Leash releases Minos: turning physical experiments into replayable worlds for agents
madhavsinghal_ · x · 2026-10-16
Leash Bio's Ian Quigley introduced Minos, a system built on the idea that agentic scientists can do science: make choices, observe physical consequences, and decide what to do next. Since real physical experiments are too slow and expensive to run thousands or millions of times, the team scaled experimentation in a way that makes the experience reusable.
They measured millions of small molecules against many proteins, then turned those results into replayable experimental worlds that agents can explore at compute speed. Across many thousands of experimental choices, different models behave very differently and use chemical feedback to inform their decisions—revealing not just what agents discover but how they behave while exploring uncertain worlds.
A preprint (MinosLeash.pdf on GitHub) and a data release (strelka.leash.bio) accompany the post, which opens with Turing's 1948 "thinking machines" proposal and Shannon's 1949 work on machine chess, arguing that games, cryptography, and math are good test cases because they require little contact with the outside world.
More from Research
- Embodied First Exam benchmarks coding agents on 1,090 tasks across 44 robots — ManlingLi_ · 2026-10-23
- Anima Anandkumar on neural operators: AI weather models beat supercomputers — AnimaAnandkumar · 2026-10-23
- Scaling GLM-5.3-Flash training to 1M token context, 64x over Megatron baseline — AndrewLBeam · 2026-10-23
- Experimental boolean-logic LM: microsecond forward pass on one CPU thread, but painfully slow training — cephaloform · 2026-10-23
- AdaDex turns seconds of human hand motion into generalizable dexterous robot skills — mangahomanga · 2026-10-23
- ReAPO paper: conditioning retries on failures beats GRPO in RLVR reasoning training — jasondeanlee · 2026-10-23