# Self-Play with Adversarial Critic: Provable and Scalable Offline Alignment for Language Models

arXiv: https://arxiv.org/abs/2406.04274 · HTML: https://mlsanity.com/paper/2406.04274/

## Key facts

- Cited by 16 papers in the map (data through 14 Oct 2026).
- Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 57 of its 85 references that are in the map.
- First posted on arXiv 6 Jun 2024.

## Placed in

- LLM alignment / RLHF / preference optimisation / [LLM preference optimization methods](https://mlsanity.com/alignment/llm-preference-optimization-methods/)
- Synthetic data and self-improvement for LLMs / [Self-rewarding alignment and self-training](https://mlsanity.com/synth/self-rewarding-alignment-and-self-training/)
- Offline RL and imitation learning from data / [Language-conditioned policies and VLAs](https://mlsanity.com/offlinerl/language-conditioned-policies-and-vlas/)

## Builds on (most cited first, 30 of 57)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems](https://mlsanity.com/paper/2005.01643/) (2020)
- [Conservative Q-Learning for Offline Reinforcement Learning](https://mlsanity.com/paper/2006.04779/) (2020)
- [Generative Adversarial Imitation Learning](https://mlsanity.com/paper/1606.03476/) (2016)
- [Off-Policy Deep Reinforcement Learning without Exploration](https://mlsanity.com/paper/1812.02900/) (2018)
- [Enhancing Chat Language Models by Scaling High-quality Instructional Conversations](https://mlsanity.com/paper/2305.14233/) (2023)
- [Introduction to Online Convex Optimization](https://mlsanity.com/paper/1909.05207/) (2019)
- [Cross-Task Generalization via Natural Language Crowdsourcing Instructions](https://mlsanity.com/paper/2104.08773/) (2021)
- [Textbooks Are All You Need II: phi-1.5 technical report](https://mlsanity.com/paper/2309.05463/) (2023)
- [Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models](https://mlsanity.com/paper/2401.01335/) (2024)
- [Contrastive Preference Optimization: Pushing the Boundaries of LLM Performance in Machine Translation](https://mlsanity.com/paper/2401.08417/) (2024)
- [The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models](https://mlsanity.com/paper/2201.03544/) (2022)
- [Emergent Tool Use From Multi-Agent Autocurricula](https://mlsanity.com/paper/1909.07528/) (2019)
- [A Theory of Regularized Markov Decision Processes](https://mlsanity.com/paper/1901.11275/) (2019)
- [Bridging Offline Reinforcement Learning and Imitation Learning: A Tale of Pessimism](https://mlsanity.com/paper/2103.12021/) (2021)
- [Bellman-consistent Pessimism for Offline Reinforcement Learning](https://mlsanity.com/paper/2106.06926/) (2021)
- [Principled Reinforcement Learning with Human Feedback from Pairwise or K-wise Comparisons](https://mlsanity.com/paper/2301.11270/) (2023)
- [FLAMBE: Structural Complexity and Representation Learning of Low Rank MDPs](https://mlsanity.com/paper/2006.10814/) (2020)
- [A Theoretical Analysis of Deep Q-Learning](https://mlsanity.com/paper/1901.00137/) (2019)

## Built on by (most cited first, 16 of 16)

- [Is Best-of-N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment](https://mlsanity.com/paper/2503.21878/) (2025)
- [Correcting the Mythos of KL-Regularization: Direct Alignment without Overoptimization via Chi-Squared Preference Optimization](https://mlsanity.com/paper/2407.13399/) (2024)
- [Pluralistic Off-policy Evaluation and Alignment](https://mlsanity.com/paper/2509.19333/) (2025)
- [Avoiding mathbfexp(Rmax) scaling in RLHF through Preference-based Exploration](https://mlsanity.com/paper/2502.00666/) (2025)
- [Learning a Pessimistic Reward Model in RLHF](https://mlsanity.com/paper/2505.20556/) (2025)
- [KnowRL: Teaching Language Models to Know What They Know](https://mlsanity.com/paper/2510.11407/) (2025)
- [From Demonstrations to Rewards: Alignment Without Explicit Human Preferences](https://mlsanity.com/paper/2503.13538/) (2025)
- [Reward Shaping for (Inference-Time) Alignment: A Stackelberg Game Perspective](https://mlsanity.com/paper/2602.02572/) (2026)
- [Discriminative Finetuning of Generative Large Language Models without Reward Models and Human Preference Data](https://mlsanity.com/paper/2502.18679/) (2025)
- [Provably Mitigating Corruption, Overoptimization, and Verbosity Simultaneously in Offline and Online RLHF/DPO Alignment](https://mlsanity.com/paper/2510.05526/) (2025)
- [Breaking the Capability Ceiling of LLM Post-Training by Reintroducing Markov States](https://mlsanity.com/paper/2603.19987/) (2026)
- [Provably avoiding over-optimization in Direct Preference Optimization without knowing the data distribution](https://mlsanity.com/paper/2602.06239/) (2026)
- [IRIS: Interpolative R'enyi Iterative Self-play for Large Language Model Fine-Tuning](https://mlsanity.com/paper/2604.20933/) (2026)
- [S-SPPO: Semantic-Calibrated Self-Play Preference Optimization](https://mlsanity.com/paper/2606.01561/) (2026)
- [When Does Trajectory-Level Supervision Permit Efficient Offline Reinforcement Learning?](https://mlsanity.com/paper/2606.18531/) (2026)
- [AI Alignment through a Game-theoretic Lens: A Survey](https://mlsanity.com/paper/2608.27910/) (2026)

## Abstract (arXiv)

This work studies the challenge of aligning large language models (LLMs) with offline preference data. We focus on alignment by Reinforcement Learning from Human Feedback (RLHF) in particular. While popular preference optimization methods exhibit good empirical performance in practice, they are not theoretically guaranteed to converge to the optimal policy and can provably fail when the data coverage is sparse by classical offline reinforcement learning (RL) results. On the other hand, a recent line of work has focused on theoretically motivated preference optimization methods with provable guarantees, but these are not computationally efficient for large-scale applications like LLM alignment. To bridge this gap, we propose SPAC, a new offline preference optimization method with self-play, inspired by the on-average pessimism technique from the offline RL literature, to be the first provable and scalable approach to LLM alignment. We both provide theoretical analysis for its convergence under single-policy concentrability for the general function approximation setting and demonstrate its competitive empirical performance for LLM alignment on a 7B Mistral model with Open LLM Leaderboard evaluations.