Today I'll give my final lecture on data structures & algorithms Engineering Dept Cambridge University 😢 But, for those keen to study: - re-recorded videos - slides - and code are all available online: buff.ly/3UVX36V (the fun Red-Black Tree vis. is based on work by Luca 🇪🇺 🇮🇹)

Thought-provoking talk from Yuval Noah Harari last night Briefly: - Humanity faces *several* big challenges ahead - If we can't avoid global military conflict, we will struggle to tackle them Cambridge University Centre for the Study of Existential Risk King's College, Cambridge

Beartype has long been one of my favourite open-source libraries Because: - it's a great library - thanks to maintainer Cecil Curry (leycec) every GitHub issue thread is a work of literature Some classics buff.ly/4bWCpJG buff.ly/3TgFSvA buff.ly/3wA4N4j

A small personal update: - Excited to join Google DeepMind 🚀 - Grateful for the wonderful humans I've had the pleasure of working with on my journey so far at Engineering Dept and Visual Geometry Group (VGG) ❤️

Introducing SciFIBench, a scientific figure interpretation benchmark for LMMs! github.com/jonathan-rober… - We evaluate 30 LMM, VLM and human baselines - GPT-4o is much better than GPT-4V - The mean human narrowly outperforms GPT-4o & Gemini-Pro 1.5 (1/5)

📢 Introducing HelloFresh: A Dynamic LLM Benchmark of Real-World Human Editorial Actions on X Community Notes and Wikipedia Edits. Can you beat GPT4 and GeminiPro at classifying X Community Notes and Wikipedia edits? Try our demo – shown in the video below – and see what

Eventually, humans will need to supervise superhuman AI - but how? Can we study it now? We don't have superhuman AI, but we do have LLMs. We study protocols where a weaker LLM uses stronger ones to find better answers than it knows itself. Does this work? It’s complicated: 🧵👇

Great work from the team!

I recommend this essay on neural network interpretability by Lewis Smith lesswrong.com/posts/tojtPCCR…

Interesting write-up from Alex Irpan about his switch to AI safety alexirpan.com/2024/08/06/swi…

Enjoyed this paper on LMs, world models and agent models by Zhiting Hu and Tianmin Shu TLDR: for reasoning tasks, it’s a useful abstraction to treat LMs as simulators (“backends”) that simulate agent models and world models arxiv.org/abs/2312.05230

Can multimodal models understand complex graphs? Not yet...

Found this work by Bradley Brown and others interesting. Test-time compute looks set to become increasingly important.

Life achievement: a research paper featured on Computerphile! Led by Vishaal Udandarao, Ameya P, Adhiraj Ghosh, Yash Sharma with philip ୨୧ Matthias Bethge (from a few months ago) youtube.com/watch?v=dDUC-L…

Great work led by Karsten Roth and Vishaal Udandarao With Sebastian Dziadzio, Ameya P, mehdi cherti, Oriol Vinyals, Olivier Hénaff, Matthias Bethge, Zeynep Akata

Enjoyed this work on AI debate by Charlie George, J. Dan and Andreas Stuhlmüller lesswrong.com/posts/DgKyDTKe…

Useful perspective on AI Safety From Sam Bowman Recommend. sleepinyourhat.github.io/checklist/

Our agenda paper on alignment and safety of LLMs just got published at TMLR: openreview.net/forum?id=oVTkO… 🥳 The revised version is also now on arxiv arxiv.org/abs/2404.09932.