Geoffrey Irving Says We Can Just Do Things
Chief scientist at UK AISI reacts to reported pauses in reinforcement learning runs.
Geoffrey Irving, Chief Scientist at the UK AI Security Institute with prior roles at DeepMind, OpenAI, and Google Brain on alignment, posted the line You see! We can just do things. The post tags a tweet and sits beside a generated headline stating OpenAI and Anthropic each implemented brief pauses during reinforcement learning training. The generated summary adds that the moves followed calls to slow frontier AI development before uncontrolled events occur.
Both OpenAI and Anthropic just had a mini-pause on RL training
Combined views
22K
4 posts, first seen
Reactions
4 postsRO 
roon@tszzlpacing the frontier success story (?) next time let’s do it proactively before there’s any absurd loss of control events
- likes: 338
- replies: 23
- bookmarks: 30
- reposts: 21
RO roon@tszzlbtw I think it’s pretty amazing how @sjgadler @/pagehedley @SnehaRevanur @j_asminewang have made “pacing the frontier” a thing. excellent policy coinage, it will be in the history books
- likes: 40
- replies: 5
- bookmarks: 2
- reposts: 2
SA Steven Adler@sjgadler@tszzl @SnehaRevanur @j_asminewang This is very kind (also cc @_NathanCalvin) but also, there were many many contributors to this I’m happy it’s been a useful concept though: It had been hard to convey that ‘speed change’ doesn’t nec mean slowing down from today’s speed (though…
- likes: 12
- replies: 0
- bookmarks: 0
- reposts: 0
GI Geoffrey Irving@geoffreyirvingYou see! We can just do things.
- likes: 16
- replies: 0
- bookmarks: 0
- reposts: 0