This work introduces DEO, which replaces challenger parameter updates with solver-guided task sampling, and shows that DEO learns distributionally robust reasoning ability.
Abstract
Self-evolving language models improve by generating tasks and learning from their own feedback, but adapting the task generator often requires a separate challenger-training loop. Can we generate tasks adapted to the current solver without explicitly training a challenger? We introduce \textbf{D}irect Self-\textbf{E}volving \textbf{O}ptimization (DEO), which replaces challenger parameter updates with solver-guided task sampling. The KL-regularized challenger objective defines an exponential tilt of a fixed base task distribution. DEO uses this distribution as a sampling target: a frozen LLM generates and mutates tasks, the solver scores them, and an approximate Metropolis selection rule refines the training pool. Only the solver is trained. Theoretically, for an idealized variant that samples exactly from the tilted distribution, and under regularity, local gradient-dominance, and initialization conditions, we show that DEO learns distributionally robust reasoning ability. In experiments, DEO achieves reasoning performance competitive with R-Zero while using over $50\%$ less wall-clock training time, and improves reasoning accuracy over a no-walk ablation. Replacing the task generator with a frozen API-only LLM further improves the local solver, illustrating a capability enabled by removing challenger training.
Reinforcement learning with verifiable rewards (RLVR) is reshaping how large language models (LLMs) acquire reasoning capabilities, yet the optimizers used remain largely inherited from teacher-forcing training. This raises a fundamental question: does RLVR induce a distinct optimization geometry that calls for a diffe...
Zhi-Peng Chen, Wayne Xin Zhao, Ji-Rong Wen· 0 citations
Self-evolving language models train on tasks they generate, so every round must decide which updated model the next round inherits. Current loops such as R-Zero, Agent0, and Skill Self-Play make this decision through the task, admitting tasks that look learnable and inheriting every resulting update. We identify a sele...
Hao-Yi Yang, Kshitij Mishra, Nils Lukas et al.· 0 citations
Activation-Conditioned Self-Distillation (ACSD), which extracts a steering vector by contrasting activations of self-generated trajectories that reach verified correct answers within a generation budget with those of all remaining trajectories, achieves the highest mean accuracy over four mathematical benchmarks among...
A holistic framework based on learnable information gain, which measures how much novel, parameterizable information a round provides relative to the previous round, and proposes ATRI (Adaptive Training Regulation via Information-gain), which reweights samples within a round and halts training across rounds when inform...
Chen-Xu Wang, Chao-Zhuo Li, Xin-Ze Shi et al.· arXiv.org· 0 citations
Self-Play Pretraining with Zero Data is introduced, an initial proof-of-concept towards realizing a more general approach to pretraining, and whether zero-shot performance on natural data improves predictably with self-play compute is tested.
Aditya Cowsik, Kfir Dolev, Michael Y. Li et al.· arXiv.org· 3 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduOct 13, 2026
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.