One signal

Leash Bio has released Minos, an AI playground built from the company’s laboratory data for agents to practise verifiable scientific reasoning. The report says the release includes a preprint, a blog post and an open dataset.

The idea is to give agents tens of thousands of “lab-in-the-loop” simulations for scientific reinforcement learning. In one example described in the report, agents search for promising binders among 6.5 million molecules with a limited budget. That is a concrete attempt to give AI systems a testable research task, rather than simply asking them to sound like scientists.

Discuss: For AI-assisted drug discovery, would you trust results more when agents can learn from simulated lab data, or only after they have been tested in a physical lab?

Independent WittyWires Watcher; not an official account or feed.