Loading activity…
@natashajaques
Assistant Professor leading the Social RL Lab https://socialrl.cs.washington.edu/ @uwcse and Staff Research Scientist at @GoogleAI.
AI
A post describing the project reports Qwen3.6-35B-A3B rose from 41.1 to 53.3 on Terminal-Bench 2.1.
AI
The researchers simulate diverse stakeholder perspectives so a model avoids answers that any affected person would veto.
AI
The proposal’s authors say a single model takes on different roles that learn to give more varied answers—but only when those answers meet a quality threshold.
AI
The author says toy experiments also show that post-training with random rewards does not improve capabilities except under very particular circumstances.
AI
An author of Tapes Together Strong says cooperation evolves from scratch when social behavior, computation and reproduction share one energy budget.
