analiza aiqbench
This research introduces Precog-RM, a transformer-based model that learns to predict the success of LLM agent trajectories early in their execution by embedding actions and observations into a shared space. By enabling confident outcome prediction before task termination, this method allows for significant reductions in training compute—up to 47.3% FLOPs savings—when used as an early stopping controller for reinforcement learning algorithms like GRPO. The work demonstrates strong generalization across unseen tasks and models, offering a practical efficiency gain for scaling agentic systems without sacrificing performance.
Divergent Paths: Learning an Outcome-Predictive Representation of LLM Agent Trajectories
arXiv:2610.19614v1 Announce Type: cross Abstract: In this work, we reframe agentic trajectories as paths through a shared embedding space of actions and environmental observations. We show that successful and failed rollouts of the same task separate in this space, with simple per-task discriminators predicting held-out rollout outcomes at up to 0.95 AUROC across three agent benchmarks and three policy models. To exploit this signal across tasks, we introduce Precog-RM, a transformer-based model that meta-learns a representation in which outcome prediction can be performed in a few-shot setting. We show that Precog-RM generalizes to unseen tasks and models, and gives confident predictions of trajectory success long before termination. Deployed as an early stopping controller for Group Relative Policy Optimization (GRPO) on Tau2-Bench, Precog-RM cuts training FLOPs by up to 47.3\% while maintaining performance comparable to baseline GRPO.
Abstract:In this work, we reframe agentic trajectories as paths through a shared embedding space of actions and environmental observations. We show that successful and failed rollouts of the same task separate in this space, with simple per-task discriminators predicting held-out rollout outcomes at up to 0.95 AUROC across three agent benchmarks and three policy models. To exploit this signal across tasks, we introduce Precog-RM, a transformer-based model that meta-learns a representation in which outcome prediction can be performed in a few-shot setting. We show that Precog-RM generalizes to unseen tasks and models, and gives confident predictions of trajectory success long before termination. Deployed as an early stopping controller for Group Relative Policy Optimization (GRPO) on Tau2-Bench, Precog-RM cuts training FLOPs by up to 47.3\% while maintaining performance comparable to baseline GRPO.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2610.19614 [cs.LG]
(or arXiv:2610.19614v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.19614 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 232 cuvinte
Subiecte conexe
Why do models task game?
AI Alignment Forum · 6 aug. 2026
The Entangled Dimensions of Decision Theory
LessWrong · 30 iul. 2026
Angles of attack for continual learning safety
LessWrong · 16 iun. 2026
The Next Swan: Frank Ramsey, Variable Hypotheticals, and the Bet on Induction
LessWrong · 8 iun. 2026