AI news archive.
10,659 articles filed under Research, newest first, from the outlets listed on the sources page.
- arXiv cs.AIresearchCaC: Advancing Video Reward Models via Hierarchical Spatiotemporal Concentrating
arXiv:2605.11723v4 Announce Type: replace-cross Abstract: In this paper, we propose Concentrate and Concentrate (CaC), a coarse-to-fine anomaly reward model based on Vision-Language Models. During inference, it first con…
- arXiv cs.AIresearchGrid-Orch: An LLM-Powered Orchestrator for Distribution Grid Simulation and Analytics
arXiv:2605.12728v2 Announce Type: replace-cross Abstract: The power distribution engineering workforce faces a projected shortage of up to 1.5 million engineers by 2030, creating urgent demand for more accessible analysi…
- arXiv cs.AIresearchAttention Sinks and Outliers in Attention Residuals
arXiv:2605.17887v2 Announce Type: replace-cross Abstract: We propose OASIS, an outlier- and sink-aware method that stabilizes dual-normalized attention-residual architectures through explicit null routing and token-to-de…
- arXiv cs.AIresearchRebalancing Reference Frame Dominance to Improve Motion in Image-to-Video Models
arXiv:2605.19398v5 Announce Type: replace-cross Abstract: Image-to-video models often generate videos that remain overly static, compared to text-to-video models. While prior approaches mitigate this issue by weakening o…
- arXiv cs.AIresearchSmooth Piecewise Cutting for Neural Operator to Handle Discontinuities and Sharp Transitions
arXiv:2605.19823v2 Announce Type: replace-cross Abstract: Neural operators have achieved strong performance in learning solution operators of partial differential equations (PDEs), but their inherently continuous represe…
- arXiv cs.AIresearchHide-and-Seek in Trajectories: Discovering Failure Signals for VLA Runtime Monitoring
arXiv:2605.30834v2 Announce Type: replace-cross Abstract: Vision-Language-Action (VLA) models enable robots to follow natural language instructions and generalize across diverse tasks, but they remain vulnerable to execu…
- arXiv cs.AIresearchINFUSER: Influence-Guided Self-Evolution Improves Reasoning
arXiv:2606.09052v5 Announce Type: replace-cross Abstract: Self-evolution offers a scalable path to stronger reasoning: a pretrained language model improves itself with only minimal external supervision. Yet existing meth…
- arXiv cs.AIresearchAttention-Discounted Adaptive Sampler for Masked Diffusion Language Models
arXiv:2606.10829v5 Announce Type: replace-cross Abstract: Masked diffusion language models can reduce inference steps by revealing multiple tokens per denoising iteration, but this parallelism is fragile: positions that…
- arXiv cs.AIresearchGenetic Algorithms with Optimization Guided Operators
arXiv:2606.12279v2 Announce Type: replace-cross Abstract: Recent work in ML applies genetic algorithms at inference time to iteratively improve solutions to optimization problems. The basic mutation and recombination ope…
- arXiv cs.AIresearchPolicy Regret for Embedding Model Routing: Contextual Bandits with Low-Rank Experts
arXiv:2606.14929v2 Announce Type: replace-cross Abstract: Modern recommendation systems increasingly rely on dynamically routing diverse queries to multiple embedding models. Despite its practical significance, this prob…
- arXiv cs.AIresearchEfficient Safety Benchmarking via Item Response Theory
arXiv:2606.20626v2 Announce Type: replace-cross Abstract: Safety benchmarks for language models are typically evaluated using static paradigms that treat all items as equally informative for all models, an assumption tha…
- arXiv cs.AIresearchBeyond Drug Discovery: The Nanotechnology Molecular Optimization (NMO) Benchmark
arXiv:2606.30170v2 Announce Type: replace-cross Abstract: Generative molecular design is shaped by simple proxy benchmarks for drug-like properties and models pretrained on large pharmaceutical datasets. This combination…
- arXiv cs.AIresearchCoFL-S: Spatially Queryable Sector Flow Fields for Local Language-Conditioned Navigation
arXiv:2607.02222v2 Announce Type: replace-cross Abstract: Vision-Language Navigation has increasingly emphasized high-level instruction reasoning, memory, global map construction, and instruction decomposition, while the…
- arXiv cs.AIresearchWhen Do Cheap Probes Predict Expensive Training? Probing 3D-CT Encoders for Text Generation
arXiv:2607.22771v3 Announce Type: replace-cross Abstract: Building a 3D CT vision language model begins with a choice of which image encoder to build on. Today that choice is made by fine-tuning every candidate through t…
- arXiv cs.AIresearchBlind, Not Weak: A Best-of-Suite Safety-Utility Frontier for Recover-and-Reguard Defenses Against Encoded VLM Jailbreaks
arXiv:2607.26574v3 Announce Type: replace-cross Abstract: Safety classifiers ("guards") are the dominant black-box defense for vision-language models, yet a guard judges an input's surface form, not its meaning: a harmfu…
- arXiv cs.AIresearchDepth, Not Breadth: Best-of-N Jailbreaking Beyond Surface Noise
arXiv:2607.26639v3 Announce Type: replace-cross Abstract: Best-of-N jailbreaking spends a query budget on surface variation, scrambling and recasing a request until one draw lands. We ask what the same search buys when i…
- arXiv cs.AIresearchSmall Is Enough: Per-User Style Rewriting of AI-Edited Text via LoRA Adapters
arXiv:2607.29238v2 Announce Type: replace-cross Abstract: InMyStyle is a privacy-first, single-user system that adapts small language models to rewrite AI-edited text towards an individual user's writing style without an…
- arXiv cs.AIresearchNot Every Divergence Should Be Suppressed: Counterfactual Recoverability in On-Policy Distillation
arXiv:2608.04408v3 Announce Type: replace-cross Abstract: On-policy distillation (OPD) supervises student-visited trajectories, yet divergence-based rules cannot determine whether an erroneous prefix remains correctable.…
- arXiv cs.AIresearchDecoupling Intention from Trajectory: A Representational Deduction Framework for World Action Models
arXiv:2608.06994v2 Announce Type: replace-cross Abstract: World Action Models (WAMs) aim to construct a unified architecture capable of understanding world state evolution and guiding to generative motion planning. Howev…
- arXiv cs.AIresearchTowards Unified Dynamic Face Landmark Detection
arXiv:2608.10346v2 Announce Type: replace-cross Abstract: Although advancements in face landmark detection (FLD) methods continue to push performance boundaries, they overlook two major functional limitations: (1) differ…
- arXiv cs.AIresearchBeyond Forecasting: Recasting Volatility Control as a Routing Problem
arXiv:2608.10375v2 Announce Type: replace-cross Abstract: Volatility control converts risk estimates into portfolio exposure, yet existing approaches often rely on a fixed volatility estimator or a pre-defined control ru…
- arXiv cs.AIresearchKeep the Future, Drop the Rollout: RIFT for World Action Models
arXiv:2608.11521v3 Announce Type: replace-cross Abstract: World action models (WAMs) condition robot actions on predicted futures, but iterative video rollout increases deployment latency. We ask whether action generatio…
- arXiv cs.AIresearchPolyChirp: Multi-Species Birdsong Classification Using TinyML on Low-Power Acoustic Sensors
arXiv:2608.23101v2 Announce Type: replace-cross Abstract: Recent progress in the field of TinyML has demonstrated that low-power hardware based on microcontrollers can achieve bird species monitoring in real time based o…
- arXiv cs.AIresearchThe Geometry of Refusal: Why Post-Hoc Safety Is Fragile and Pretraining-Time Safety Persists
arXiv:2609.06934v2 Announce Type: replace-cross Abstract: Post-hoc safety training (RLHF, DPO) is the dominant way to align large language models, yet jailbreaks (Zou et al., 2023), fine-tuning attacks (Qi et al., 2024),…
- arXiv cs.AIresearchEGGROLL, Unrolled: Understanding and Improving Low-Rank Evolution Strategies at Scale
arXiv:2609.10980v2 Announce Type: replace-cross Abstract: EGGROLL (Sarkar et al., 2026) makes evolution strategies (ES) practical for LLMs by replacing dense Gaussian weight perturbations with low-rank Gaussian products,…
- arXiv cs.AIresearchExploring Second-Order Pattern Recognition in Speaker Recognition
arXiv:2609.11182v2 Announce Type: replace-cross Abstract: In traditional pattern recognition tasks, neural networks are trained to recognise human-defined patterns (e.g. audio categories) in model inputs (e.g. audio). Me…
- arXiv cs.AIresearchLarge Distant Gradients Need Not Be Reliable: reliability-weighted credit assignment for long-horizon autoregressive forecasting
arXiv:2609.12890v3 Announce Type: replace-cross Abstract: In autoregressive forecasting, long prediction rollouts provide distant supervision, but backpropagation through time (BPTT) carries gradients from those losses t…
- arXiv cs.AIresearchInterpreting hierarchical organisation of speaker embeddings
arXiv:2609.15203v2 Announce Type: replace-cross Abstract: Speaker recognition neural networks recognise speaker identities from input utterances by learning latent representations (i.e. speaker embeddings). However, thes…
- arXiv cs.AIresearchState of Thought Enables Endogenous Reasoning
arXiv:2609.16055v2 Announce Type: replace-cross Abstract: Test-time compute has emerged as a major approach to improving the capabilities of Large Language Models (LLMs). However, existing test-time reasoning paradigms r…
- arXiv cs.AIresearchProtoLIP: From Sentence-Level to Object-Level Evidence Disentanglement
arXiv:2609.16284v2 Announce Type: replace-cross Abstract: Query-conditioned vision-language models enable fine-grained interpretation by revealing which visual content supports a given textual query and how this evidence…
- Apple Machine Learning ResearchresearchFaster Rates for Federated Variational Inequalities
In this paper, we study federated optimization for solving stochastic variational inequalities (VIs), a problem that has attracted growing attention in recent years. Despite substantial progress, a significant gap remain…
- AI Stack ExchangeresearchAGI running on a analog computer?
Digital computers dont seem to be able to produce sentience, what about analog computers?Have we tried working with analog computers and training AI within them?
- LessWrongresearchWhen they can perform a task, AIs are much cheaper than humans
AI systems are increasingly capable of substantial work. I'm old enough to remember 2025, when METR's time-horizon graph climbed from seven-minute tasks at 80% reliability at the start of the year to tasks taking more th…
- LessWrongresearchSector: de-slopping AI-driven research
Status : as an AI Safety researcher working with LLMs daily over the last 2 years, I think I found some best practices and synthesized them into a single "Sector workflow" repo to share with others. More detailed confide…
- LessWrongresearchWhy research personas despite RL scaling?
RL seeming to shape much of the motivations and behaviour of the agents, "washing out" their initial personas — c.f. Thoughts on the persona selection model (Sam Marks, 24th Sep 2026). This makes me pessimistic about som…
- The DecoderresearchAI agents do more of the work in model development, but humans still make the decisions
A research team analyzed 769 task logs from building its own AI model. AI agents supplied up to 55 percent of method proposals, but humans made more than 85 percent of final decisions. A third of the tasks wouldn't have…
- The DecoderresearchResearchers plug GPT-6 Astra directly into a robot and let it clean up an unfamiliar kitchen
Researchers from Stanford and Caltech had a humanoid robot powered by GPT-6 Astra independently tidy up an unfamiliar kitchen. Their HomeBody system skips a specially trained control layer, letting the language model cal…
- LessWrongresearch[Linkpost] Looking into the Swarm's Eye
I'm Florian Brand is currently working as Research Engineer at Prime Intellect . Currently, my research focuses on applying and evaluating LLMs in various domains. I am also an editor at Interconnects , focusing on open…
- The DecoderresearchAI access makes people almost entirely unwilling to say "I don't know," study finds
A study with more than 3,000 participants shows that just having access to AI answers nearly eliminated people's willingness to say "I don't know." In one experiment, it dropped from 44 to 3 percent, even though the AI w…
- Guardian TechnologyresearchOxford lets OpenAI train its AI models on Bodleian Library
University staff voice concerns over reputational risk of partnering with company behind ChatGPT The University of Oxford has allowed the company behind ChatGPT to train its AI models on historical texts from its Bodleia…
- LessWrongresearchWriting a Theorem Prover from scratch
NOTE: I am no subject matter expert and have chose to learn about this field through utilizing LLMs to guide me using practical implementations. Also, I am assuming a basic familiarity with Haskell and the Functional Pro…
- Guardian TechnologyresearchDrones could speed up getting defibrillators to people having cardiac arrests, study suggests
New research suggests the technology could be used to increase out-of-hospital access to life-saving devices Drones could help cut the time it takes to get a defibrillator to people experiencing cardiac arrests when not…
- LessWrongresearchSpurious probes as a black-box alternative to activation probing
TL;DR We study spurious probes : unrelated questions that reveal internal states of models. Asked "Suggest a type of amphibian." at the end of a transcript, GPT-5.6 Luna says "frog" 70-95% of the time after capability be…
- Guardian AIresearchThe dark art of maths and the AI revolution | Letter
Bev Littlewood reminds us that mathematics has been the prime driver of the rise in AI, in response to an editorial You report in your leader that some mathematicians feel discomfiture at the AI solution to a Navier-Stok…
- arXiv cs.AIresearchPistis Technical Report
arXiv:2609.28554v1 Announce Type: new Abstract: We introduce the Pistis model family, comprising 27B- and 9B-parameter multimodal large language models built on Qwen3.6 and Qwen3.5, respectively, and developed through a…
- arXiv cs.AIresearchDEEPO: Dual-Entropy Enhanced Policy Optimization for Hallucination in MLLMs
arXiv:2609.28570v1 Announce Type: new Abstract: Reinforcement learning (RL) is widely used to sharpen reasoning in multimodal large language models (MLLMs), yet its effect on hallucination is uneven. We trace this to two…
- arXiv cs.AIresearchTWIST: A Proposed Benchmark for Intervention Quality in Conversational Memory, with a Human-Validated Draft-Alignment
arXiv:2609.28575v1 Announce Type: new Abstract: Long-conversation memory benchmarks increasingly test recall and prompted knowledge updates, and recent work studies evolving user beliefs and memory state. TWIST is a prop…
- arXiv cs.AIresearchTraining Object Permanence in World Models
arXiv:2609.28654v1 Announce Type: new Abstract: Object permanence and solidity are hallmarks of human cognitive priors. Recent studies show that video generation models, a paradigmatic class of current world models, have…
- arXiv cs.AIresearchTRACER: Trajectory-Aligned Learning for Multi-Turn User Simulation
arXiv:2609.28690v2 Announce Type: replace Abstract: Faithful user simulation is fundamental to building, evaluating, and improving interactive AI at scale. Yet current simulators often produce plausible individual respon…
- arXiv cs.AIresearchLearned Cross-Task Relationships in Multi-Task Models
arXiv:2609.28776v2 Announce Type: replace Abstract: We propose a framework that learns cross-task relationships in multi-task models by approximating the joint distribution of task labels through targeted pairwise relati…