For those interested in advanced AI safety work that directly impacts policy globally, I highly recommend checking out the UK's AI Safety Institute. Very talented, driven team that's accomplished a lot in the past year, catalyzed by the UK AI Safety Summit. They're hiring lots

Proud to have contributed to @AISafetyInst's testing of Claude 3.5 Sonnet (new), focused on assessing Claude's refusal behaviour in the face of adversarial prompting strategies! 🧵 on our findings 1/N

Join my team! One of the most important and interesting functions in AISI imo

🎉 Huge congratulations to AISI researcher Hannah Rose Kirk for winning a Best Paper Award at #NeurIPS2024! 🏆

If you're at NeurIPS tonight, come to the UK AISI social to meet our teams working on agents, post-training, cyber misuse evals, safety cases, societal impacts, science of evals, and building our SF office. Prof Yoshua Bengio & British Consul General Thomas Codrington will also

Control work seems very promising to me and UK AISI is a great place to work on it. Consider joining!