Google Research posts Harness2, a recursive agent harness
The only available text is a Hacker News listing with 43 points and nine comments.
9 stories, newest first · one card per story
The only available text is a Hacker News listing with 43 points and nine comments.
Google Research's RRSI uses noise bands, cost rules, and leakage screens for safer self-improving AI agents.

A coding guide demonstrates Google Research's Kauldron, a modular JAX trainer using plain-data configs and string wiring.

Google Research open-sourced RRSI, letting frozen-weight agents improve their harness without overfitting benchmarks.

A tutorial implements sound encoders for Google Research's MSEB and scores them on four audio tasks.
Google Research's RRSI regularizes recursive agent-harness edits, lifting held-out scores while cutting policy tokens.
Google Research's RRSI regularizes recursive agent-harness evolution, lifting out-of-distribution scores while cutting policy tokens 30%.

Google Research introduced R4T, an RL-trained fan-out pipeline distilled into a 53.9M-parameter diffusion retriever achieving 12x-20x faster query fan-out.

Google Research and partners introduce ToolGrad, a verified tool-chain-first data generation framework reaching 99.8% pass rate and boosting Gemma-3-12B to 83.1 on BFCL.
Today 1.6K readers · 2.4K views · 775 reads · all time 189K views