Harness^2 operates at two recursive levels.
Recursive harness edits help agents adapt without answer keys
Harness^2 tests competing changes to an agent’s runtime scaffolding, then retains lessons that can guide adaptation on later tasks.
Big Tech
Ruiyao Xu · Yanfei Chen · Zhongying CuiZhu · Bhavana Dalvi Mishra · Yifei Ming · Han Yu · +3 more
Google Cloud AI Research · Northwestern University
Research Digest··2 min read
Xu and colleagues introduce Harness^2, a framework that modifies the prompts, memory, tools and orchestration around a frozen language model while it tackles previously unseen tasks.
Why this paper
From Google Cloud AI Research and Northwestern University
In one line
Recursive harness improvement without ground-truth feedback increases LLM agent performance by up to 11.8 percentage points.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (3 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§