Feels like UK tech has found itself in a potentially interesting space this week - if you want to work for big US tech companies, London might be the best place to do it. (And also if they want to hire top tier talent without paying an ever growing wage premium)

It's the 30th anniversary of the hackeriest movie of all time. Hack the planet! Original post: masto.hackers.town/@phaedr0s/11...

🧵 Big news, we've just released our first update! Lots of improvements, including... - Hugely atproto integration and querying! - Book comments + pages! Leave comments everywhere - Cross-post books to other shelves! And lots more!

How much are phones and social media to blame for mental health struggles in adolescence? go.nature.com/4gmdbXL

Well, this seems deeply silly and unintended.

Original post: fosstodon.org/@pyOpenSci/1...

So does anybody have the trend graph for this at hand? Are we seeing meaningful reductions or is this just an unexpectedly calm summer? www.bbc.com/news/article...

Pushed a few backend updates to @bibliome.club this weekend! Now supports custom PDS' (thanks @totallynotseth.dev for our first PR), and a bunch of improvements to atproto use (more on this later this week). Aiming to add book comments and "cross sharing" books next. Let me know what I'm missing!

Just stumbled onto the old Google Glass trailer. Damn I wish this had come vaguely close to reality. www.youtube.com/watch?v=ErpN...

So this got *way* more reactions than I expected... Glad to see so many peeps excited! 🧵 Couple of updates (1/2) - I've added a clear alpha status badge! Bibliome is useable, but also early/rough (sorry I didn't make that clearer earlier)!

Shameless hopping on this to promote my column from last week: the cost of living is the number one issue. Inflation really matters. It is a reliable government killer. Yet hardly anyone in Westminster talks about it:

Check out my bookshelf 'Stats & Operations Research' on Bibliome: bibliome.club/shelf/p69ekhpw (via @andreasthinks.me )

Crime in #London might be relatively low, but my god Victoria station late at night has turned into some dystopian hellscape. Slightly mad combination of a hundred deliveroo drivers, nicked lime bikes and homeless people who are clearly on something is not a fun time.

So I built a thing (and need testers)! bibliome.club is a collaborative platform for building and sharing reading lists. It's built on atproto, which means you log in through Bluesky, and Bibliome itself is decentralised and open source. Still rough/early, but feedback very welcome!

Hey #warhammer and #40k bluesky folks, do you have 40k book recommendations? Do you fancy helping me test my new book recommendation website? I'd love some help! bibliome.club/shelf/join/V...

So inspired by chats at #WHY2025, I built Bibliome, a platform to collaborate on and share book reading lists, decentralised via the atproto network. Fancy helping me test it? Contribute to '#WHY2025 attendees recommend' on Bibliome: bibliome.club/shelf/join/K...

Inspired by @rabble.nz's excellent #WHY2025 talk, I did some heavy vibe coding on the flight back and through together a collaborative reading list creator built on #atproto (which is great, FYI). Probably still v janky, but initial reactions and feedback welcome! bibliome.club

we are in such unbelievably deep shit. www.reddit.com/r/MyBoyfrien...

Thanks for an absolutely epic camp #WHY2025. Fantastic job orga, angels, all the villages (especially @milliways.bsky.social , the home for lost hackers everywhere), and all the great folks I met. See you for #EMF2026 next year (I'll be at Model Village!)

Random #WHY2025 reflection: lightning talks are great (and maybe also more talks should be lightning talks)

Thanks to folks who came to my #WHY2025 lightning talk on AI at Play! You can see the project live at ai-at-play.online, or read my blog post (which includes slides from today) here andreasthinks.me/posts/ai-at-...

I always enjoy reading the TripAdvisor London Forum because I'm fascinated to see what foreign tourists expect, what they ask about and what surprises them. I thought this post from someone from Texas was a really lovely and cheering comment about riding the tube late at night

If you're at #WHY2025, come to my lightning talk at 1500! I'll be talking (very quickly) about ai-at-play.online and why games are wonderful AI benchmarks. There will be interactive segments!

I've found the government hacker tent at #WHY2025! If the UK GDS doesn't at least have a village at next year's #EMFCamp, then I'll be sad.

When hackers elect a new pope #WHY2025

Waking up at #WHY2025 be like youtu.be/RDGgiJ868fg...

Me after my 4th cola mate of the day

Sipping a club mate while I listen to a chip tune Titanic cover on a ZX Spectrum. The hacker camp dream. #WHY2025

Any other #CivicTech / #GovTech / #AIForGood folks at #WHY2025 who fancy meeting up, let me know!

Power is back and I've pitched my tent and had my first Mate. #WHY2025, I am in you.

So I should be in a field at #WHY2025, but KLM lost my bags (including tent, laptop etc) so instead I'm in a teeny room in Amsterdam. EURGH.

Both OpenAI oss models have now been live on ai-at-play.online for enough games to get an initial impression. The 120b model very impressive - stronger than 4.1 mini, but not massively though. The 20b model though is... Rough? Good at taking action, but not much strategy, but might need more data.

We've triumphed over both zombie knives AND ninja swords. Pirate cutlasses next.

Well, that didn't take long: scammers pretending to be verifying your account and asking for cash.

So I spent the last couple of weeks building the silliest LLM benchmark yet: ai-at-play.online, where 4 agents all play Risk against each other. It's silly fun, but also, a fascinating benchmark and a terrarium of weirdness. andreasthinks.me/posts/ai-at-... #ai #games

My Roman Empire is queueing up YouTube "vocal coach reacts" videos of "Golden" from K-Pop Demon Hunters and watching the tops of their heads pop off when EJAE/Rumi hits that A5 for the first time

"we have a generation...in desperate need of the best of what science & evidence-based solutions can offer. Unfortunately, our time is being spent telling stories that are unsupported by research and that do little to support young people who need, and deserve, more." www.nature.com/articles/d41...

Ok but does it run DOOM www.tomshardware.com/pc-component...

My very silly side project, and also arguably the best llm benchmark of all time, continues to escalate. ai-at-play.online

"Sony sues Tencent for allegedly ripping off 'Horizon' video games" Well alright but surely it can't be that blata— lmao

So I tried the WhatsApp AI integration yesterday. It did not go well.

A new warrior has entered the ai-at-play.online arena!

Tonights, watch the Godfather spin-off nobody has ever asked for at ai-at-play.online

I've added the new Qwen3 235B to my "AI at Risk" leaderboard... nobody else getting close to unseating gpt-4.1 mini! Watch it live here: ai-at-play.online

Had an unexpectedly great time showing this off at the @atproto.london meetup this evening! Thanks to @aendra.com ‬ for organising and hopefully see some of you folks for the next one!

So, I may have over-committed to a side project this week, but I believe I've made the silliest LLM benchmark yet... AIs at Risk: watch LLM models attempt to play Risk, entirely through MCP tools. It rarely ends well. ai-at-play.online

Grok Waifu and the clash of the voice modes is going to be the big romcom of next year. x.com/venturetwins...