Twitxy Saved

William Isaac

Verified account This account is verified. Learn more

@wsisaac

Principal Scientist @DeepMind | Previously @OSFellows & @hrdag. RT != endorsements. Opinions Mine. Pronouns: he/him| @williamis.bsky.social
London, England wsisaac.com Joined April 2008
1.3K Following 6.3K Followers 11K Posts

Latest posts 5

William Isaac Verified account @wsisaac
Great thread on our new research understanding how to bring workers back into the training of AI systems!

@charvvvv_ @SolomonMg @kweku_gdm @kl
Charvi Rastogi @charvvvv_ 11h
In which my colleagues @SolomonMg, @kweku_gdm and I discussed AI and jobs with @econormist . As I highlight in this piece, the question is markedly that about AI and workers! 

aipolicyperspectives.com/p/the-jobs-que…
View quoted post on X
William Isaac Verified account @wsisaac
Very interesting set of papers and thinking about multi-agent + coordination!
Rohan Paul Verified account @rohanpaul_ai Nov 7
New Oxford paper finds that frontier LLM agents fail to coordinate with unfamiliar partners because each one treats its own made-up rules (arbitrary convention) as the obvious choice.

Agents from different companies increasingly must work together with no agreed rules. Past studies used shared scaffolding or complex games that hide why coordination fails.

They tested text and code agents built on models like GPT-5.2 and Gemini 3 Pro in 9 small cooperative games. In Cat & Dog, a safe move guarantees 5 points, while a signal convention pays 10 only if both agents pick the same one.

Paired with other models, text agents scored well below that safe 5, each calling its own arbitrary convention natural. A short shared statement of coordination principles, called a Coordination-friendly definition, led agents to reliably take the safe 5.

If your agent must work with agents you did not build, tell it to avoid unstated conventions and prefer actions any partner reads correctly.
View quoted post on X