getbasis.ai has the most stunning website and demo video

Genuinely very impressed by the SVG of a pelican riding a bicycle I just got out of Google's new Gemini 3 Deep Think model simonwillison.net/2026/Feb/12/...

coding assistance lets you fast-track important projects that improve your life, such as reformatting every single site you read into the old vbulletin 3.x default template

Anatole becoming really bloody useful

very interesting seems 'Explore' agent being used a lot more now

media matters www.economist.com/culture/2026...

Two new speech-to-text models (similar to Whisper) from Mistral today - one of them is API-only, the other is a 8.9GB Apache-2.0 licensed open weights model for "realtime" transcription. They're both very good! simonwillison.net/2026/Feb/4/v...

subscribe to the media you want to support sometimes it won't make a difference but sometimes it will

The CIA just stopped publishing their World Factbook and took every page, including the archived copies of previous versions! This sucks. It was public domain, so I recovered the 2020 edition (the last one published as a zip file) and shared it to GitHub simonwillison.net/2026/Feb/5/t...

I cover quite a lot of ground in this talk, but the bit I hope people take away from it is that AI-driven development poses dangers that we're not talking about nearly enough. This clip was recorded before we knew how prominently Elon features in the Epstein files: 🎞️ youtube.com/clip/UgkxImh...

Hottest ticket in town @protocolsforpublishers.com

Codex keeping the tone warm and secure is maddening.

tragedy of the commons

Big day! www.producthunt.com/products/ana... Been testing Anatole with a group of patient, thoughtful beta testers for some time now And it never feels ready to launch. But at some point, I think you just have to put it out there. After all:

Jordan at Takara is using embeddings for drug discovery. He's working with labs to to turn drug discovery into something more like a language problem. This is one of the really fun things about working on media recommendations with Anatole:

Never thought I'd enjoy using Slack again, but here it is. Pressing the 'plus' tells Anatole that this article about the new X algorithm was good. So I get more like that.

slack literally covering up the other half of the menu with 'AI' now

Try using Gemini to track your competitors, I dare you.

intelligencedividend.substack.com/p/the-intell... by @jamierumbelow.bsky.social is excellent

First draft. anatole.fyi

Dropping a new landing page on Monday.

the thinking man's 'may I meet you'

We released a Svelte Skill! It will teach your agent to use the svelte CLI to invoke the same tools as the MCP! We also released a subagent so that docs and iterations with the autofixer don't take up precious context! And you can get all of that easily because the repo is now a Marketplace! 🧡

Thanks to @mbanerjeepalmer.bsky.social feedback on Svalettab (Palettab browser extension in Svelte) there a couple of new features now 🤌 A load more new fonts and disable the loading spinner (which is literally there for whimsy alone 😅) github.com/spences10/sv...

"What a stunning photo essay, what a shame it isn't built in...hang on!" www.economist.com/interactive/...

fun: cloudflare outage means mistral is responding with html

We are finally making a svelte skill for claude (and turning the repo into a claude marketplace so you can install the MCP and the skill with two commands). Please check out this PR, there's a zip file with the skill that you can download..try the skill without the MCP and report if it works! 🧡

Government consulting on foreign state ownership of 'Newspaper Enterprises'...bizarre www.gov.uk/government/c...

I hate literally every second of using Google Cloud

20% of US adults say they now regularly get news on Tiktok, up from 3% in 2020. By age, 43% of 18-29s often get news on Tiktok, 25% of 30-49s, 10% of 50-64s and 3% of 65+. www.pewresearch.org/short-reads/...

I accidentally launched a podcast search app on Product Hunt this morning it's currently closing in on #9: www.producthunt.com/posts/vol-2 why accidental? after hitting publish I wanted to polish the svelte app a bit more, but couldn't cancel so yeeted it into the future then forgot about it...

Finally: social features limping back into Spotify newsroom.spotify.com/2025-08-26/i...

Substack is only site to report year-on-year traffic growth in August among the top 50 English-language news sites in the world

We're working to fix common issues when working with Svelte and AI. If you use AI with Svelte, we want to hear about the specific issues you've encountered. Your feedback will help us build an official MCP server and provide suggestions to AI models through static analysis. forms.gle/FhVGVAMCLBpw...

I think most of my bsky network is either 'ML researcher' or 'AI sceptic'. If you're in the latter group you're going to violently disagree with this: mbanerjeepalmer.substack.com/p/should-pro...

Lose the ability to read complex prose and he fears you may also lose the ability to develop complex ideas that “allow you to see nuance and to hold two contradictory thoughts together”. The medium is the message, and the message is currently 280 characters long. www.economist.com/culture/2025...

Must-read. Eliot is spot-on. Notes: 1. Every media consumer I've seen wants a truthful, quality, diverse media diet. (Yes, even *those* ones.) 2. But base motivations (curiosity, status, anger, mood) shape behaviour. Aspirations come from System 1 but we live in System 2. So we need to align them

It seems clear that legacy publishers are growing more and more comfortable building editorial products around the personal brands of their individual journalists. Yes, there's risk to this strategy -- the most prominent being that the journalist could eventually leave to launch their own competing

Gmail bizarre spam filtering reaches new level as it filters out Google's own culture newsletter

Finding key information shared at council meetings takes time and effort. Today, we've launched a tool that allows you to search verbatim quotes from council meetings in the UK and Ireland, without the need to watch long recordings or sift through extensive notes. council-search.bellingcat.com

Mightily tempted to build 'study LLM behaviours as a service'. So many of these scientific papers are literally just evals. 1. We wrote a set of questions 2. We assigned scores to types of responses 3. We asked the questions to a set of LLMs and scored the answers 4. Journal article

On the other hand: - Friends who are good at writing React are very productive with LLMs. - But (1) you need to be good at React in the first place (2) you can't ask it to write something unfamiliar. - We should help LLMs unlock similar productivity for people who can already write Svelte.

We just presented at Protocols for Publishers on building an internet that actually works for news 📰 Big thanks for @unternet.co for having us and @neeldhanesha.com for the writeup!

Using LLM via API to write a huge pile of regexes offline so that the online classification is dirt cheap and lightning fast (once translated to Rust, 13k items classified in four seconds). by @alexparsons.bsky.social www.mysociety.org/2025/08/12/u...

Haven't tried, but very excited to

new `catalina` model on @lmarena.bsky.social stands out, need to find out what it is

Shout out to @logfire.pydantic.dev for being surprisingly helpful. Certainly lots of rough edges. But compared to other platforms, which seem to be built for an o11y in-group, Logfire says "read the docs? lol we know you cba, copy-paste this instead"

probably worth reading in full!