AI
EvoDuet: Bilevel Co-Evolution of Web Searching and Task Solving for Scientific Discovery
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 29/09/2026, 20:00:00 UTC
Briefing created: 01/10/2026, 09:20:57 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 03:18:24 UTC.
Briefings
EvoDuet is a bi-level optimization method that co-evolves solutions and search queries using fixed LLM parameters. It uses a retrieval gate to assess knowledge gaps and dynamically select new or reused documents, while an inner loop ranks documents by predicted solution scores. The method was tested across 21 optimization tasks with one candidate per iteration, using GPT-5.6-Luna and Gemini-3.8-Flash, achieving a 74.1% to 78.0% normalized discovery gain increase on OpenEvolve. It also improved performance on Sums/Diffs and Denoising tasks with other scaffolds like Top-K and EvoX.
Key points
- EvoDuet uses a bi-level optimization method to co-evolve solutions and search queries with fixed model parameters.
- The method achieves a 74.1% to 78.0% normalized discovery gain increase on OpenEvolve across 21 tasks using GPT-5.6-Luna and Gemini-3.8-Flash.
Context and limitations
Performance varied by model; Qwen3.5-9B did not benefit. Best runs surpassed previously reported best scores on eight tasks but did not match them on three others.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 09:20:57 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeia5xaqjfwo67mp3z22aqwope2xxtjubemk5ao4de3cj7zto2gotti