← Back to all news
arXiv AI By Yanze Jiang, Mingxuan Li, Yuhao Wang, Shengfang Zhai, Jiaheng Zhang

Don't Solve, Just Compare: Tiny Advisors for Runtime Intervention in LLM Agents

Read the original on arXiv AI →

The paper introduces COTA, a lightweight comparison-only framework that enables runtime intervention in large language model agents by using a tiny comparator to evaluate whether alternative continuations outperform the agent’s current proposal. Instead of generating full solutions, the comparator only judges sampled alternatives, and repeated comparisons decide when to intervene. Experiments on WebShop, ALFWorld, and tau^3-Retail show that COTA improves performance across nine settings and outperforms existing baselines, demonstrating that constructive intervention can be effective even with a much weaker auxiliary model.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

Related stories

arXiv Computation and Language
Sep 1

CAST: Critique-Aware Supervision for Training Reliable Long-Horizon Tool-Calling Agents

CAST is a critique‑aware training framework that transforms sparse task outcomes into action‑level supervision for both critique learning and policy optimization. By analyzing agent trajectories, CAST synthesizes structured rationales that explain action validity under partial observability, enabling the creation of richer training data. Fine‑tuned Qwen3‑family models trained with CAST show significant reliability gains, outperforming GPT‑OSS‑120B by over 10% on Retail tasks and improving Telehealth performance by 9% in an out‑of‑domain setting.

By Amir Saeidi, Zehua Zhang, Rishitosh Singh, Naman Ahuja, Vivek Gupta, Ali Payani, Gaowen Liu, Jayanth Srinivasa, Chitta Baral
llmsagentsfine-tuningbenchmarks
More like this →
arXiv AI
Oct 12

MENTOR: Learning from Failure with On-Policy Pedagogical Intervention

The paper introduces MENTOR, a multi‑agent framework that provides on‑policy pedagogical intervention to address zero‑reward collapse in reinforcement learning from verifiable rewards. MENTOR diagnoses failures in real time and generates targeted prompts to guide the learning agent, integrating seamlessly with existing policy optimization pipelines such as GRPO. Experiments on math and code generation tasks show that MENTOR significantly reduces collapse, achieving up to 78.3% relative gains on Llama‑3.2‑3B‑Instruct.

By Guanlan Yue, Hang Yu, Xiaoyong Huang, Ruipeng Li, Wei Zhang, Bingchang Liu, Rui Wang
llmsagentsreinforcement-learningbenchmarks
More like this →