- Type
- Provider research
- Publication
- Google DeepMind
- Version used
- arXiv v2
- Source date
- 4 October 2024
- Recorded in Atlas
- 23 October 2026
- License
- Original source rights apply; Atlas explanations and artwork are original.
Date of the consulted version, not a claim of independent replication. No separate validity period has been established for the finding.
How Atlas uses this source
When agents learn from experience
This changes the trained policy, unlike adding a note to a model input.
Open the source passageSupports this claim:Training is a different route to better correction.