Bootstrapping Task Spaces for Self-Improvement
About
Progress in many task domains emerges from repeated revisions to previous solution attempts. Training agents that can reliably self-improve over such sequences at inference-time is a natural target for reinforcement learning (RL), yet the naive approach assumes a fixed maximum iteration depth, which can be both costly and arbitrary. We present Exploratory Iteration (ExIt), a family of autocurriculum RL methods that directly exploits the recurrent structure of self-improvement tasks to train LLMs to perform multi-step self-improvement at inference-time while only training on the most informative single-step iterations. ExIt grows a task space by selectively sampling the most informative intermediate, partial histories encountered during an episode for continued iteration, treating these starting points as new self-iteration task instances to train a self-improvement policy. ExIt can further pair with explicit exploration mechanisms to sustain greater task diversity. Across several domains, encompassing competition math, multi-turn tool-use, and machine learning engineering, we demonstrate that ExIt strategies, starting from either a single or many task instances, can produce policies exhibiting strong inference-time self-improvement on held-out task instances, and the ability to iterate towards higher performance over a step budget extending beyond the average iteration depth encountered during training.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Machine learning engineering | MLE-bench (held-out task instances) | Accuracy (%)58.6 | 6 | |
| Mathematical Reasoning | Math held-out task instances (test) | Accuracy20.4 | 6 | |
| Tool Use | Tool-use multi-turn (test) | Accuracy76.8 | 6 |