arXiv:2606. 18089v1 Announce Type: new Abstract: Post-training pipelines that combine supervised fine-tuning (SFT) with reinforcement learning (RL) have emerged as the key recipe for transforming large language models (LLMs) into robust reasoners.
arXiv:2505.22635v2 Announce Type: replace-cross Abstract: A common approach for teaching large language models (LLMs) to reason is to train on chain-of-thought (CoT) traces of in-distribution reasoni...
The paper investigates how different forms of compressed chain‑of‑thought (CoT) reasoning—Explicit, Composed, and Implicit—affect large language model (LLM) performance after supervised fine‑tuning (SFT). Using a synthetic compositional reasoning task, the authors show that coarser CoT requires more SFT data, that Composed and Implicit CoT benefit more from data scaling (with Composed also benefiting from repetition), and that reinforcement learning with verifiable rewards (RLVR) can decompose compressed steps learned during SFT. Additionally, unidirectional CoT ordering improves generalization on longer sequential tasks.
arXiv:2607. 07646v1 Announce Type: new Abstract: Does RL post-training merely amplify primitive skills already latent in a base model, or can it compose primitive skills into new higher-level strategies?
arXiv:2606. 05402v1 Announce Type: cross Abstract: Large reasoning models (LRMs) produce reasoning traces with non-linear structures, such as backtracking and self-correction, that complicate the evaluation and monitoring of the reasoning process.
The paper introduces a dependency‑graph framework to formalize compositional reasoning in language models, defining three increasing levels of compositionality. Using data‑structure tasks with deterministic rewards, the authors observe a consistent asymmetry: training on decomposed skills does not reliably transfer to composed tasks, whereas training on composed tasks transfers more readily to decomposed ones. They provide a theoretical explanation for this asymmetry and evaluate its effects under length extrapolation, structural distribution shift, and transfer to unseen skills, concluding with a pilot study on real‑world tool‑calling benchmarks that suggests the phenomenon extends to practical settings.
arXiv:2604. 06628v2 Announce Type: replace Abstract: A prevailing narrative in LLM post-training holds that supervised finetuning (SFT) memorizes while reinforcement learning (RL) generalizes.
arXiv:2606. 02994v1 Announce Type: new Abstract: ReAct-style LLM agents often rediscover the same reasoning routines across problems, yet leave those routines trapped in transient scratchpads.
arXiv:2606. 10646v1 Announce Type: new Abstract: Token-level credit assignment remains a key obstacle for reinforcement learning (RL) in large language models (LLMs), where RL recipes typically treat all tokens equally, failing to distinguish decisive reasoning steps from routine formatting or fluent filler.
arXiv:2604. 08477v2 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has substantially improved reasoning in formal domains such as mathematics and code, but extending these gains beyond STEM remains challenging.
arXiv:2607. 12771v1 Announce Type: new Abstract: Reaction mechanisms consist of the step-by-step sequences of elementary reactions that explain chemical transformations.
arXiv:2505.16782v3 Announce Type: replace Abstract: Large Language Models (LLMs) have shown impressive performance on complex tasks through Chain-of-Thought (CoT) reasoning. However, conventional CoT...
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.