Policy optimisation, reward modelling and RLHF — how models are trained by feedback rather than by labels.
arXiv:2608.29831v1 Announce Type: new Abstract: Localizing issue-relevant code regions is a critical step in automated software engineering. However, due to their reliance on sparse trajectory-level...
arXiv:2608.30765v1 Announce Type: new Abstract: Multi-task reinforcement learning (MTRL) is a technique to train multiple tasks simultaneously, where previous works usually train a single model to so...
arXiv:2608.30597v1 Announce Type: new Abstract: Direct Preference Optimization (DPO) simplifies alignment through pairwise comparisons but assumes all observed preferences are reliable. Real data oft...
arXiv:2608.29356v1 Announce Type: new Abstract: Artificial Intelligence (AI) has long been inspired by studies of biological intelligence. Reinforcement learning, for instance, drew inspiration from...
arXiv:2608.29461v1 Announce Type: new Abstract: Sandbagging models strategically underperform on evaluations while retaining the capabilities being measured. The evaluations that guide frontier-model...
arXiv:2608.29458v1 Announce Type: cross Abstract: Sandbagging, in which a model deliberately underperforms on an evaluation despite retaining the underlying capability, threatens the safety evaluatio...
arXiv:2608.29953v1 Announce Type: new Abstract: Flat retrieval-augmented generation treats a corpus as a bag of chunks, discarding document hierarchy and cross document structure. We introduce Search...
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:2608.30841v1 Announce Type: new Abstract: Large language models can often generate plausible mathematical reasoning traces, but reliably identifying the correct solution among multiple candidat...
arXiv:2601.03468v2 Announce Type: replace Abstract: Reinforcement learning (RL) has become a standard approach for post-training large language models and, more recently, for improving image generati...
arXiv:2608.29850v1 Announce Type: new Abstract: E-commerce platforms increasingly personalize user experiences through machine learning, yet page layout decisions remain dominated by static rules and...
arXiv:2608.30399v1 Announce Type: cross Abstract: Large language models (LLMs) exhibit strong semantic reasoning and open-ended generation abilities, but aligning these abilities with structured sequ...
SkillForge is a framework that breaks down formal code synthesis into reusable atomic skills, each handling a specific subtask such as specification inference, body synthesis, invariant generation, error diagnosis, or repair. A verification-driven harness coordinates these skills by submitting candidates to the Dafny verifier, diagnosing failures, and routing them deterministically to the appropriate repair skill until correctness is achieved or a budget is reached. On a curated benchmark, SkillForge outperforms state‑of‑the‑art agentic and iterative baselines, requiring fewer tokens and lower latency, with ablation studies showing each skill’s measurable contribution and rapid convergence.
arXiv:2608.28600v1 Announce Type: new Abstract: Large language models (LLMs) achieve strong performance on mathematical reasoning benchmarks, yet the mathematically meaningful skills underlying their...
The next generation of mobile networks is envisioned as fully AI-native, with AI-RAN architectures embedding small language models (SLMs) to perform reasoning over real-time telemetry. The state-of-th...
Recent advances in large reasoning models (LRMs) have shown that reinforcement learning with verifiable rewards (RLVR) can substantially improve reasoning in mathematics and code, where outcomes can b...
Chemistry questions often demand exact computation and database lookups that a language model cannot supply from its parameters, so it must reach for external tools. Tool use here is a three-part prob...
Existing approaches for synthesizing Partial Differential Equation (PDE) solvers predominantly rely on autoregressive models, yet their global left-to-right decoding incurs substantial redundancy when...
Recent offline reinforcement learning (RL) studies report policies that outperform physician decisions on clinical outcomes. We conduct a systematic, partially crossed evaluation of five offline RL al...
The paper introduces VICT, a method that leverages the internal structure of verifiable tasks to perform fine‑grained credit assignment for long‑horizon LLM agents. VICT exposes executable or evidence‑backed atoms from a task’s terminal verifier and traces them back to actions via dependency‑valid proof edges, redistributing advantage only along these edges. This approach improves performance on ALFWorld and WebShop compared to outcome‑only training and matches recent fine‑grained credit methods without requiring additional critics, labels, or inference‑time verifier access.