Sat 31 Oct
LiveGet it free
ZOTPAPER

News without the noise

Research

New AI papers from arXiv, picked daily and explained with their claims, benchmark results and code.

2566 articles · page 6 of 54 · The Research front page

AllAgentsLanguage modelsReinforcement learningVisionGenerative mediaSpeech & audioRobotics & embodied AISafety & securityEvaluation & benchmarksEfficiency & systemsCode & softwareAI for science & healthTheory
Viewgridlistcompact
Language models

Constrained decoding does not inherently degrade LLM reasoning, contrary to common assumption

The authors revisit the assumption that constrained decoding inherently harms LLM reasoning. A theoretical proof and causal ablations across eleven open-source models show that apparent performance drops stem from prompt-level differences, not decoder-level masking. Validity constraints improve accuracy by up to 17.4%, while reasoning constraints help only when they enforce beneficial trajectories.

Evaluation & benchmarks

AI research agents with complex strategies can lose to simple searches

The authors introduce Auto-Research Arena, a multi-target benchmark for evaluating autonomous research systems in language modeling, spanning nine objectives that include model size, data, training, and inference efficiency. Across six LLM-based research harnesses and a Tree-structured Parzen Estimator baseline, a simple sequential search ranks second overall, indicating substantial room for improvement in current auto-research systems. The benchmark also surfaces a broad frontier of discoveries, some of which, when reused, yield better models under matched compute.

Efficiency & systems

Scalar two-bit weight quantization can preserve language model reasoning accuracy

The authors present ThinkBit, a five-stage framework for quantizing reasoning language models to two bits per weight in a strictly scalar format, where each weight is an integer codeword times a bfloat16 scale shared by 128 weights. On Qwen3-1.7B, their 0.52 GB checkpoint scores up to 73.0 on MATH-500 and is smaller than every published 2-bit GGUF of the model; at 14B, a 3.74 GB model scores 94.8 versus 97.5 for the 29.5 GB bfloat16 baseline.

Generative media

AI can render moving scenes from new viewpoints in real time

Hong and colleagues present FL4N, a model that turns monocular video into a representation for rendering a moving scene from different viewpoints and at different times. Rather than learning scene structure and motion from scratch, it reuses a pretrained static-scene model and learns temporal changes as corrections to its internal representation. The authors report leading or competitive rendering quality, depending on the camera information supplied, alongside real-time rendering.

Reinforcement learning

Language models find more valid solutions without supplied correct answers

The authors identify a 'Frequency Conflict' in test-time reinforcement learning (TTRL) for tasks that admit multiple valid solutions: generation frequency simultaneously serves correctness estimation (favoring frequent outcomes) and diversity preservation (favoring rare outcomes). They propose MS-TTRL, which decouples these roles by using outcome-level self-verification as an independent correctness signal and reserving frequency for diversity reweighting. Across two benchmarks, four multiple-solution tasks, and three backbone models, MS-TTRL improves the Recovery Rate by 19.13 percentage points on average over the best label-free baseline and approaches the performance of the supervised IPS-GRPO.

Efficiency & systems

Language models can learn nearly as well with four-bit numbers

The authors trained the 20-billion-parameter mixture-of-experts model GPT-OSS-20B under the MXFP4 4-bit microscaling format on the MLPerf Small MoE benchmark. By composing existing techniques with a simplified weight de-oscillation scheme adapted from TetraJet-v2, they achieved a final validation loss of 3.3350 on C4, only 0.007 above the BF16 baseline (3.3283). De-oscillation halved the remaining loss gap relative to MXFP4 without it (which gave 3.3418) and reduced the steps-to-target overhead from +20% to +5%.

Agents

AI video assistants answer more accurately with better timing as events unfold

The authors introduce SyncPro, a synchronous dual-stream framework for proactive streaming video understanding. It consists of an Inner branch for evidence tracking and a Response branch for user-facing outputs, which share causal history but activate independently, allowing concurrent decoding. Experiments on three proactive benchmarks show improved response timing and more accurate answers compared to prior methods.

Evaluation & benchmarks

Systematic audit reveals pervasive data quality issues in RVL-CDIP document classification benchmark

Larson and Leach systematically audited the 400,000-document RVL-CDIP benchmark and found multiple defects: label errors, train-test leakage, spurious cues, OCR sensitivity, and sensitive personal data. After remediating all defects, retrained models were 0.9 to 2.9 points more accurate; the most accurate model, LayoutLMv3, reached only 96.5%, far below the 98.8% claimed on the original benchmark.

Safety & security

Image generators forget unwanted content more reliably while preserving image quality

Xiahou and colleagues introduce Concept Acupuncturist, a method for suppressing selected concepts in text-to-image diffusion models by fine-tuning one internal block. On SDXL-Turbo’s standard one-step generation setting, the authors report an average forgetting-success improvement of 36.34 percentage points over the best-performing baseline, while preserving image quality and adherence to prompts.

Language models

Open-weights German-English model claims Pareto frontier for quality and serving cost

Aleph Alpha presents Kolibri, a 78.1B-parameter mixture-of-experts transformer with 3.46B active parameters per token, trained on 24T tokens including a German-heavy data mix and released under Apache 2.0. The authors report that among the open models they evaluated, Kolibri sits on the Pareto frontier of quality versus serving cost in both English and German, while covering regulation-sensitive domains such as public administration, industry and aerospace.

Evaluation & benchmarks

Short-horizon PDE model accuracy does not ensure long-rollout reliability

The authors evaluate seven PDE models on seven time-dependent datasets, comparing short-horizon and full-horizon autoregressive rollouts. They find that short-horizon advantages of foundation models often preserve rankings but absolute errors grow substantially, with median error amplification of 2.04 across 144 configurations. Over 40% of configurations that fall below a VRMSE of 1 at short horizon exceed it at full horizon, and 12 exceed VRMSE 10 despite none doing so locally.

Vision

Camera software certifies best-fit position and orientation estimates under moderate noise

The authors propose PnP*, the first practical solver that directly optimizes the Maximum Likelihood Estimation (MLE) for Perspective-n-Point (PnP) and returns a certifiably globally optimal solution. By reformulating PnP as a quadratic program and deriving a convex semidefinite relaxation that is provably exact under moderate noise, they achieve a stable and scalable solver that runs at over 10 Hz.

Language models

Language models can produce better writing without sacrificing variety

The authors propose PRISM, a training-time framework that internalizes output diversity in LLMs. By pairing each prompt with diverse multi-teacher solutions and injecting Gaussian noise into special-token embeddings during training, the model learns to use that noise as a signal to produce varied outputs. On writing and reasoning benchmarks, PRISM improves both quality and diversity, shifting the Pareto frontier outward.

Generative media

Image generation improves through verified repairs without retraining the generator

Han and colleagues introduce ImageRSI, a system that checks generated images, repairs their prompts and saves repair strategies only after comparative testing. With the image generator held fixed, the authors report Checklist Strict Accuracy of 52.6% on MindBench and 65.66% on their new ImageReq-Bench, exceeding the strongest compared baselines by 10.2 and 9.28 percentage points, respectively.

AI for science & health

AI model estimates treatment effects faster without retraining for each dataset

The authors introduce PICLE-PFN, the first foundation model for longitudinal causal effect estimation that supports both static and dynamic treatment policies. The model represents target policies as in-context demonstrations, is pretrained on synthetic structural causal models, and predicts conditional average potential outcomes directly. On benchmarks it matches or outperforms conventional estimators while being about 100 to 5000 times faster at deployment.

Theory

Language models can transfer specialized skills across sizes without retraining

The authors propose Ariadne, a training-free approach that transfers task-specific behavior across Transformer architectures by focusing on the activation changes induced by fine-tuning rather than moving parameters directly. They show that aligning these activation displacements via local Procrustes maps and then synthesizing a target update through closed-form ridge regression improves target-model performance while requiring less time and memory than prior cross-architecture rebasin methods.

Robotics & embodied AI

Hand-held skill interface enables portable data collection for dexterous robot hands

The authors present DexPort, a hand-held interface for collecting demonstrations for high-degree-of-freedom robot hands. Operators move the device to specify wrist motion and select from a small set of skills to coordinate fingers, while the interface records native commanded joint targets and measured states. Behavior cloning policies trained on these demonstrations completed four real-world dexterous tasks and outperformed a state-as-action baseline.

Reinforcement learning

Regularised reinforcement learning boosts policy robustness in sim-to-real transfer

The authors propose SMM-AC, a regularised RL algorithm extending soft mellowmax to continuous control, and evaluate it against PPO, SAC, and TD3 on MuJoCo benchmarks and real robots (Franka arm, ANYmal-D). They find that regularised methods, particularly SMM-AC, yield higher real-world returns and maintain stronger performance under perturbations to observations, environment, and dynamics, demonstrating that regularisation-induced robustness translates to more reliable real-world deployment.

Evaluation & benchmarks

AI annotators better reflect human labeling disagreements after learning from feedback

The authors construct ANT-A, a multi-annotator benchmark spanning 7 datasets across domains and modalities, each with detailed guidelines and 5 independent human labels per instance. They propose ANT-OR, a multi-agent annotation framework that refines prompts by distilling contrastive decision rules from positive and negative feedback. ANT-OR improves agreement@>=2 accuracy by 4.5 points over the strongest baseline and achieves Fleiss kappa within 2 points of human annotator distributions.