The EU should launch a talent scheme ↘️

🚨I’ve got a new paper out! External evaluations are often cited as a critical tool in the AI governance toolbox. However, their efficacy depends on evaluators having sufficient access to the system being evaluated and accompanying information.

I remember when technical safety folks obsessed with 'solving technical alignment first' gradually started acknowledging the role of governance, policy and regulation. Now gradually many are also seeing all the other concerns relating to agency, labour, power etc. This is great

For those interested in EU AI Policy, applications are now open for the Talos Fellowship Autumn 2025. talosnetwork.org/fellowship (There's also an international AI governance specialisation) I went through this program years ago, and it was great. I'm happy to answer DMs on this!

Ariane 6 launch with two Rafale trying to race it.

My biggest takeaway from this is that Jack Clark has great taste in fiction because There Is No Antimemetics Division by qntm is top-tier stuff (Also AI is a big deal and governments should take it far more seriously)

Pour sauver la France il va falloir boire le champagne, le cognac et la chartreuse que 350 millions d’américains n’auront plus

Introducing Mistral Small 3.1. Multimodal, Apache 2.0, outperforms Gemma 3 and GPT 4o-mini. mistral.ai/news/mistral-s…

Happy St. Patrick’s Day to the powers that be

Applications to our Q3 Research Fellowship are now open! → June 30 – Aug 29 in London at the London Initiative for Safe AI → Work on AI safety with the guidance of your experienced mentor and research manager → £5,000 stipend + meals, travel & housing support (link in bio)

Every time AI timelines shorten I get 5% closer to buying the Elizabeth line couch

Perplexity hiding all the best models from us

Check out AI Policy Bulletin, a new digital magazine - I've appreciated the variety of content on whistleblowers, chemical standards, export controls, among others, by great writers like Anton Leicht I recommend following their page: aipolicybulletin.org

As the article gets at, this is spooky for two big reasons: 1. It again opens the floodgates to a huge lobbying battle on privacy 2. If a policy as fundamental as GDPR can be reopened, then almost anything is on the table

new Consistently Candid episode with Conjecture co-founder Gabriel! we talked about why alignment is hard, what it would take to solve it, what ordinary people can do to help reduce AI x-risk and more link below

📢 Applications for our annual Research Scholar roles (US, EU, and Legal Frontiers) are closing on June 22! 📍 Washington, D.C. / Cambridge, UK / London preferred, remote options available 💼 $100,000 – $175,000 📅 Full-time, flexible hours More details in thread 🧵

There are half a million civil servants in Britain. I used to be one of them. They have a huge impact on government. But they don't have a public voice on policy, and their leaders rarely make public interventions about their work. So we surveyed them! Here's what we found🧵

It’s like u go 2a shop,when you come out, ur money’s gone but you’ve bought things. What’s important’s what u purchased is worth the fortune