This work reframe agentic trajectories as paths through a shared embedding space of actions and environmental observations, and introduces Precog-RM, a transformer-based model that meta-learns a representation in which outcome prediction can be performed in a few-shot setting.
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.
An end-to-end success rate alone cannot distinguish failure to learn an action from failure to apply it as task state changes. We study this distinction in deterministic tool workflows that provide the current artifact state and candidate action recipes. Using 1.7B, 8B, and 27B Qwen models, we compare supervised learni...
Dong Nie, Rui-Yi Lan, Yi-Xin Nie et al.· 0 citations
Reinforcement learning enables large language model (LLM) agents to learn multi-step behaviors through interaction with their environments. However, rewards in many interactive tasks reflect only the final outcome, providing limited guidance on which intermediate decisions advance the task. Successful training trajecto...
Bo-Wen Zhang, Jun-Wei He, Mao-Qi Liu et al.· 1 citation
Language models are moving beyond generating answers to pursuing long-horizon goals in interactive environments. Post-training these agents requires long, heterogeneous trajectories, and synchronous systems leave learner engines idle until rollout and verification finish. To squeeze out these pipeline bubbles, asynchro...
Seil Kang, Hangoo Kang, Tarun Suresh et al.· 0 citations
This work introduces ActObs, which also supervises the observation tokens already present in each trajectory, and traces this difference to SFT: action and observation gradients rapidly become orthogonal, while action-only training leaves a large residual observation gradient and degrades environment prediction below t...
Joint-embedding predictive architectures (JEPAs) learn compact latent world models without pixel reconstruction. Existing JEPA pipelines typically train visual representations and action-conditioned dynamics before adding a control policy or solver. We introduce LeWAM, an end-to-end JEPA for control with: (1) a joint t...
Minghao Fu, Tavis Siebert, Eryk Halicki et al.· 0 citations
LePlanner is an amortized iterative controller that learns to construct and refine latent action sequences through a frozen world-model predictor, enabling fast, horizon-aware, nonlinear physical control without online optimization.
Saksham Bansal, Om Naphade, Chayan Aggarwal et al.· 1 citation· ⚡1
Machine-learning networks can learn without detailed instructions from a designer. Those that do may offer insight into how the brain reorganizes itself during learning.
MIT News · Artificial Intelligence· news.mit.eduSep 29, 2026
Professor Sherry Turkle’s new book, “Artificial Intimacy,” offers a withering critique of chatbots and the antisocial dynamics she believes they encourage.
