🤩We are hiring a postdoctoral researcher to work on AI safety and multi-agent safety. This is a fully funded position in collaboration with Toyota Motor Europe (TME). More info and apply here: tinyurl.com/5c9jjcdy Closing 5th September Adel Bibi Engineering Science, Oxford

Time to update the LLM speed runner benchmark? Minqi Jiang (thanks Edward Grefenstette for spotting this in time for editing)

🚨job alert🚨Foerster Lab for AI Research is looking for a postdoc, deadline is 1st of September (my.corehr.com/pls/uoxrecruit…) By many reports we are on the global Pareto frontier of talent density, agency and ambition. Join us!

LLudites

🚨 Research Scientist Hiring alert! 🚨 Applications close THIS FRIDAY for Google DeepMind research scientist roles to work on autonomous assistants and human-facing agentic capabilities, self-improvement, and open-endedness.

I recently had a lunch time conversation with a very senior AI researcher about how are multi-agent problems differ from single agent (their starting point was they do not). One point that made them think: As computers scale, the rest of the world (i.e. no agentic parts) is not

When I stared my phd in deep learning in 2015 I thought I was late to the party. When I bought a few nvidia shares over a beer at NeurIPS Conference 2016 I was sure I had missed the boat (given a 3x run-up) but told my peers "better late than never". which boat did you miss?

Multi-agent AI is taking over the world.

We deployed 44 AI agents and offered the internet $170K to attack them. 1.8M attempts, 62K breaches, including data leakage and financial loss. 🚨 Concerningly, the same exploits transfer to live production agents… (example: exfiltrating emails through calendar event) 🧵

Excited to present this work at RL_Conference next week!! Please message if interested in coffee, lunch, etc. 😃

🇨🇦 Next week, I'll be visiting Alberta to present our paper at RLC 2025! Please reach out if you're around and want to meet up to chat about algorithm discovery, meta-RL, or anything else!

Random "Hot Take": air bubbles in milk foam add insulation and keep cappuccino hot longer. Same principle works for pastries with fillings (Cornish pasties), pigs-in-a blanket, or having a layer of fat on top of a soup. Even baking potatoes develop a layer of insulation on the

This is cool! Would be great to know how adversarially robust it is. As in - if I train an RL agent to complete a task inside the world model, how exploitable is the model? Bringing this to real time is quite amazing, too.

🥳 It’s an honour to have been awarded the Outstanding Paper for Scientific Understanding in RL at RLC for our work, ‘How Should We Meta-Learn RL Algorithms?’ Thank you to the organisers RL_Conference for putting on a great conference, and congratulations to the other winners!

Check out the full story of attention sink, written by my former intern Guangxuan Xiao, who did the work in Meta😀 two years ago. This is now used in OpenAI's new OSS models. As the main mentor I am really proud of the streamlined research procedure: Getting curious about a