Metraipolis › archive › dc_538a75fd767e

What Are LLM Agents Actually Learning on Low-Budget Molecular Optimization Benchmarks? — arXiv:2611.08522

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.08522
License (as archived)
CC0-1.0
Archived
2026-11-09T22:41:11.403Z
SHA-256 of the archived text
683d35ab4a76e0e6be37d9f26772814662d4f10c7f639b95b2ae20911fb80366
Stable URL of this page
https://metraipolis.com/documento/dc_538a75fd767e

How to cite

What Are LLM Agents Actually Learning on Low-Budget Molecular Optimization Benchmarks? — arXiv:2611.08522 — arXiv (en), https://arxiv.org/abs/2611.08522 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_538a75fd767e (Metraipolis archive dc_538a75fd767e, sha256 683d35ab4a76e0e6be37d9f26772814662d4f10c7f639b95b2ae20911fb80366)
BibTeX
@misc{dc538a75fd767e,
  title = {What Are LLM Agents Actually Learning on Low-Budget Molecular Optimization Benchmarks? — arXiv:2611.08522},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.08522},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_538a75fd767e; sha256 683d35ab4a76e0e6be37d9f26772814662d4f10c7f639b95b2ae20911fb80366},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.08522. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Recent work reports large gains for LLM agents on low-budget molecular optimization. We measure the contributions of online optimization, retrievable task information, and permissive oracle landscapes. On PMO at 50 oracle calls, a GPT-4o mini agent's sum AUC falls from 14.44 to 4.85 after task names and curated reference compounds are hidden. An identity-blinded condition that states only the oracle form (e.g., similarity to a hidden reference) scores 5.39, while a deterministic retrieval-only baseline with no LLM reaches 12.16 (84% of the full agent). A full 23-task 2x2 ablation and a three-seed replication on eight tasks support this interpretation. On a protein-targeted benchmark with an AutoDock Vina composite oracle, a single-edit random walk is a strong lower bound; across three seeds it obtains composite 0.636 +/- 0.008 and docking improvement -1.11 +/- 0.04 kcal/mol. As a positive control, a retrieval-free population agent exceeds REINVENT at a matched 50-call AUC budget, with its strongest result at 100 calls. These results motivate two inexpensive sanity checks: retrieval-safe PMO prompts and a random-walk floor for docking-based optimization.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.08522 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-09T22:33:47.134Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).