CORTEXA
← Browse
arxivcs.LGcs.AIstat.ML2026-06-30

On the Convergence of Self-Improving Online LLM Alignment

Xudong Wu, Pangpang Liu, Vaneet Aggarwal, Jiayu Chen

The Self-Improving Alignment (SAIL) algorithm addresses distribution shift by reducing a bilevel formulation of the problem to an efficient, single-level method. Empirically, SAIL has demonstrated strong performance on this task. However, a formal analysis of its convergence properties has been lacking. We identify a key theoretical challenge: the standard SAIL objective function is not guaranteed to be strongly concave due to unfavorable properties of its Hessian. To address this limitation, we propose a regularized objective, SAIL-RevKL, which incorporates a reverse Kullback-Leibler (KL) divergence penalty to improve the optimization landscape. Our central theoretical contribution is to prove that this regularized objective satisfies the Polyak-Lojasiewicz (PL) condition within a bounded parameter space. We establish global convergence guarantees, achieving a near-linear sample complexity. We further validate the effectiveness and stability of SAIL-RevKL through empirical evaluations, demonstrating that it outperforms the vanilla SAIL on both MuJoCo benchmarks and LLM alignment tasks.

View free PDFSource page

Related papers

arxivstat.MLcs.AIcs.LGecon.EM2026-07-24

CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal Inference

Jiyuan Tan, Vasilis Syrgkanis

Automating theoretical research is constrained not only by the generation of candidate results, but also by their reliable evaluation. A common approach is to close the research loop with a large language model (LLM) reviewer. However, such reviewers remain empirically unreliable…

View free PDFSource page
arxivcs.LGcs.AIq-bio.BMstat.ML2026-07-01

Active-GRPO: Adaptive Imitation and Self-Improving Reasoning for Molecular Optimization

Xuefeng Liu, Mingxuan Cao, Qinan Huang, Thomas Brettin, Rick Stevens, Le Cong

Scientific reasoning is an increasingly important capability of large language models, yet improving the robustness and efficiency of training such reasoning remains a key open challenge. We study this problem in instruction-based molecular optimization, where answer-only supervi…

View free PDFSource page
arxivcs.CLcs.AIcs.LGstat.ML2026-07-05

Telescope: Improving Zero Shot Detection of LLM Generated Content By Measuring Token Repetition Probability

Christopher Nassif, Josh F. Cooper

Distinguishing Large Language Model (LLM) generated text from human writing is a critical and difficult challenge. While LLMs are trained to write like humans, we hypothesize that this training leaves an indelible mark. LLMs develop a particularly strong aversion to token repetit…

View free PDFSource page
arxivcs.LGcs.AIstat.ML2026-07-09

Reinforcing the Generation Order of Multimodal Masked Diffusion Models

Yidong Ouyang, Zhe Wang, Sourav Bhabesh, Dmitriy Bespalov

Diffusion Language Models (DLMs) have recently achieved substantial progress in natural language generation tasks. Recent research demonstrates that adaptive token generation ordering can significantly improve performance in mathematical reasoning and code synthesis applications.…

View free PDFSource page
arxivcs.AIcs.CLcs.LGstat.APstat.ML2026-07-02

Online Safety Monitoring for LLMs

Mona Schirmer, Metod Jazbec, Alexander Timans, Christian Naesseth, Maja Waldron, Eric Nalisnick

Despite alignment training, LLMs remain prone to generating unsafe outputs at deployment time. Monitoring outputs online and raising an alarm when safety can no longer be assumed is therefore critical. We study a simple real-time monitor that turns a verifier signal from an exter…

View free PDFSource page
arxivcs.LGcs.AIcs.CLstat.ML2026-06-26

What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefs

Nhi Nguyen, Shauli Ravfogel, Rajesh Ranganath

Large language models (LLMs) are increasingly deployed in high-stakes domains, where free-text explanations such as chain-of-thought and post-hoc rationales are used to justify model outputs. Yet it remains unclear whether these explanations are sufficient, i.e., if they contain…

View free PDFSource page