The latest AI news, in focus.

How Does Post-Training Shape LLM Value Preferences?

arXiv:2610.33371v1 Announce Type: new Abstract: During post-training, large language models (LLMs) learn value preferences from corpora that may contain imbalanced proportions of opinions associated with different demographic groups. Amplifying these imbalances may further underrepresent some groups' opinions. However, how opinion mixtures shape value preferences across post-training paths remains unclear. To answer this question, we conduct controlled experiments using post-training corpora with different opinion mixtures and five post-training paths based on SFT, DPO, and GRPO. We find that post-training amplifies opinion imbalances differently across post-training paths. Once value preferences are amplified, whether further post-training can remove them depends on both the post-training

Searching with Sparse Histories: Cross-Task Architecture Search from Incomplete Search Logs and Zero-Cost Proxies

arXiv:2610.32179v1 Announce Type: new Abstract: Neural architecture search (NAS) is expensive, but cross-task transfer can reduce costs by reusing knowledge from previous tasks. However, existing methods struggle with sparse search histories and task-specific transfer failures. We propose a method, a risk-robust strategy that combines sparse source logs with zero-cost proxies through champion-based initialization and joint pairwise-ranker pretraining. On TransNAS-Bench-101, the proposal mitigates severe worst-case failures, reducing worst-10% ranks from 290 to 43 for Micro autoencoding and from 465 to 45 for Macro room layout. Across 13 cell-space targets, it outperforms a proxy-only ranker on 10 targets in both mean and worst-case performance, while maintaining competitive average perform

When to Stop Penalizing Actions? Martingale GRPO for VLA Reinforcement Learning

arXiv:2610.32431v1 Announce Type: new Abstract: A failed vision-language-action (VLA) rollout does not imply that every action contributed to failure. Yet in group-relative policy optimization (GRPO), a single trajectory-level advantage is assigned to all actions in a rollout. Potentially useful recovery attempts can thus be penalized alongside the errors they aim to correct. Ideal stopping conditions for failure feedback are characterized using success-probability and policy-score martingales. At these stops, the removed gradient contribution has zero conditional expectation given the retained information. Truncation therefore preserves the expectation of a reference GRPO group-gradient estimator and can reduce its covariance. One such stop occurs when the conditional probability of succe

Import AI 478: Math for alignment; ProgramBench; and Mirendil shows what RSI means

What is “night” to a machine?

Cognitively Constrained Partners for Zero-Shot Coordination in Real-Time Partitioned Control

arXiv:2610.31726v1 Announce Type: new Abstract: Effective human-machine teaming requires agents that coordinate with unpredictable human partners. Reinforcement learning from live human feedback or large behavioural datasets is expensive, while purely simulated training overfits to artificial conventions and fails at Zero-Shot Coordination with real people. We propose a framework for training teaming agents against Cognitively Constrained Partners: simulated partners whose diversity emerges from procedural constraints grounded in human physiological literature, specifically reaction delays and perceptual limitations, rather than from reward shaping or historical training checkpoints. In a modified Lunar Lander environment, we report preliminary evidence that these partners play more like r

New research finds: The existing critic in reinforcement learning can be directly repurposed as a confidence estimator for large language models.

An AI research paper published on October 26, 2026, on arXiv (ID 2610.31820v1) states that, for large language models (LLMs) trained using value-based reinforcement learning (value-based RL), their existing critic modules can serve as high-accuracy, highly calibrated confidence estimators without requiring additional training or data.

Why it mattersThis study reveals the mathematical equivalence between confidence estimation and Critic training in reinforcement learning (under a well-defined correctness metric), and empirically demonstrates the effectiveness and generalization capability of reusing the Critic across more than ten domains, including mathematics, code generation, tool invocation, and general reasoning. These findings hold direct engineering value for lowering the barriers to trustworthy LLM deployment.

POS-Signature Demonstration Retrieval for In-Context Dependency Parsing

arXiv:2610.31966v1 Announce Type: new Abstract: Demonstrations for in-context learning are often retrieved by semantic similarity, which may be a poor proxy for relevance in structured prediction. We study demonstration selection for Universal Dependencies parsing with a POS-signature policy that represents sentences with weighted part-of-speech statistics and retrieves three length-compatible, gold-parsed examples from the labeled training split. We evaluate two open-weight language models without fine-tuning on all 2,077 sentences of the UD English Web Treebank test set, comparing the policy with shot- and format-matched fixed and semantic three-shot baselines, as well as zero-shot, chain-of-thought, and critique-refinement systems. POS-signature retrieval yields the highest labeled atta

Financial Context-Rich Strategy for Actor-Critic based Agents: A Deep Reinforcement Learning Approach for Stock Trading

arXiv:2610.32237v1 Announce Type: new Abstract: In modern financial markets, more than 3/4th of the large volume trade is executed by algorithmic trading based on advanced trading strategies to attain higher precision and speed than traditional human traders. The conventional strategies focused on price-based indicators are inadequate in capturing the market dynamism arising from real-world events such as social sentiments, political instability, or sudden economic shifts. With the rise of Deep Reinforcement Learning (DRL) and data availability, there is a requirement for more efficient algorithms that can incorporate these external factors in trading strategies. DRL has shown potential in simulating real-world scenarios and in simplifying complex decision-making since DRL agents can autom

TaxoSAE: Rethinking Sparse Autoencoders through Taxonomic Logic

arXiv:2610.31894v1 Announce Type: new Abstract: Recent Sparse Autoencoders (SAEs) decompose the representations of large language models (LLMs) into interpretable features, enabling precise readout and control of encoded knowledge. However, existing SAEs treat each feature as a standalone unit, whereas real-world knowledge is often taxonomically organized. As a result, readout carries no guarantee of taxonomic consistency, and control lacks the granularity to specify the scope of an edit. We propose TaxoSAE, which explicitly encodes taxonomical information, built on is-a (hypernym-hyponym) relations, into the SAE feature space. TaxoSAE mirrors a taxonomy with a stack of SAEs, one per taxonomy level, and introduces a taxonomic alignment loss that trains each concept to be active whenever on

KnitTrans: Coordinated Trajectory Transfer for Effective and Efficient LLM Reinforcement Learning

arXiv:2610.32389v1 Announce Type: new Abstract: Trajectory transfer from rollout to training affects both learning quality and system efficiency in large language model reinforcement learning (RL). Enforcing batch-wide training requirements can delay the admission of completed trajectories, resulting in higher architectural staleness: the additional parameter-version gap from disaggregated pipelining relative to colocated execution. Beyond admission, misalignment between data access and training consumption causes redundant movement and unnecessary waiting when transferring large trajectory payloads. We present KnitTrans, a system that coordinates the rollout-to-training path through three designs. First, constraint-preserving mini-batch admission allows flexibility in early mini-batch com

Don't Forget What You Just Learned: Self-Directed Repair Turns Agent Mistakes Into Lasting Lessons

arXiv:2610.31846v1 Announce Type: new Abstract: Intelligent agents should learn from their mistakes so that experience leads to continuous improvement. Yet today's agents recover from a mistake only temporarily: in-context reflections are easily forgotten across sessions and compete for limited context, while full-parameter fine-tuning is expensive and can interfere with existing capabilities. We introduce Self-Direction, a mechanism for agents to learn from mistakes that combines the benefits of temporary in-context retries and policy updates. When an agent fails a task, it reflects on its own trajectory in order to extract a lesson and immediately distills it into a lightweight low-rank adapter. By keeping the original weights frozen, Self-Direction prevents adapter training from overwri

Spatial Heuristic Strategy based Diffusion Reinforcement Learning for Online 3D Bin Packing

arXiv:2610.32241v1 Announce Type: new Abstract: Traditional heuristic methods have accumulated mature experience in core objectives, such as space utilization and layout compactness of 3D bin packing, relying on artificially designed rule-based logic. However, they exhibit obvious limitations in generalization ability and adaptability to dynamic scenarios. Diffusion reinforcement learning, which integrates the spatial search capability of reinforcement learning and the data distribution modeling capability of diffusion learning, has achieved certain progress in action distribution modeling and decision compliance constraints. Nevertheless, there remain issues including insufficient perception of the spatial topological structure in 3D bin packing and limited layout optimization capability.

Distributional Alignment of LLMs and Its Generalization Properties

arXiv:2610.31687v1 Announce Type: new Abstract: Large language models are increasingly used to simulate and probe human behavior, e.g., for surveys. For these applications, a key property is that outputs satisfy distributional alignment: to match its output distribution with some reference (in this case, human) distribution. An obstacle in achieving distributional alignment is generalizing from small training data. In this work, we study (1) how to distributionally align LLMs; and (2) how aligned LLMs might generalize to unseen samples and unrelated topics and tasks. Using two large-scale benchmarks, we find that LLMs can indeed achieve distributional alignment and generalize to unseen questions of the same domain. Cross-topic generalization depends on the relationship between topics (e.g.

LazyFlash: Forward-Dense, Backward-Sparse Attention for Accelerated Long-Context Fine-Tuning

arXiv:2610.31911v1 Announce Type: new Abstract: As large language models (LLMs) process increasingly long sequences, attention becomes a major computational bottleneck. Sparse attention mostly targets inference, yet many methods still train densely or cannot support sparse training, and their added complexity makes sparsity hard to turn into wall-clock speedups, leaving dense attention widely used in practice. This paper introduces LazyFlash, a forward-dense, backward-sparse attention method built on the observation that attention training is dominated by the backward pass. LazyFlash keeps the forward pass exactly dense, so the trained model remains a standard dense model that existing systems such as vLLM can serve without any change. For the backward pass, LazyFlash proposes a gradient-a

PPO vs. SAC: A Comparative Study for Cooperative Motion Control of Dual-Arm Space Robots

arXiv:2610.32034v1 Announce Type: new Abstract: The growing population of orbital debris has intensified interest in active debris removal and on-orbit servicing, for which dual-arm space manipulators offer improved grasping stability and load distribution relative to single-arm systems. However, coordinating two manipulators under inter-arm collision avoidance, orientation alignment, and sensing uncertainty remains challenging. This paper develops a coordinated reinforcement learning framework for cooperative motion planning of a dual-arm space manipulator in a target-reaching task with inter-arm collision-avoidance constraints. A centralized state representation combines joint-space, end-effector, and target-relative information, while a multi-objective reward function promotes target-re

Sebastian Raschka has released an eight-video course titled “Building a Reasoning Model from Scratch.”

On October 24, 2026, Sebastian Raschka published an eight-part, over-11-hour hands-on video course on Python and PyTorch in Ahead of AI magazine (magazine.sebastianraschka.com), systematically covering the evaluation of inference models, scaling during inference, reinforcement learning (RLVR), and knowledge distillation, with all methods implemented using the pre-trained Qwen3-0.6B model.

Why it mattersThis course is the author’s advanced technical educational offering, following “Building Large Language Models from Scratch.” It focuses on the core evolutionary direction of today’s mainstream LLMs—enhancing inference capabilities—covering the entire technical pipeline from setting up the foundational environment to GRPO reinforcement learning and self‑distillation. The author explicitly states that all implementations are “hand‑crafted from scratch,” fully reproducible, and highly instructive with significant engineering reference value.

The study proposes a self-assessment reinforcement learning framework, SARL, which enhances the search state-awareness capability of retrieval-augmented large models.

On October 24, 2026, arXiv published the paper arXiv:2610.30435v1, introducing the Self-Assessment Reinforcement Learning (SARL) framework, which aims to address the issue in existing retrieval-augmented large language models where search agents lack the ability to perceive search states.

Why it mattersThis study addresses a critical limitation of retrieval-augmented large language models—namely, their tendency to generate hallucinatory reasoning and prematurely terminate searches when they fail to recognize previously acquired or missing information in complex tasks. It proposes an end-to-end reinforcement learning approach that requires no additional evaluation during testing, achieving significant performance improvements on multi-hop reasoning and deep-research benchmarks.

Release of a new benchmark for Nepali–English mixed-language speech recognition and a two-stage adaptation framework from synthesis to real-world conditions

On October 24, 2026, researchers published a paper on arXiv (ID 2610.30860v1), introducing the first Nepali–English code-switching ASR benchmark for spontaneous podcast scenarios, comprising 13,044 real-world utterances (15.45 hours), and proposing a two-stage adaptation approach that combines synthetic and real data: first, fine-tuning with synthetic speech generated by a 0.95-hour multi-speaker TTS model as supervision; then, incorporating group relative policy optimization (GRPO) with script-based rewards.

Why it mattersThis work directly addresses the core challenges of script obfuscation and high error rates at switching points in code-switching ASR for low-resource languages, by rigorously auditing data to prevent machine translation leaks and acoustic loops, and by making all validation texts, timestamps, episode metadata, and reconstruction scripts publicly available. The experimental results are statistically significant (p ≤ 0.001) and demonstrate generalization stability on monolingual benchmarks.

Researchers have reengineered a neural machine translation system for Sicilian, improving the quality of translations for low-resource languages.

A technical paper published on October 24, 2026, on arXiv (ID 2610.30374v1) reports the reconstruction of a small-scale neural machine translation (NMT) system for Sicilian: it leverages high-quality parallel corpora extracted from PDFs of the bilingual journal “Arba Sicula,” improves sentence alignment, tokenization, and subword segmentation that is sensitive to word stems, and migrates the original Perl‑dependent pipeline to a Perl‑free implementation. Experimental comparisons between a small Transformer trained from scratch and a fine-tuned pre-trained NLLB model demonstrate that the latter performs significantly better, particularly when translating into Sicilian.

Why it mattersThis paper presents a reproducible, engineered NMT pipeline for the endangered/low-resource language Sicilian, covering key stages such as corpus construction, model adaptation, and optimization of training strategies. All evaluations are conducted on the frozen “Arba Sicula” test set, with transparent methods and clearly defined limitations, making it a valuable technical reference.

The study proposes Delta N-grams: a forward-only, continuous-writing method for native N-gram tables in language models.

On October 24, 2026, arXiv published the paper arXiv:2610.31125v1, introducing Delta N-grams—a novel mechanism that does not modify the pre-trained N-gram table T⁰, but instead writes sparse incremental patches Δ via forward propagation, enabling efficient and reversible continual memory updates.

Why it mattersThis work addresses the issue of large perturbations to a model’s base parameters in continual learning by proposing a memory-enhanced pathway that is lightweight in structure (4.1M parameters), computationally efficient (purely forward pass), and fully invertible (resetting Δ restores the original model). Its superior document-specificity and robustness against interference have been validated on models such as Qwen3.8-Flash-Next, making it a cutting-edge contribution to trustworthy AI systems research.