The paper presents a Bayesian framework that unifies several large‑language‑model training and evaluation paradigms—supervised fine‑tuning (SFT), few‑shot in‑context learning (ICL), and KL‑regularized reinforcement learning (RLHF/RLVR). It shows that each method can be viewed as a two‑step process: first constructing a Bayes or Gibbs posterior over outputs or actions using a prior and a utility signal, then approximating this posterior via a forward‑KL projection onto a parametric family. The authors formalize ICL and SFT as amortized weight projections, and demonstrate that reward‑weighted SFT, reward‑weighted ICL, and advantage‑weighted SFT are all special cases of forward‑KL projection onto reward‑induced posteriors, while also outlining where these equivalences hold and where they break down.
arXiv:2606. 01682v1 Announce Type: cross Abstract: Selecting the best response from multiple small-model samples using a stronger scorer is a simple inference-time strategy, but fails when the small model has already committed to incorrect reasoning paths.
Selecting the best response from multiple small-model samples using a stronger scorer is a simple inference-time strategy, but fails when the small model has already committed to incorrect reasoning paths. PRM guided search avoids this by scoring candidate continuations during generation, but requires a reward model trained with step-level labels.
arXiv:2607. 16097v1 Announce Type: cross Abstract: Reinforcement learning (RL) has become central to improving large language models (LLMs) on complex reasoning tasks, yet RL post-training is largely studied in isolation from the pretraining that precedes it.
arXiv:2605.30327v2 Announce Type: replace-cross Abstract: Frontier reasoning models are produced by post-training base language models with reinforcement learning. Recent work has challenged this by...
arXiv:2607. 09693v1 Announce Type: cross Abstract: Reinforcement learning (RL) has become the dominant paradigm for improving the reasoning capabilities of large language models, but it requires expensive training, curated data, and reward signals.
Whether large language models perform genuine algorithmic reasoning or mere pattern completion is hard to test, because most benchmarks lack a ground truth for correct inductive inference. We introduce F-ICL, an in-context-learning benchmark that supplies one exactly.
arXiv:2606. 17024v1 Announce Type: new Abstract: Sparse reward reinforcement learning (RL) has become a standard tool for improving LLM reasoning, but its success depends critically on the coverage present in the base model.
arXiv:2605.31378v2 Announce Type: replace Abstract: Large Reasoning Models (LRMs) still struggle with fine-grained translation quality estimation (QE), even with long reasoning chains. We argue that...
The paper introduces Stepwise Marginal Information Gain (MIG), an intrinsic process reward that evaluates how each reasoning step of a large language model (LLM) or vision-language model (VLM) improves the likelihood of the reference answer. MIG rewards only new likelihood maxima, preventing duplicate credit, and is combined with outcome, format, and self‑distillation objectives to guide training. Experiments on eight benchmarks show that this method outperforms outcome‑only reinforcement learning and improves accuracy by up to 4.8 points over binary‑reward training, including a 12.6‑point gain on MathVerse and a 12.9‑point advantage on vision‑language transfer at 7B parameters.
arXiv:2606. 11709v1 Announce Type: new Abstract: On-policy self-distillation (OPSD) provides dense, token-level supervision for reasoning models by aligning a model's own distribution with the distribution it produces under privileged context, typically a verified solution.
arXiv:2606. 09092v1 Announce Type: new Abstract: Theory of Mind (ToM) is a must-acquire skill for modern foundation model systems to operate effectively and safely in the real world.
The paper investigates whether the high costs of training chain-of-thought reasoning models can be reduced through algorithmic design. It introduces an autocurriculum approach that lets the model select which problems to focus on during training, showing that this method provably improves both supervised fine‑tuning and reinforcement learning. For supervised fine‑tuning, autocurriculum requires exponentially fewer reasoning demonstrations by targeting prompts where the model struggles, while for reinforcement learning it decouples computational cost from the quality of the reference model, making the burn‑in cost nearly independent of target accuracy.