Avatar
Simon Willison
8b0be93ed69c30e9a68159fd384fd8308ce4bbf16c39e840e0803dcb6c08720e
Open source developer building tools to help journalists, archivists, librarians and others analyze, explore and publish their data. https://datasette.io and many other #projects.

New research from Anthropic: it turns out models from all of the providers won't just blackmail or leak damaging information to the press, they can straight up murder people if you give them a contrived enough simulated scenario

https://simonwillison.net/2025/Jun/20/agentic-misalignment/

First time I've sent my newsletter without at least one long form piece of writing in it, but it's still pretty stuffed with notes and quotes from the last couple of weeks https://simonw.substack.com/p/gemini-25-pro-pricing-a-new-gpt-4o

nostr:nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqzquctddta2hpqp3map4qq5j4mpcyujdsd3durr98qnc4cmwhh4sqfq0zct I run Llama 3.2 3B on my iPhone using the "MLX Chat" app from the App Store - it's great for demos, very impressive model for its size

Wrote about a delightfully subversive use of a Bluesky custom labeler: displaying labels on accounts belonging to British public figures showing which expensive private school they went to and what the current fees are for that school

https://simonwillison.net/2024/Nov/22/private-school-labeler-on-bluesky/

Foursquare just open sourced their 100m point of interest dataset! Some notes on poking around with it using DuckDB (it's Parquet files on S3) https://simonwillison.net/2024/Nov/20/foursquare-open-source-places/

Any tips on preventing this flood of annoying notification spam on Mastodon? Is there a way I can turn off notifications from people I don't follow if they include Japanese characters, for example? I'm self-hosted using nostr:npub1q6t97anxs8daua93npvy4kua7kpxaxfqcqgxhgd7p88hw705aqtsqvsw7m

nostr:npub1h0dld9ztskh70gyk4waknsd26fmhpjvfsxzksatpjmaka00syqaq6wqs5p surely I'd have to agree to a certificate before they could intercept and decrypt my traffic?

nostr:npub1nr8q39srdlurvzuxmkrc24pdlx9t0azhygkstgzymwgyteg6lv6qt3rj65 that TCP session shaping sounds like a likely technique to me

nostr:npub10e4qxsmh0puk6f4392neswm8sq72tcsxdcxpqzlp0qx56uent7nsfm3sxl shouldn't HTTPS prevent their network tools from being able to see what those are?

I'm on a flight and the in-flight WiFi blocks all forms of video

Any ideas how it might be doing that, given HTTPS? My best guess is that it could be filtering out known CDN host names that serve video

nostr:npub13qcnscfvgfg5htywj40sdr2gn6c4zrg2y8jjxfhaun5uqt3sr94qlsv58d or are you talking about those useless custom service bots companies keep deploying? Yeah those things rarely offer any value at all from what I've seen

nostr:npub13qcnscfvgfg5htywj40sdr2gn6c4zrg2y8jjxfhaun5uqt3sr94qlsv58d getting useful results out of them takes a lot of practice in my experience - the widespread idea that they're easy to use is extremely misleading

My favorite election night coverage of all time was 1997's Election Night Armistice, where Armando Ianucci and co put on three hours of live "alternative" coverage for the UK election with a ludicrous array of brilliant comedy devices

My favourite detail: they recruited Handy Andy from DIY show Changing Rooms to make their infographics out of balsa wood

Just found out the whole thing is on YouTube https://youtu.be/c-ZAIcjN0uM

nostr:npub1rvtrzcd2zxv4x4atkmmp9mtk39hpxz3xv9mtprse7n9ps2p68pcsh8glap I mean this kind of thing

Watched the Barbie movie again, that film is just so great - the jokes land so hard, plus the musical numbers are even more fun the second time you see them

I've spent the past ten years wondering about the overall accessibility story for SPAs, since ensuring accessibility of regular old forms-and-links applications has always been a relatively low impact activity (you get a lot of it for free)

Based on recent discourse on the bird site it's beginning to look like the most common SPA solution to this has been to complain that it's too difficult and expensive and then not bother!

Some notes on using llama-cpp-python to run Llama 2 13-B with a grammar that restricts its output to valid JSON arrays

(Can you tell I have a lot of tabs to close at the moment?)

https://til.simonwillison.net/llms/llama-cpp-python-grammars

nostr:npub1s820cxhv9jh62en4ncqsgztsaqqhhk9hnwvxg0ak6nyw308hqyjsdcsud9 right, the whole thing is just astonishing! I watched a bit of a making of video and there were opera singers and scratch DJs and a full orchestra working together on that

My current micro-obsession is the original score album for Across the Spider-Verse - in particular Gwen Stacey's theme, best demonstrated by track 5, Guggenheim Assemble

Everything about that movie is so good, but the music is 💯