This is a linkpost for https://time.com/7204670/uk-ai-safety-institute/
Seems like quite a well-informed piece, discussing among other things:
- AISI's relationship to the major AI labs
- The challenges of evaluations research
- What impact AISI could have