Combing a multi-agent objective (e.g. number of humans alive ever) with a single-agent action space (what can I do at the margin, when almost everything else is a given) is a self-fullfilling recipe for disempowerment.
Last week I won my first poster prize @ICSA_senescence 2026!! The study is now on bioRxiv!
https://t.co/wtQa4Scsd5
With @Avelar_Ageing & @HiroFujiwara@RIKEN_BDR we created ECMExplainer, a novel tool to interpret fibroblast phenotype. mini 🧵; (1/5)
Zero-Shot Coordination among LLM Agents
Adrian Hayler, Shashank Reddy Chirra, Andrei Lupu, Johannes Forkel, Bidipta Sarkar, Siheng Feng, Jakob Foerster
https://t.co/mvNNOLueWQ [𝚌𝚜.𝙻𝙶 𝚌𝚜.𝙰𝙸]
humans: I don’t know
papers: it is uncertain
Astra/Opus: the epistemic landscape remains undetermined and awaits clarification
as models grow increasingly capable they have learned to speak “research paper” English on steroids!
I work at the intersection of neuroscience, machine learning, and genetics. Empiricism and modelling are essential to understanding, and these skills may atrophy over time if we simply pursue short-term output-oriented goals! (that's not to say one needn't be productive!)
AI can help us search in certain domains and problem situations, but we conclude that human hands should stay deeply involved in the work of exploration and judgement to avoid the individual and societal costs.
Blog: https://t.co/GZpCX7Sczm
Paper: https://t.co/BZVaCDYhLF
(11/n)