This is a big deal - and further underlines the amazing progress the UK has made on state capacity in AI. Congratulations to the whole AISI team!

I haven't read in full yet, but this risk framework by Center for Research on Foundation Models seems pretty robust. On the uncertainty side, very nice to see support for forecasts of how model capabilities will improve would influence assessments of misuse efficacy and scalability. crfm.stanford.edu/open-fms/

What I’m reading this week 📚
- Kapoor et al. on the impact of open foundation models: bit.ly/socimpactfound…
- RAND’s emerging tech risk analysis for ‘synthetic pandemics’: rand.org/pubs/research_…
- The preprint for Evo, a new ‘genomic foundation model’: bit.ly/EVOpreprint

Come work with me! I can't emphasise how fun I find the various challenging problems I get to work on, and think this is a great opportunity to develop valuable skills at the interface of research and policymaking alongside a caring, hard working team.

Deadline: March 8, 2024

'We need a science of evals'

If we want to be able to trust the result of evals in high-stakes situations, we really need to understand the process and methodology deeply.
apolloresearch.ai/blog/we-need-a…

AI:bio risk recommendations from FAS is now @scientistsorg policy sprint are out now—check out mine and Richard Moulange's proposal, and the work of my very talented colleagues here: fas.org/publication/bi…

This is a useful (and impressively quick!) analysis, though this is mostly assessing companies' *stated intention* to comply with best practices, not evidence of them actually doing these things - so a bit overoptimistic to say they are already meeting best practices

Excited about helping shape the UK’s approach to responsible innovation through the UK’s Biosecurity Leadership Council!

(And great to see Department for Science, Innovation and Technology already moving on implementing aspects of the BSS!)

gov.uk/government/new…