Each DuoDiscover worker selects an optimization target and constructs a fast proxy evaluator, called a yardstick.
Agents devise cheaper evaluations to scale autonomous engineering search
DuoDiscover pairs optimization targets with fast proxy evaluators, while periodically checking those proxies against expensive real measurements.
Big Tech
Shubham Agarwal · Akshat Gupta · Mike Heddes · Dilara Soylu · Rui Meng · Patrick Tser Jern Kon · +18 more
UC Berkeley · Stanford University · Google
Research Digest··3 min read
Agarwal et al.
Why this paper
From Google and 2 others
In one line
DuoDiscover scales autonomous engineering optimization by co-discovering targets and cheap proxy evaluators, while fidelity checks and true evaluations prevent proxy gaming.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks (6 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§