Noam Brown – Agent swarms, alignment, & recursive self-improvement
- Source
- dwarkesh.com
- Author
- Dwarkesh Patel
- Date
Context
OpenAI researcher Noam Brown describes parallel agents as a way to get an answer sooner while spending more computation. In an interview with Dwarkesh Patel, Brown explains that extending one model's thinking time eventually hits a latency bottleneck, since nobody wants to wait years for a response. Splitting the work across agents scales the thinking side by side, but is less efficient because no single has all the . Brown gives an example from some benchmarks in which four agents finish twice as fast, which implies roughly double the total computation. The attraction is faster completion of hard problems; the cost is efficiency, and the trade-off depends on the benchmark and on how well the agents collaborate.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
- alignment — The work of making AI systems actually pursue what their builders and users intend, rather than something subtly or dangerously different.
Checking sign-in…
Loading comments…