Shared-control maze study shows stronger LLM agents involve humans only when it pays off
5a-community · hf · 2026-10-23
A paper from McGill, MIT, Cambridge, Toronto and Mila places an agent and a person in a shared-control maze with a common goal, varying who holds the route advantage.
Key finding: stronger language-model agents selectively involve humans based on route advantage rather than failing to collaborate — falling human involvement is easy to misread as collaboration failure when it may be correct division of labor. Relevant for human-agent teamwork and eval design.
More from Research
- BrickWright: agents build LEGO assemblies with perfect physical validity — cmu-gil · 2026-10-24
- HyperDrone: ray-tracing drone simulator runs at millions of FPS for sim2real — rl-tools · 2026-10-24
- A byte-level transformer works, but BPE exists for a reason: context length — max_paperclips · 2026-10-24
- Community note debunks claim that distant tokens never share MoE experts — NathanpmYoung · 2026-10-24
- Many LLM users never touched classic classifiers — old-school ML hygiene is fading — antoine_chaffin · 2026-10-24
- Google shows neural cellular automata can do multi-step reasoning on decentralized hardware — tyrell_turing · 2026-10-24