AI news archive.
19,861 articles, newest first, from the outlets listed on the sources page.
- arXiv cs.AIresearchLearning, Fast and Slow: Towards LLMs That Adapt Continually
arXiv:2605.12484v3 Announce Type: replace-cross Abstract: Large language models (LLMs) are trained for downstream tasks by updating their parameters (e.g., via RL). However, updating model parameters forces them to absor…
- arXiv cs.AIresearchGQLA: Group-Query Latent Attention for Hardware-Adaptive Decoding
arXiv:2605.15250v4 Announce Type: replace-cross Abstract: Decoding efficiency depends on KV-cache traffic, computation, and parallelism. Multi-head Latent Attention (MLA) offers absorbed and head-expanded forms, but its…
- arXiv cs.AIresearchWhen and Why Adversarial Training Improves PINNs: A Neural Tangent Kernel Perspective
arXiv:2605.15959v2 Announce Type: replace-cross Abstract: Physics-informed neural networks (PINNs) are powerful surrogates for differential equations but are notoriously difficult to train due to spectral bias, stiffness…
- arXiv cs.AIresearchSymmetry-Compatible Principle for Optimizer Design: Embeddings, LM Heads, SwiGLU MLPs, and MoE Routers
arXiv:2605.18106v5 Announce Type: replace-cross Abstract: A striking geometric disparity has long persisted in the practice of deep learning. While modern neural network architectures naturally exhibit rich symmetry and…
- arXiv cs.AIresearchHARP: Efficient Data Selection for Finetuning Large Language Models
arXiv:2606.07690v2 Announce Type: replace-cross Abstract: Finetuning data selection requires balancing two competing goals: selecting examples that improve the downstream objective, and doing so without repeatedly finetu…
- arXiv cs.AIresearchActivation Steering Induces Emergent Misalignment: Steering Vector Construction and Comprehensive Evaluation
arXiv:2606.08682v2 Announce Type: replace-cross Abstract: Activation steering (AS) has emerged as a popular inference-time technique for modulating the behavior of large language models (LLMs). AS modulates model behavio…
- arXiv cs.AIagentsSecureClaw: Clawing Back Control of LLM Agents
arXiv:2606.09549v3 Announce Type: replace-cross Abstract: Tool-using large language model (LLM) agents face two distinct security failures: unauthorized external actions and exposure of sensitive plaintext inside the run…
- arXiv cs.AIagentsDIRECT: When and Where Should You Allocate Test-Time Compute in Embodied Planners?
arXiv:2606.12402v2 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) are increasingly deployed as high-level planners for embodied agents, with an emerging strategy of scaling test-time compute to impr…
- arXiv cs.AIagentsDriving, Fast or Slow? Neuro-Symbolic Guidance for Motion Prediction in Multi-Modal Ground Mobility
arXiv:2606.15251v2 Announce Type: replace-cross Abstract: Accurate and interpretable motion prediction for heterogeneous traffic spaces, including pedestrians, bicycles, cars, and trucks, is essential for safe autonomous…
- arXiv cs.AIresearchPoLAR: Factorizing Extent and Mode in Latent Actions for Robot Policy Learning
arXiv:2606.21139v2 Announce Type: replace-cross Abstract: Robot actions vary in both extent and mode: how much change they produce and what kind of change they produce. A latent action space that reflects this structure…
- arXiv cs.AIresearchG$^3$VLA: Geometric Inductive Bias for Vision-Language-Action Models
arXiv:2606.24472v2 Announce Type: replace-cross Abstract: Vision-language-action (VLA) models have rapidly advanced generalist robot manipulation by harnessing the semantic knowledge of pretrained vision-language backbon…
- arXiv cs.AIresearchSelf-Organized Conformal Prediction: Reducing Regional Coverage Gaps with Unsupervised Group Discovery
arXiv:2606.29403v3 Announce Type: replace-cross Abstract: Conformal prediction guarantees marginal coverage, but a pooled calibration quantile can hide systematic undercoverage across heterogeneous regions of the feature…
- arXiv cs.AIresearchWhat Does a Discrete Diffusion Model Learn? One Reverse Process in Three Coordinates
arXiv:2607.05381v2 Announce Type: replace-cross Abstract: What does a discrete diffusion model actually learn? The literature trains denoisers, bridge plug-ins, and concrete scores, each deriving its own ELBO objective u…
- arXiv cs.AIresearchRhyMix: A Lightweight Adaptive Multi-Rhythm Network for Long-Term Time Series Forecasting
arXiv:2607.08234v2 Announce Type: replace-cross Abstract: Real-world time series exhibit complex dynamics characterized by multiple simultaneous temporal patterns: short-term fluctuations, periodic seasonal cycles, long-…
- arXiv cs.AIresearchCorrelation-Aware Contextual Bandits with Surrogate Rewards for LLM Routing
arXiv:2607.09015v2 Announce Type: replace-cross Abstract: We study contextual bandit problems with correlated arms and access to surrogate reward signals produced by a machine learning model, motivated by applications su…
- arXiv cs.AIresearchDoes Depth Survive Composition, and Can a Router Use It? Adaptive Compute in Latent World Models
arXiv:2607.10203v4 Announce Type: replace-cross Abstract: Adaptive-compute world models run fewer predictor blocks on some rollout steps than on others. This rests on two assumptions: that a deeper predictor gives a more…
- arXiv cs.AIresearchFormalizing Flag Algebras in Lean
arXiv:2607.23500v2 Announce Type: replace-cross Abstract: Razborov's flag algebra method is one of the most successful tools for proving asymptotic density inequalities in extremal graph theory. It reduces an inequality…
- arXiv cs.AIresearchEvery Time I Hire a Linguist, Inference Costs Go Down: On Linguistic Rules as Effective Prompt Compressors
arXiv:2607.25335v3 Announce Type: replace-cross Abstract: Prompt compression shortens LLM input to reduce inference cost, yet existing methods score token importance through LM forward passes. It remains questionable whe…
- arXiv cs.AIresearchCommit Locally, Exit Globally: Coordinating Adaptive Sampling and Early Exit in Diffusion Language Models
arXiv:2607.28166v3 Announce Type: replace-cross Abstract: Diffusion language models predict tokens in parallel but still require many iterative forward passes. Adaptive sampling accelerates local token commitment, while…
- arXiv cs.AIresearchIACM-RL: Intent-Aware Context Management and Reinforcement Learning for Complex Tool Invocation under Dynamic Intent Fluctuations
arXiv:2608.02110v2 Announce Type: replace-cross Abstract: Executing long-horizon tool invocations in real-world environments is severely challenged by dynamic user intent noise. Existing methods attempt robustness via im…
- arXiv cs.AIresearchFACT: Failure-Aware Causal Training for World-Action Models
arXiv:2608.10232v2 Announce Type: replace-cross Abstract: Recent world-action models (WAMs) show that co-training policies with future prediction can provide physical priors for action generation. Building on the future-…
- arXiv cs.AIresearchRubric Dropout: A Simple Way to Mitigate Reward Hacking in Rubric-Based RL
arXiv:2608.11669v4 Announce Type: replace-cross Abstract: Rubric-based reinforcement learning, in which an LLM judge scores responses against prompt-specific criteria, has become a standard way to post-train language mod…
- arXiv cs.AIresearchUncertainty-Aware Probabilistic Constrained Clustering from Entangled Pairwise Supervision
arXiv:2608.12027v2 Announce Type: replace-cross Abstract: Pairwise constrained clustering typically relies on hard must-link/cannot-link labels, whereas realistic pairwise supervision may be real-valued and entangle intr…
- arXiv cs.AIresearchEquilibrium Forcing: Adaptive Video Generation Without Noise Conditioning
arXiv:2608.14706v2 Announce Type: replace-cross Abstract: Standard autoregressive video generation algorithms based on Diffusion and Flow Matching rely on rigid training objectives and static sampling schedules, limiting…
- arXiv cs.AIresearchMechanistic Circuit Identification for Controllable Data Generation
arXiv:2608.24065v2 Announce Type: replace-cross Abstract: While recent advances in data synthesis aim to curate high-quality datasets, most generation pipelines still rely on heuristic prompt-based control. This black-bo…
- arXiv cs.AIresearchDCGC: Draft-Conditioned Global Correction for Complex Reasoning with Masked Diffusion Models
arXiv:2608.25428v2 Announce Type: replace-cross Abstract: Correcting flawed reasoning traces remains a significant challenge for Large Language Models (LLMs), whose autoregressive generation can propagate early mistakes…
- arXiv cs.AIresearchAdaVDR: Adaptive Tool Use and Reflection for Video Deep Research
arXiv:2608.25559v2 Announce Type: replace-cross Abstract: Video deep research aims to answer questions by jointly understanding video content and retrieving external knowledge from the open Web. However, these tasks ofte…
- arXiv cs.AIresearchTEAR: Table Extraction with Attribute Recommendation from Texts via Large Language Models
arXiv:2609.15205v2 Announce Type: replace-cross Abstract: Table extraction from texts is an important task for information systems, and recent approaches that prompt large language models (LLMs) with instructions have dr…
- Simon Willison's WeblogcommunityLaTeX renderer
Tool: LaTeX renderer Inspired by Paul Ford and Aboard Foundary's new web app Tiny Little Office , which provides a suite of web app content editors (a Word Processor, Outliner, Presentation Builder, Spreadsheet, Calculat…
- SiliconANGLE AIstartupsAhead of IPO, Anthropic reveals its revenue grew 45% sequentially to $16.8B in Q3
Anthropic PBC pulled in $16.8 billion in revenue during the third quarter, according to a report today by the Financial Times — and also achieved profitability on an adjusted basis for the first time. The revelations cam…
- BBC TechnologystartupsAI giant Anthropic woos potential investors ahead of stock market debut
It comes as some investors have become concerned about the amount of money being spent on AI infrastructure.
- TechCrunch AIchipsRicursive Intelligence is betting on AI-designed chips to help solve the compute crisis
The founders detailed how their startup is developing AI that can design chips, learn from each iteration, and improve subsequent designs, all with the goal of automating and accelerating the chip design process.
- Engadget AIchipsBee unveils two AI note-taking devices, its first new products since being acquired by Amazon
The Eclipse and Frontier are cheap and colorful, but you'll need a Pro subscription to make the most of them.
- TechCrunch AIstartupsGudea leverages AI to help organizations prevent social media crises before they erupt
Gudea, a company that tracks how stories and rumors spread online that describes itself as the “storm tracker for the internet,” is using AI to help organizations spot trouble brewing online before it turns into a crisis…
- Hugging Face Blogopen-sourceWelcome FLUX.3 Image Klein in Diffusers
Open-source model releases, tooling, and community updates.
- Apple Machine Learning Researchopen-sourceLocal and Distributed Large LLM Inference with MLX and RDMA on Mac Studio
MLX on Mac with Apple silicon lets AI researchers and developers experiment locally with the latest models and methods in a private environment. The open-source array framework leverages the unified memory architecture o…
- Hugging Face Blogopen-sourceWelcome Muse Spark Open: open-weight frontier multimodal mixture-of-experts model from Meta
Open-source model releases, tooling, and community updates.
- Apple Machine Learning ResearchresearchSelection-Based Structured Reasoning: Toward Efficient Multimodal Search Agents
Multimodal agents commonly generate freeform reasoning before each action. For small models, limited model capacity can result in lengthy reasoning that provides little useful guidance for action generation while incurri…
- Apple Machine Learning ResearchresearchPlan or Retrieve First? A Scale Study of Tool Discovery Architectures
Tool-augmented LLM agents discover relevant APIs from large registries before planning execution, yet finite context windows prohibit injecting all tool definitions into a single prompt. The dominant approach, which we t…
- OpenAI NewsagentsSullivan & Cromwell brings legal expertise into agreement review
In collaboration with OpenAI, Sullivan & Cromwell developed Agreement Analyzer, an AI tool for identifying issues in a draft agreement.
- LessWrongresearchI trained GPT-2 with full-batch gradient descent so you don't have to
Note: The results of this are not as clean as they would naively appear, since a lot of training parameters are tuned for non-full-batch gradient descent. See discussion in the comments. I've left this up in case it's st…
- The Verge AIcommunityMcKinsey is making job applicants spar with a fake AI client
Consulting firm McKinsey & Company is going to ask some entry-level candidates to have a 10-minute conversation with an AI simulation of a client as part of the interview process, according to The Wall Street Journal. Th…
- LessWrongsafetyThe basic argument for making AI better at moral philosophy
In this post, I try to briefly make the case for making models good at moral philosophy. To be up front, aside from some niche areas of moral philosophy, I don’t actually think this is the most important thing to work on…
- The Verge AIstartupsAmazon’s always-listening Bee wearable can now help you buy stuff
Last year, Amazon bought Bee, a relatively unknown startup with a $50 always-listening AI wearable that summarized every conversation you had. Today, the e-commerce giant is announcing two new Bee devices - and a new sub…
- SiliconANGLE AImodelsGrok Voice 3.0 from SpaceXSI outscores OpenAI and Google on speech-to-speech index
Elon Musk’s artificial intelligence company SpaceXSI, formerly xAI, today released Grok Voice 3.0, a new version of the speech-to-speech model behind its Voice Agent API. SpaceXSI is pitching the model to businesses that…
- Latent SpaceagentsCrossing the Threshold: How OpenAI and Anthropic Approach Autonomy and Safety in Financial Agents
At AI Engineer New York, frontier labs and financial firms explored how to give agents useful autonomy while keeping consequential decisions under control.
- Latent SpaceagentsCrossing the Threshold: How OpenAI and Anthropic Approach Autonomy and Safety in Financial Agents
At AI Engineer New York, frontier labs and financial firms explored how to give agents useful autonomy while keeping consequential decisions under control.
- LessWrongcommunityWhen bureaucracy gives you something it shouldn't
Epistemic status: two cases and the statute behind them. The first case rests on one side's account in a lawyer's call-in video; I have not seen her file. I am not a lawyer. I could not find a base rate. AI use: a Codex…
- LessWrongcommunityAI Marketplaces Can Contribute to, and Profit From, the AI 2040 Plan: Research Directions and Strategy
Epistemic status: Draft-Quality. Partially motivated reasoning. After my “ Not Pinning Your OpenRouter Provider Might Invalidate Your Research ” post, a representative of an AI marketplace reached out to me about provide…
- TechCrunch AIsafetyFlock CEO says its microphones detect ‘just gunshots’ — its website says different
Langley’s comments come amid mounting scrutiny of Flock from privacy advocates, including the ACLU, and members of the public who are concerned about the misuse of its technology.