2026
- Your Agent's Notes Are the Closest Thing It Has to Training Data
TL;DR: A new paper from Meta Superintelligence Labs, UC Berkeley and others measures how well frontier-model agents improve when their weights stay frozen and everything they learn goes into files: …
- Why Your Cheapest Model Should Write the Harness
TL;DR: A May 2026 paper separates two capabilities that self-improving agents usually conflate: writing harness updates and benefiting from them. Writing is flat across model tiers: a 9B open model …