# On The Statistical Limits of Self-Improving Agents

arXiv: https://arxiv.org/abs/2510.04399 · HTML: https://mlsanity.com/paper/2510.04399/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Self-evolving LLM agents, one of 13 directions in Synthetic data and self-improvement for LLMs.
- Builds on 26 of its 51 references that are in the map.
- First posted on arXiv 5 Oct 2025.

## Placed in

- Synthetic data and self-improvement for LLMs / [Self-evolving LLM agents](https://mlsanity.com/synth/self-evolving-llm-agents/)

## Builds on (most cited first, 26 of 26)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks](https://mlsanity.com/paper/1703.03400/) (2017)
- [Overcoming catastrophic forgetting in neural networks](https://mlsanity.com/paper/1612.00796/) (2016)
- [Parameter-Efficient Transfer Learning for NLP](https://mlsanity.com/paper/1902.00751/) (2019)
- [Toolformer: Language Models Can Teach Themselves to Use Tools](https://mlsanity.com/paper/2302.04761/) (2023)
- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [DARTS: Differentiable Architecture Search](https://mlsanity.com/paper/1806.09055/) (2018)
- [Neural Architecture Search with Reinforcement Learning](https://mlsanity.com/paper/1611.01578/) (2016)
- [Continual Lifelong Learning with Neural Networks: A Review](https://mlsanity.com/paper/1802.07569/) (2018)
- [Meta-Learning in Neural Networks: A Survey](https://mlsanity.com/paper/2004.05439/) (2020)
- [Regularized Evolution for Image Classifier Architecture Search](https://mlsanity.com/paper/1802.01548/) (2018)
- [Hyperband: A Novel Bandit-Based Approach to Hyperparameter Optimization](https://mlsanity.com/paper/1603.06560/) (2016)
- [Neural Turing Machines](https://mlsanity.com/paper/1410.5401/) (2014)
- [Train faster, generalize better: Stability of stochastic gradient descent](https://mlsanity.com/paper/1509.01240/) (2015)
- [Meta-Learning with Implicit Gradients](https://mlsanity.com/paper/1909.04630/) (2019)
- [Introduction to Multi-Armed Bandits](https://mlsanity.com/paper/1904.07272/) (2019)
- [AutoML: A Survey of the State-of-the-Art](https://mlsanity.com/paper/1908.00709/) (2019)
- [Population Based Training of Neural Networks](https://mlsanity.com/paper/1711.09846/) (2017)
- [Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents](https://mlsanity.com/paper/2505.22954/) (2025)
- [Generalization Bounds For Meta-Learning: An Information-Theoretic Analysis](https://mlsanity.com/paper/2109.14595/) (2021)
- [Information-Theoretic Generalization Bounds for Meta-Learning and Applications](https://mlsanity.com/paper/2005.04372/) (2020)
- [A Theory of PAC Learnability under Transformation Invariances](https://mlsanity.com/paper/2202.07552/) (2022)
- [Revisiting Agnostic PAC Learning](https://mlsanity.com/paper/2407.19777/) (2024)
- [PAC Learning with Improvements](https://mlsanity.com/paper/2503.03184/) (2025)
- [Computable learning of natural hypothesis classes](https://mlsanity.com/paper/2407.16663/) (2024)
- [Metagoals Endowing Self-Modifying AGI Systems with Goal Stability or Moderated Goal Evolution: Toward a Formally Sound and Practical Approach](https://mlsanity.com/paper/2412.16559/) (2024)

## Built on by (most cited first, 5 of 5)

- [Hierarchical Self-Improvement: A Framework for Task-Specific Evolvable Agent Harnesses](https://mlsanity.com/paper/2608.08466/) (2026)
- [A Theory of Reliable Self-Evolution for Agent Harnesses](https://mlsanity.com/paper/2609.08175/) (2026)
- [Why Code, Why Now: An Information-Theoretic Perspective on the Limits of Machine Learning](https://mlsanity.com/paper/2602.13934/) (2026)
- [Which Self-Improvements Should We Trust? Reliable Self-Improvement When Agents Reuse Their Benchmarks](https://mlsanity.com/paper/2609.33180/) (2026)
- [Learning without Overwriting: A Theory of Self-Distillation and Supervised Fine-Tuning in Continual Reasoning](https://mlsanity.com/paper/2610.05200/) (2026)

## Abstract (arXiv)

We develop a learning-theoretic framework for analyzing self-improving agents by decomposing self-modification into five axes. Within this framework, we prove a sharp boundary: under standard i.i.d. assumptions, distribution-free PAC learnability is preserved if and only if the policy-reachable family remains uniformly capacity-bounded. If reachable capacity can grow without bound, utility-rational self-changes can make learnable tasks unlearnable. We further introduce a simple Two-Gate guardrail -- a validation-improvement requirement plus a capacity cap -- that preserves this boundary and yields standard VC-rate guarantees. The broader implication is that self-modification must be constrained not only by objectives, but also by structural conditions that preserve the statistical prerequisites for learning. As AI systems become increasingly intelligent and autonomous, this framework provides a foundation for the statistical theory of self-improvement.