Noesion Radarnoesion.ai
Research2026-09-16

Researchers from Google, Google DeepMind, University of Maryland and University of Virginia published Dream-RSI. It treats past discovery histories as replay simulators for refining exploration policies. A meta-layer loop collects histories, builds simulators and redeploys improved policies online. It showed gains in algorithm engineering, mathematical optimization and GPU kernel engineering. The paper and project page are live. The arXiv posting is in progress. Code and discovered programs are being prepared for release.

What it unlocks

Using past discovery histories as replay simulators to refine exploration policies without rerunning discovery agents.

  • 1.22x faster downstream runtime
  • 1.74x less discovery compute
  • 162x fewer calls than SimpleTES
  • 2 of 3 tasks at or above baseline

What you need to act on it

  • code release pending
  • discovered programs release pending
  • reproduction scripts release pending

Sources

Send this to someone who needs it

Shares the story and its sources. Nothing about you.

What does this mean for your job?

This is the story as everyone gets it. Once a week we send you the version written for your role — what changed, why it matters for the work you actually do, and one thing to try. Free while we tune it.