Neel Nanda (at ICLR)

I don't think I've mastered the skill of interpreting these visualisations, but they're SO PRETTY. Great work by Catherine Yeh!

Great paper and elegant set up! This is another nice illustration of how it is so, so easy to trick yourself when interpreting LLMs. I would love an interpretability project distinguishing faithful from unfaithful chain of thought! Anyone know what the smallest open source model…

Great new work from @ch402 on the differences between representing activations with directions for independently varying features vs a dense yet structureless code. I really appreciate the commitment to writing great exposition!

Great article from @DavidSKrueger on the various dynamics that makes the community underrate AI X-Risks. My favourite line:

I really enjoyed @yonashav's paper on how we might create a world where all large training runs can be monitored - feels like the best AI governance proposal I can recall seeing! Makes me optimistic there are policies simple enough to be realistic, but useful enough to matter