this book took me nearly 60 days to read, but it did which makes the events at the start of the book feel so utterly far away from the end.

I'm not sure what drew me to the book but I picked it up with zero expectations.

It discusses feminism, the pressure society puts on women to behave and appear certain ways (though this is Japanese culture, I do think a lot of this applies/is experienced by western white women too).

There's a strong feature of food and butter in particular (as you'd expect from the title) - and the author (and translator) does an excellent job of writing in a way that left me reading late at night and really wanting to taste the food (and I'm not a foodie).

There's also a story of friendship and support through those connections. This part of the book was really lovely to read, but seemed to all be crammed in at the end.

The story also includes a very manipulative character, Kajii, and whilst I read I felt sorry (and frustrated) for the protagonist, Rika, that she couldn't see the obvious manipulation (but perhaps that's the point, when we're close up to manipulation, we can't see the wood for the trees).

There really is a lot (of meaty subjects) in this book, and perhaps this contributed to my slow reading but it did leave me struggling to carry all the ideas through to the end of the book.

I've written about urgency in writing, and that's what helps me turn the page. This book doesn't have that, which isn't bad, but doesn't help me.

I suspect I would have enjoyed this more had I been able to digest it a little faster than two months.

Originally published on Remy Sharp's b:log

The code calendar

The advent of code is a daily code challenge, comprising of two parts based around the same problem, with increasing complexity. There's a wonderful little story that's weaved through the days, usually relating to Christmas and rouge elves that you can help.

I use it as a yearly chance to try to bend the jq language into shapes it definitely wasn't intended for. It's another way of clearing out some of the cobwebs that gather in my head (these are my attempts so far).

Perhaps you're new to software development or a particular language, then this is a great way of having real problems to solve where there's an absolute answer. There's also a wealth of solutions posted up on the Reddit channel if you (or I) get stuck and need inspiration.

Changes to the programme

As of this year, there's now only 12 challenge days (previously it was 24 - one for each day until Christmas eve). For me this is welcomed, partly because I'd usually break by day 16, but also, I'm not sure I want to spend hours tinkering on a hard problem on Christmas eve (having never gotten past 16, I hadn't yet!).

The global leaderboard has also been removed. I personally never made it anywhere near the leaderboard, but some people would obsess over it. Sadly, with the rise and ease of access of AI tools, it meant quickly the time from challenge release to posting a correct answer started turning up in seconds long. That's single digit seconds to solve the problem - which really isn't in the spirit at all.

AI was banned/asked to not join in with the leaderboard, but completely removing the global leaderboard completely takes the pressure off (and that idea that "why can't I solve it that fast").

You can still create a leaderboard for your friends or team - which makes sense.

Speaking of AI

Here's where my final suggestion might be controversial: why not try to solve using AI?

By that, I mean, if you're in a similar camp as me and have been sceptical of AI and a little wary, this might be a good opportunity to dip your toe in. I don't mean to paste in the challenge and have AI spit out the answer - that doesn't help anyone.

It could be seen as a chance to practise controlling the context and the problem space with AI doing the work. Perhaps testing different models from providers, but perhaps trying out local LLM models to see if they can do the work (so perhaps we could be a little more in control of the power usage).

As for me, I'm having to actively disable copilot in VS Code when solving the advent of code, because it's so desperate to help me!

Originally published on Remy Sharp's b:log

Whilst I sit squarely in my AI-sceptic seat, I was recently prompted to try a different tact by two post I read a few weeks back.

The first was Bruce's colonoscopy post, yes, that one. It was in fact that he was using a local LLM to create a generative image to commemorate his visit. I'd been using chatgpt (for code and electronics, basically a NLP version of a search engine), and hadn't considered that perhaps I could host myself to take some responsibility for one of the two problematic aspects of LLMs today (first being power consumption, second being the global theft).

The second post was Simon Willison's post on how he created a mini tool (the post is much broader than that, a useful read). It wasn't the post so much but the tool he was using: Claude Code. So far with chatgpt I'd copy back and forth errors and tweaks when it was helping me. Although I also have copilot enabled in VS Code it really holds down to autocomplete at a pretty junior level.

I'll often have typescript errors (for work) that copilot claims it can help fix, only resulting in even more typescript errors - so as a rule, I tend to avoid generating code fixes with copilot.

But what Simon showed with a shared transcript between himself and Claude Code was the software making the changes and offering diffs.

So that was what prompted a mini journey, and here's how it went.

Offline LLMs

Previously I had installed command line tools and even Ollama on my Mac without having the faintest idea how to use them effectively - so they sat idle and unused.

I'm not sure how, but I came across Msty, a tool that purported to make using local LLMs very easy. For a change, that seems to be true.

Since I still had Ollama running (though I should have probably ejected it), Msty quickly linked up to this and discovered (though I'd forgotten) that the DeepSeek R1 model was loaded already.

I figured since I was primarily using LLMs for software development, it made sense to find a model that suited those requirements.

Not quite that simple, or certainly not if you (ie. me) doesn't have the LLM lingo and terminology down.

I did some searching online, found a few articles, which themselves read as if they had been generated by AI, offering feature reviews of LLM models, but I couldn't find anything that said "model X is excellent for Y software development".

Additionally, I found some models on huggingface.io, and figured that 7B was going to be easier/more viable to run than a 70B (big numbers need big machines? I was guessing…). However putting the name of the model into Myst yielded a lot of results and I couldn't really tell you what the difference was. More importantly, I couldn't quite be sure of the provenance (and whether that even matters…I'd assume it did).

I ended up installing the following models:

1. llama 3.2 (2gb)
2. deepseek coder v2 (which blew up)
3. codellama (3.8gb)
4. codellama 13b
5. WizardCoder (I think the 3gb one?) (it just didn't reply on the query)
6. Wavecoder-ultra-6.7b (literally never worked for me)

I tried a number of technical problems without much success and did capture the output for a number of these models to try to solve a Home Assistant automation process (which you can see here).

It's entirely likely there's better models out there for what I wanted to do. I did know I didn't need a model to know everything, so some focus (and thus: reduced size) would make sense.

However, not one of these gave useful replies to my questions.

My favourite (fail) was giving code-llama a question that read: this query gets all the lessons matching these lesson slugs (I gave it the actual SQL query), how do I get a list of all the slugs that didn't match?.

The LLM happily reported that I was trying to learn Spanish:

The provided code contains a list of sentences that are written in Spanish, and the goal is to translate them into English. However, there are some errors in the provided code that need to be fixed before it can function properly. Here are some issues with the current code:

Missing curly braces around the translate function definition: The translate function should be defined as a separate named function expression, like this: { translate } = require("google-translate");. This will ensure that the function is properly closed and can be used in other parts of the code.

For some reason it was using the slugs in the query as the most important part and completely ignoring the technical query.

For the most part, the responses I got were fairly hand wavy, text heavy (which I didn't want since I was asking about code), and in most cases irrelevant to my task.

I think em0ry42 on BlueSky sums up what I was seeing:

Smaller self hosted models will always under perform the larger ones. I think your experience and those of the other commenters are consistent with the current reality of these technologies. They can't do what people are promising. Yet.

I'm sure there are people who can tune the hell out of their setup, but sadly, running any decent LLM locally as a useful code assistant, is just not here for the rest of us.

So I parked that for a while, and turned my attention to Claude Code.

Coding without touching code

I've no clue how new Claude Code was at the time, though I've gathered it's fairly new. It's a solid product from my experience (where I even managed to lose track as to which company owns which weirdly named AI thingy).

Setup and interface is entirely on the command line, so already we're speaking my language.

I'd seen demos of developers who've been able to join up their entire codebase to the LLM but each time I'd dabbling, I would quickly get lost and give up.

Claude Code does exactly this without the walls I'd experienced in the past.

I am however, acutely wary that Claude is running on a remote machine, and likely to be chonking through so much power that we're just throwing away water to keep machines from burning up. Let's stick a pin in that (and gosh, I loath myself already for that).

The very first problem I wanted it to solve was where I was trying to download 1,000s of videos and they all needed to be added to one massive tarball (context: this is for work, to allow users to bulk download our assets).

I'd hit a problem where the tar process kept throwing an unhelpful exception the evening before and no amount of documentation on the library I was using helped me.

Overnight I had a suspicion as to the cause and it gave me an idea to try - but I thought I'd let Claude try first, see what it does.

Without any specific direction (ie. my idea for the fix), and only the name of the file and the function the problem happened, Claude Code suggested the same solution I had in my head.

The UI then offered a syntax highlighted diff of the change it wanted to commit to disk. I was able to review it (very much how I'd approach a code review) and all I then needed to do was hit enter to accept.

I tested the code in a separate terminal and indeed the change worked.

Given this positive start, I then spent most of the working time split between the Claude Code UI and in the terminal to run the main program (which was sequencing a very large dataset).

The code changes for the most part were always good and code that I accepted.

The experience was…weird. I'd heard of LLMs being referred to as junior developers but when I was going back and forth between chatgpt and vscode (again, for me, copilot never really came in useful), because of the amount of interaction that was required from me, it felt even less than working with a junior.

But this was a much closer experience. I'd describe the change and logic, sometimes pointing to filenames that would offer useful context, and Claude would spend some time (and money) thinking, then it would ask me to check a diff.

Weirdly I spent more time sitting and staring out the window waiting for code to come back than I did looking at code. It was a weirdly hands off experience. I can't tell where I sit on that.

The main criticism that I had is that, because we use specific rules for typescript (no any and types are defined, which I think seems okay), Claude wouldn't really follow those strict rules, so I needed to go in at the end to clean that part up.

The secondary criticism is more a matter of taste. The code (and logging) was verbose to my taste. Additionally, being outside the code for the majority of the work period felt really strange. Sort of like a self-driving car took me the majority of my journey, deciding itself the navigation, for me only to be needed for the final arrival through some tight country lanes. Or something!

A cost

Since I was freestyling my way on Claude Code, I did manage to rattle through $5 of credit. I did think this was (somehow) linked to my Google business account, but I'm now suspecting it was free credit to introduce me to their API.

After running through this credit now twice (I switched to my personal account for a second run) I've discovered there's tools to help manage that sprawling cost (such as /compact and /clear to reduce how much context the LLM is fed before giving me a result). I'd like to play with this more to get an idea of how much I'm really prepared to pay.

Also after writing (most of) this post, I came across an interesting project that takes the Claude UI and lets you connect up your own backend. I've not tried it yet, but I'd be interested to see if I can connect to a local LLM and try out results (though going by the current experience, it's going to have a hard time competing).

Since it was conversational...

I decided to hack together a simple keyboard keycap (I had spare) with an ESP32 board to emulate a keyboard.

Then this would send a (fairly) unique keycode that then launched a python command which started a whisper based script that let me talk, then pasted the text into whatever was focused.

This meant I had: press button, say the thing, press the button, wait for it to be done.

It wasn't great because it was a little clunky, but it definitely felt futuristic!

How I felt afterwards

(I'm now writing this 8 months late, but I remember how it felt on the day).

Even though I was surprised at the progress of the work, both for how terrible the local code solving was and how impressed I was with Claude Code - it did leave me with a feeling of disconnect.

There's certainly the issue with the maintainability of pure vibe-coded software, but this was something more.

There's a creative input that I put into my coding process. A sense of purpose and achievement in solving some complicated problem, or writing a line of code that I'm particularly pleased with. There wasn't really of that feeling of connection with the output.

Having written this retrospectively I know that my perspective has changed somewhat, but I do remember have this weird dissonance between the outcome and the experience of getting there.

Originally published on Remy Sharp's b:log

So now as I sit writing this a full eight days later - sat on the side of the swimming pool as many kids, including my own, do their swimming lessons - I'm trying to collect my thoughts on the day.

The talks

Every year it's always an amazing experience, hearing the stories of speakers from across different walks of life.

The videos of the talks will be released in the coming weeks and our newsletter will highlight each in turn.

I curated the talks so that up front we had the heavier subjects: diversity from Hellen and data privacy from Chetan.

From the evening before we were talking about JavaScript, React, standards and new developers coming to the industry. We were talking specifically about whether the lessons had been learnt from the past or needed to be repeated.

I'm of the belief (though only in the last decade) that we need to repeat the learnings over and over and over. There's no point when the teaching is done. A fun example of this is seeing the CSS creative coders having their hay-day whilst hearing from the sidelines that the Flash developers have already done a lot of this. It's not new. But if you don't continue to pass down your lessons, the next generations will make the same mistakes and have to learn from scratch.

That's a very long way to say that we've covered the subjects of diversity and privacy before, but new perspectives and different experiences keep the subjects fresh and not a "done" task in the world of the web worker.

The next segment talked about web components from Hannah and about cults from Sèrges.

Hannah talked about her experience bringing web components to a React based team and the process integration both from the software side but also from the team adoption perspective.

Unsurprisingly (see previous notes on curation 😉) Sèrges brilliant (fun and funny) talk about cults dovetailed perfectly into Hannah's talk. She argued that we're all in a cult (to some degree) and touched on the gatekeeping that was also touched on in Helen's talk about team diversity.

Post lunch was time to discuss AI. When friends of mine who know me from outside the web, as about the conference, they understandably assume it's all about AI. It's all we hear about these days.

The two talks were very different perspectives on AI. Asim gave a talk (obscurely but made sense in the end) entitled Don't be an Idiot on how AI can, and is, being used to democratise big decisions, putting much more nuance in the decisions offered to people.

The following was Jess and Eda on AI coding tools and skills. Told from the perspective of a senior and junior and what are the lasting impacts on relying (or being pushed into) AI tools. Very much the argument of shallow learning versus deep (human) learning through experience.

Finally to close the event two talks from the vast spread of experience starting with Surya on the journey of 6 to 16 years old and coding. As our industry continues to exist, it's the young people who will breathe life into it, so I wanted to hear about Surya's experience (even if he is a bit of a black swan - making many of us old hats feel rather like underachievers!).

Then to close, we (FFConf) were lucky to have Sacha Judd (because she resides in New Zealand and it's a long old journey) filling our hearts with the joy of what the Good Internet is. Not just a nostalgic memory but alive today and just as weird as always.

Here are all the photos from FFConf 2025 if you'd like to see the lovely time people had.

Looking for meaning

The last few years of FFconf have been hard for me to put into context in a world that honestly frightens me. War. Genocide. Famine. Racism. Fascism. Hate. They're alive and well.

Sometimes I think it's ridiculous to put on a web event when things outside are so hard on people. In part it feels like a horrible privilege to be able to sink my head down and listen to these amazing speakers.

On reflection though, I do believe we need to recharge. That we need a little self care if we're going to continue fighting for a fair and equal world, however we achieve it. So maybe that's important for FFConf.

At the top of the event, I asked everyone to try to find one thing that you'll take away. There's so much in a day of talks - it's hard to remember what got you excited in the earlier talks let alone make sweeping changes.

So the (first) one thing I decided to do, relatively simple, was to move to more paid services (I've switched to Kagi already). I've already started paying for my news (404 Media, Novara Media, Guardian and a few others) so I'm continuing to look for these. I also know that this works if you have the funds. For me, it's part of my work, so my business pays for it. I get that it's not as simple for others (that's to say: this is my one thing!).

In my back pocket I'm going to make myself some either silly or partially useful little toys. I may or may not share these; I like that these can be for me and family/friends: low pressure nonsense.

Next

FFConf will return next year, Friday 13 November. I've already got a couple of speakers in mind so I need to approach them with my proposals.

I'll be attending more events myself next year too, including State of the Browser, Web Day Out, (the final) All Day Hey, Beyond Tellerand and WDC (tickets aren't live yet). I'm on the lookout for more events and meet ups in particular (though I have a stinking record of staying at home).

I know this might read as a hopeful note, and I do want it to be, but I also know that I don't 100% buy my written down optimism (I should!). Still, I hope to see you out there.

Originally published on Remy Sharp's b:log

Source: thingsithinkithink.blog

Originally published on Remy Sharp's b:log

The extremely quick read is, add /* HTML */ to the front of the template. Case sensitive and space sensitive (though hopefully one day it won't be so strict). Now highlighting and prettier (with save and fix) works.

Note that you need the es6-string-html VS Code extension for this to highlight correct (something I had forgotten I had installed).

The slightly longer read

What I wanted to achieve was that I could include my template in the source of the web component (or at least in the same directory - i.e. physically near to the application logic).

The problem is that unless I use a build function (or tools), the markup for the template is a string.

Here's the starting point:

customElements.define(
  'hello-world',
  class extends HTMLElement {
    connectedCallback() {
      const n = this.getAttribute('name') || 'World';
      this.attachShadow({ mode: 'open' }).innerHTML = `
      <style>:host{font:600 16px system-ui;color:#222}</style>
      <div>Hello, ${n} 👋</div>`;
    }
  }
);

Seen here with VS Code's syntax highlighting - note that the template itself is just green, plain text:

Using a tag function solves the main issue, but requires extra code or a magic build process which I really don't want or need - and look at that nasty red snake telling me I need to write more code:

VS Code does support highlighting if you give it a hint using a comment: /*html*/, but prettier doesn't format it - close, but still no dice:

Finally, if you get the comment syntax exactly right, that's uppercase and with spaces around the text, /* HTML */, then you'll get both highlighting and syntax tidy support without the need for build tools:

I do have the VS Code option "prettier.embeddedLanguageFormatting": "auto" in my settings, but if I've understood correctly, this should be the default for prettier and not required.

A large hat tip to Timothy Leverett on Mastodon for helping me.

Originally published on Remy Sharp's b:log

Source: automerge.org

Originally published on Remy Sharp's b:log

This question came up recently by someone who had either attended back in the early days or that knew of our event, but they (understandably) saw us as it was in 2009.

I think the vibes you came away with in the 2009 and 2010 events would be recognised, but content and the core messaging has changed quite significantly.

The event was founded on the premise that there weren’t any (dedicated) JavaScript conferences in the UK at the time, and I wanted something to fulfil my own desire to see that change. I wanted an event that I would want to attend.

The first line up in 2009 was mostly brought about by asking the question: who do I know or could reach that talk about JavaScript things? I can’t say for certain, but I suspect the outreach was "can you speak" rather than "can you speak about…".

Immediately I knew I wanted a rule that prevented speakers from repeating (this was a preventative measure to get me to reach outside of my circles).

The follow 2010 event still had a JavaScript flavour but was already showing signs of morphing into something different (albeit I wouldn’t start working on the diversity until 2012).

I remember Dan Webb (who spoke in 2010) saying that he realised at the end of the event he was the only speaker to show code in his slides.

By 2018, the 10th conference, I think the entire shape had changed. Talks like "Mentoring: Being the help you wish you'd had", "Dear Developer, the Web Isn't About You" and "Weird Web & Curious Creation" were all based in the idea of helping others and ourselves to become better Web Citizens.

People who had attended told me that the talks they saw at FFConf were talks they never expected to see - in a positive way. "Capitalism, The Web, And You" was an excellent example of that in 2019 along with "Working towards a greener world from behind the keyboard" or "Programming with Yarn" (although I've only included 3 talks, they're all superb, or certainly to me).

The talks at FFConf continue along this line of thinking. How can we learn to be better versions of ourselves - and so that we can help others.

From the talks on the day, you won't learn how to add aria roles to a tabbing system, but you will learn how your junior developers are being affected by learning from LLMs spouting code at them. You won't learn what the latest additions to JavaScript syntax is, but you will learn how the latest technology has been used to exploit your privacy and the ethics of how developers' decisions landed them there.

You won't learn what a blog post, a tiktok or chatgpt can give you. You'll come away inspired and fired up and (hopefully, usually!) excited to work. It's a hard one to measure, and hard to explain to a business "what's the ROI?", but if we are happy, keen or even excited to work, then the ROI is always a net positive.

So that's what FFConf is in 2025. It's an event chooses people over technology, that aims to inspire and motivate, and that champions a better web.

Originally published on Remy Sharp's b:log

Source: blog.johnozbay.com

Originally published on Remy Sharp's b:log

I do struggle reading non-fiction, and this book was no different. It took me nearly 2 months to finish reading (in my evening snippets) - but it was definitely worthwhile. It was this 90 minute interview by Novara Media with Karen Hao that made me immediately purchase the book.

I don't think I can fully review this book and do it justice, but I can share what I learnt from the book where my original assumptions were wrong about AI and AI companies.

The book is focused primarily around OpenAI and Sam Altman, and in particular how his ousting in November 2023 came to be, through hundreds of interviews and documents, and paints a very insightful picture of the messiah complex (though not in her words) that Altman has.

Here's a short summary of things I didn't know but learnt through reading this:

1. Altman and friends started conversations (and business, albeit initially as a non-profit, though that didn't last) in 2015 because they were going to build AGI and that it was inevitable.
2. The company was formed with scientific researchers originally, again as an open company with the intent of sharing, though spoiler alert: this all changed
3. I had always assumed it was cowboy bro developers working on the code, but it was, originally, academic based engineering
4. AI safety was constantly there and initially significant - but as we all know now, eventually lost a battle to make an impact, left "hobbled" and thrown rather to the wayside.
5. The training data was, after GPT-2, ingested wholesale and attempt to clean/sanitise would happen on the results coming from prompts - i.e. the inputs were not cleaned, which means applying dizzying array of filters to catch on the output and edge cases.
6. Common Crawl was introduced at GPT-3 - which is also where the input filtering stopped happening
7. AI, or Western AI companies including OpenAI (but also Google and Microsoft) put their data centres in the Global South, additionally sourcing their data annotators from the poorest countries allowing them to pay (via third parties) literal pennies per hour for the work (which could also come with terrible mental health side effects as the worker would read and view the generative content that AI could come up with based on the unfiltered dark corners of the web)
8. Sam Altman lies. Little lies, but from a great deal of documentation, a lot and often to tell people what they want to hear whilst (we guess?!) having some ulterior motive
9. The path that OpenAI decided to take to head towards what they believe will be AGI, effectively requires unlimited compute power, when in reality, there are lots of different applications of AI that don't need that level of power, Stable Diffusion being one such example trained using 256 GPUs (still not a desktop computer, but not hundreds of thousand GPUs either)
10. OpenAI's approach, to close off it's scientific findings, close it's source and refusing to share methods means that there's no way to verify any of their progress, but more importantly is stripping the academic scientific community of it's researchers (as someone who has visited CERN on two occassions, seeing science being shared is incredible and incredible for society)

My only complaint about the book (and it's likely to be my own fault) is I had trouble with the jumping backwards and forwards in time - I'd often be unsure where we were in the timeline.

If you work in tech, I'd absolutely recommend this book. If it's not possible, then definitely the interview I linked above.

Originally published on Remy Sharp's b:log

Source: nielsleenheer.com

Originally published on Remy Sharp's b:log

Source: gradient.style

Originally published on Remy Sharp's b:log

Unblocking these different services just to get a physical product to work (naively I thought it worked mostly offline) felt like I was exposing myself to a pretty gross network.

Then had to make an account for my child, whom if there were over the age of 10 and under the age of 13, could be managed under my account. Once they turned 13 (in the eyes of Meta) they were completely unrestricted (which I've got A LOT to say about, but I digress).

This combined with having already having to lift the protection I had in place was too much. I quickly deleted both accounts and parked the problem for the day.

After some reflection, I decided that I would create a single account, that we'd all share that way at least I could monitor all activity and the data Meta collected would be (hopefully) extremely mixed due to different people using the account.

Which reminded me of a story I heard about years ago where villagers in some areas of Africa (and likely other places in the Global South) were sharing a single device to visit Facebook and the like, which made it nigh impossible for algorithms to profile the user and made targetting and ads effectively useless.

Why don't we do this now? Why don't we share accounts to large social media sites to pollute the signal we're putting out?

I understand if you want to create an account so you can build your brand or post what you're up to, this doesn't really work. But if you're visiting these sites to keep in touch with your communities and post short comments or questions - then it seems to me that this is a viable solution.

If I were to propose implementing this, it would be that each shared-user would have 10-20 real people using the account, I can see how it might be chaotic with more people - plus there's an aspect of trust required in your group. If someone decides to change the password you're all of a sudden blown up.

Funnily enough, I've written about this same thought some 6 years ago, prompted by this fun pollution tool.

What do you think? What downsides am I missing here?

Originally published on Remy Sharp's b:log

So today, the 30th of August, grief wakes me up in the morning. It presses hard downwards on me. And I need it, even if it's just for a day.

15 years didn't fix us, or heal us, time doesn't do that, but it does allow for new memories. For more memories. For life to fill in the space. It's that time, and that life, and our amazing two children who create the life and family we needed. It keeps us busy and it keeps us alive.

We only got one day with Tia, to hold her. Then we had to leave her. Alone in the hospital. That's my memory that's fraught with pain and longing, and all kinds of complicated feelings. For a long time, in those early years, it was a dark memory, a memory I hated myself for, for leaving her. It took a year of therapy to put light into that room again, to not cast judgement on myself, to leave the memory as it is. Just that one day. The one day that she was real.

But that's why I write these post each year. It's why I write this for myself, and in part to share with you. Tia is a memory to me, but she was real. I have to write about that to bring her existence into reality, into the world I still inhabit.

I was worried this year I wouldn't find the words, that the grief wouldn't find me. I know that there's an instinct in me to shove the grief away, to busy myself, but I also know that every other day of the year can do that with no effort at all. The little one starting secondary, the big one starting his options, new achievements, new problems, new arguments, new memories. So, I'll try my best to sit in the grief today. Share it with Julie, and try to be gentle on ourselves.

And I'll remember Tia today.

Originally published on Remy Sharp's b:log

I think I expected the same kind of murder spree from the first book, and it's not that at all (although Rhiannon, the protagonist, desperately wants that). It's actually written on the tin, In Bloom is Rhiannon's experience of being pregnant - which, for a change, rolls up all the shitty experiences into one.

I enjoyed, and laughed at, the cutting turn of phrases - the kind of brutal honesty that you might reserve for the closest of friends knowing that it's for the "inside voice".

I can see myself reading the next instalment in the book (though perhaps not so soon) and I'm not entirely sure where the character can go given the way the book ends, so that's definitely intriguing.

But just for the the record (my record!), Rhiannon is a terrible person!

Originally published on Remy Sharp's b:log

Since I'm getting my Amazon Kindle book highlights already but am also actively trying to buy books from anywhere that isn't Amazon it means my highlights are in KOReader.

There's a multi-step process that requires special command line tools and navigating KOReader into the "right" state to allow for syncing (I'm certain there's other ways to do this process, but I've not found it yet myself).

1. Export the highlights

Probably the simplest part, and if you're happy to connect the Kindle via a USB port, probably a lot easier to get the exported notes off (but I'm taking the wireless route, so it's trickier!).

Whilst the current book is open, open the "tools" menu (the spanner and screwdriver icon), then Export highlights and Export all notes in current book.

I've already selected the format as JSON and selected my export folder (arbitrarily my exports are going to /mnt/us/koreader/clipboard).

Now the file is waiting to be lifted to my computer.

2. WebDAV

From the machine I want to upload the JSON file to, I need to run a WebDAV server (a technology I've always known about but never really had an need or use for until now).

There's quite a few options for WebDAV servers, but I wanted something very lightweight and that I could run on the command line. I ended up using a Go based server, via brew install webdav.

Which ever server you use, you'll need to make sure it has write permissions. To run this go version in write, in the directory I want the JSON file uploaded to, I run:

WD_PERMISSIONS=CRUD webdav -p 8181

This says:

1. Run on port 8181
2. Run with create, read, update and delete permissions

Once I'm done with the process, I then terminate the WebDAV server.

However, before I shut the server down, I need to upload my JSON notes from the Kindle.

3. Uploading to WebDAV from KOReader

This is where KOReader's UI gets clunky again. You need to use the "Cloud Storage" option.

Cloud Storage is only visible in the tools menu when you're in the file browser (not in the book reading mode).

Once the Cloud Storage is open, tap the plus icon on the top right, then fill out the details. Assuming the Kindle is on the same network as the computer (because it needs to be), the URL/host is http://{IP}:8181 and the folder is just /.

Once this is saved, tap on the connection name. If the directory is empty, you'll see a new screen with no files. If there's any .pub ebooks, then those will be listed (and could be downloaded).

From this new screen, tap the top left plus icon to upload a file. Navigate to the file, and long press on the file to "choose" the file to be uploaded.

If all is successful, KOReader should say so. If it fails, it could be related to permissions. I had earlier luck with a python WebDAV server.

4. Transform

The JSON structure is fairly simplistic, but I have this specific jq transform to get into the format I use:

def slug:
  ascii_downcase
  | gsub("[^a-z0-9]+"; "-")
  | gsub("(^-|-$)"; "");

. +  { slug: .title | slug } | {
  "\(.slug)": {
    title,
    author,
    highlights: .entries | map({ text, page, note })
  }
}

Alternatively, if I'm lifting the annotations (highlights) directly from Calibre, this works:

def clean: tostring | gsub("\\\\\\n"; "\n\n") | gsub("[’‘]"; "'"; "g");
def percent($total): . / $total * 100;

def slug:
  ascii_downcase
  | gsub("[^a-z0-9]+"; "-")
  | gsub("(^-|-$)"; "");

. as $_ | { slug: .stats.title | slug } | {
  "\(.slug)": {
    title: $_.stats.title,
    author: $_.stats.authors,
    highlights: $_.annotations | to_entries | map(.value | { text: .text | clean, page: "\(.pageno)/\($_.stats.pages)", note: (.note | clean) })
  }
}

That's how I get my notes from non-Amazon bought books into my own blog.

Originally published on Remy Sharp's b:log