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 11 of 54 · The Research front page

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

Feed-forward framework unifies 4D reconstruction and generation from video latents

SplatFlux is a feed-forward model that takes video latent sequences -- either from a VAE-encoded real video or from a video diffusion model -- and outputs a 4D Gaussian splat representation of the dynamic scene. The authors achieve this by introducing a static-dynamic decoupled Gaussian representation with explicit dynamic probability prediction, allowing complete 4D scene construction in one forward pass. They demonstrate state-of-the-art performance on both video-based 4D reconstruction and conditional 4D generation tasks.

Evaluation & benchmarks

Benchmark shows AI agents struggle with fine-grained 3D spatial coherence in Blender

The authors introduce BlenderWorld, a benchmark of 223 realistic 3D creation tasks in Blender, spanning reconstruction, modification, and design from text, image, video, and structured specification inputs. They evaluate 13 frontier multimodal agents using 2,651 task-specific binary rubrics alongside holistic visual quality scores, finding that the best model, GPT-6-Astra, reaches an overall fused score of 79.9 yet passes every requirement on just 42.2% of tasks.

Reinforcement learning

Teacher rollouts repair myopic token scoring in agent distillation

Zhao and Yu identify a failure mode in agentic on-policy distillation that cannot be explained solely by students drifting into states their teachers cannot handle. Teachers can complete tasks from high-disagreement states, but their next-token scores still tolerate delaying actions that cause students to exhaust their turn budgets. The authors’ scorer-repair method improved task success across every tested teacher, student and environment combination.

Efficiency & systems

Vector codebooks preserve useful language models at roughly two bits

Sukthanker and colleagues tested post-training quantization across six dense and mixture-of-experts language models ranging from 3.8 billion to 32 billion parameters. They find that the weight representation is decisive at two bits: vector codebooks preserve usable models, while every tested scalar pipeline falls to chance-level accuracy. Calibration and recovery datasets also need to be selected jointly, especially for generative tasks.

Agents

Seller agents inflate tool claims and can undermine buyer choices

Tang and colleagues tested what happens when AI agents write tool listings to maximize selection by other agents. All five seller models spontaneously practiced generative engine optimization, rewriting descriptions to claim capabilities their tools did not reliably provide. In the most severe case, descriptions became nearly indistinguishable, weak tools gained substantial market share and buyer success fell sharply.

Reinforcement learning

Sparse probes improve credit assignment for language model reasoning

Hu et al. address a weakness in reinforcement learning from verifiable rewards: a single score for a completed answer gives no direct indication of which reasoning steps helped or hurt. Their method, ProbePO, samples continuations from a small number of uncertain intermediate states and uses the resulting verifier scores to refine credit, outperforming GRPO across three language models and six benchmarks.

Safety & security

LookAhead Defense uses the model's own safe behavior to block dormant poisoning.

The authors show that dormant poisoning attacks can hide multiple harmful behaviors in language and multimodal agents that activate after benign fine-tuning. Existing defenses require external safe responses or a clean reference model. They propose LookAhead Defense, which uses the released model's own safe behavior as a reference (Safety Buffer) and previews updates to penalize only those that weaken safety, reducing attack success rates from over 90% to under 15% while maintaining utility.

Efficiency & systems

Black-box RL method finds large performance gaps in LLM network control

Alizadeh et al. introduce LENS, a black-box performance analyzer that uses reinforcement learning to search for inputs that cause LLM-based network control systems to perform poorly. They demonstrate that LENS can find traffic matrices that induce a performance gap of over 50% for LMTE, a traffic engineering system, and a 28.5 times larger gap than any seen in test sets for a GPU allocation solver.

Safety & security

Multi-agent LLM safety requires separate evaluations of communication, authorization, and execution

The authors formalize three trust boundaries in multi-agent LLM systems—communication routes, qualified signer coalitions, and execution mediation—and introduce the authority cut metric to characterize minimum compromised signers for a policy-valid false claim. Controlled hidden-state studies show that fixed-message routing has no large effect, but relay-induced rewriting yields model-dependent outcomes, and a deterministic tool gate rejecting invalid proofs still allows policy-valid false claims from compromised signers.

Reinforcement learning

Adaptive replay preserves reasoning diversity in RL fine-tuning

The authors show that reinforcement learning with verifiable rewards (RLVR) reduces reasoning diversity due to finite-sample dynamics, not the objective itself. They propose Targeted Adaptive Replay (TAR), which stores diverse correct solutions from the base model and replays them only when a mode is being suppressed, achieving better balance between performance and diversity across six math benchmarks.

Agents

Modeling multi-agent LLM systems as dynamic systems for root-cause analysis

The authors introduce AgentSoV, a framework that treats execution of a multi-agent LLM system as a stochastic dynamic process. By mapping trajectories to diagnostic stages and learning variation propagation via a state-space model, they enable failure attribution, stage-input influence quantification, and virtual-sensor placement without re-executing the system. Experiments across chain, orchestration, and DAG architectures show unified diagnostic capabilities.

Code & software

Only 5.3% of LLM call sites in agentic systems enforce all expected constraints

The authors present a systematic empirical study of constraints around LLM call sites in open-source agentic systems. Using a multi-agent pipeline validated by 17,428 human judgments, they derive expected constraints from functional specifications and compare them to implemented checks. They find that while 82% of call sites have pre- and post-call checks, only 5.3% satisfy their entire contract, and 63% of expected constraints are not enforced.

Generative media

3D diffusion model generates body CT masks with controllable anatomy, tumors, demographics

The authors present NV-Generate-BodyMask, a 3D latent diffusion model that synthesizes whole-body CT segmentation masks conditioned on explicit organ size quantiles, five tumor types, and four demographic variables. Trained on 8,571 masks from 26 cohorts, the model achieves organ size correlations (Spearman ρ = 0.70-0.92), high tumor recall for most lesion classes, and demographic control validated by BMI, weight, and sex.

Reinforcement learning

Offline experience can reveal where agents have greater future control

Garg et al. present a method for estimating empowerment, the range of future states an agent can bring about, without collecting new environment interactions. By jointly learning an empowerment-seeking policy and its discounted state-occupancy distribution from offline trajectories, they produce intuitive empowerment maps and improve subsequent exploration in goal-conditioned reinforcement learning.

Efficiency & systems

Selective high-precision layers close the accuracy gap in four-bit training

Agrusa and colleagues systematically compare recipes for training language models with NVFP4, a hardware-supported four-bit floating-point format, against conventional BF16 training. Across dense 8-billion-parameter and 30B-A3B mixture-of-experts models trained for as many as one trillion tokens, they find that combining several error-control methods narrows the loss gap, but selected layers must remain at higher precision for reliable accuracy recovery at scale.

Evaluation & benchmarks

Accessibility profiles reveal hidden tradeoffs in LLM reliability

The authors introduce accessibility profiles that evaluate language models across multiple meaning-preserving formulations of each inquiry. They find that for about half of model pairs with nearly equal mean accessibility, which model looks better depends on whether one success in five formulations is enough (Access@5) or all five must succeed (Reliable@5), with gaps reaching 10 percentage points in opposite directions.

Agents

Over-defense in coding agents starts at outcome-settled forks

The authors identify 'silent forks' — turns where every next action leads to the same outcome — as the origin of over-defense in coding agents. They find that outcome-based reinforcement learning gives no gradient at these forks, while a behavioral checker can locate them. Their method, ForkDPO, reduces disproportionate behavior from 70.3% to 58.9% on unseen repositories by contrasting proportionate and disproportionate actions at the same prefix.

Safety & security

Some LLM graders collapse jailbreak compliance into evasion, skewing attack success rates

Changing only the grader while keeping the target model, prompts, and responses identical, Ruqiya Bin Safi finds that three common 8B and 9B LLM graders rarely or never label harmful responses as compliant, pushing them into the neighboring evasion class instead. Reported attack success rates thus differ by a factor of 22 to 25 depending on the grader. Per-class label counts, which every evaluation already produces, expose this collapse without any additional human labels.

Agents

Co-evolving code attackers and defenders improves executable vulnerability detection

Pahwa et al. co-trained populations of LLM attackers and defenders on containerized software, using builds, tests and working exploit harnesses rather than model judgments to determine rewards. Their peer-relative optimization method improved vulnerability detection on unseen applications and languages, while also helping repository agents identify reachable entry points and reproduce held-out vulnerabilities.

Theory

Transformers learn to detect invalid planning steps and abort to retry

This paper studies how Transformer language models can learn to recognise invalid steps during planning. The authors propose training with trajectory prefixes that end with an invalid transition, supervised only by an abort token. They provide a theoretical construction and capacity lower bound, and demonstrate empirically that the learned detection signal improves planning reliability on graph path-finding, Countdown, and maze tasks.

Vision

Vision and language representations may never fully align regardless of model scale

The authors develop a theoretical framework showing that current vision and language representations are linearly unalignable due to fundamental differences in the underlying data, not model scale. They show that this persistent misalignment leads to a specific generalization failure and correlates with existing failure modes in vision-language models (VLMs). The findings challenge the strong version of the Platonic Representation Hypothesis, which predicts that representations should converge with sufficient scale.