RL Environments Explained: How AI Agents Learn Real-World Work | Brendan Foody, Mercor
- Source
- youtube.com
- Author
- Sequoia Capital
- Date
Context
A reinforcement learning (RL) environment, as Mercor's Brendan Foody explains in a Sequoia talk, is a practice setting for AI agents with three parts: a simulated world of messages, documents and files from a realistic project; clones of common work apps the can operate; and tasks pairing a with a verifier, either a rubric or unit tests, that scores the result. Experts are needed because, outside areas like math, a cannot reliably find its own mistakes in something like a slide deck, much as a student cannot grade their own homework.
Foody's example is a legal environment written by lawyers from top law firms. Foody reports that further of a GLM model on 1,800 such tasks, using about $500,000 of compute, raised a corporate-law score from 4.7 percent to 26.6 percent, with smaller gains on other benchmarks. That is Mercor's own result, from a company that sells this data, and the talk does not say whether the tasks were held out. Foody also discusses Mercor's business in a Lenny's Podcast interview.
- Agent — A system in which a model repeatedly chooses operations, a harness carries them out, and the results return to the model to inform its next choice.
- Prompt — Information or instructions supplied to a model; in a chat, the request you type is often called your prompt.
- Model — A computational system that uses learned patterns to produce an output from information supplied to it.
- Training — The process that adjusts a model using examples and a measure of how well it performs.
videoBringing Continual Learning to Every Company: Trajectory’s Ronak Malde
videoCrusoe CEO Chase Lochmiller: Turning Stranded Energy Into Intelligence
videoDavid Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience
videoCracking the Code on Offensive Security With AI ft XBOW CEO and GitHub Copilot Creator Oege de Moor
Checking sign-in…
Loading comments…


