AI news archive.
10,659 articles filed under Research, newest first, from the outlets listed on the sources page.
- arXiv cs.AIresearchA Training-Free Proactive Defense Against Partial Speech Manipulation via Self-Embedding Steganography
arXiv:2608.25285v2 Announce Type: replace-cross Abstract: Partial deepfake speech, where only limited segments of an utterance are synthesized or manipulated, poses a significant challenge to existing deepfake detection…
- arXiv cs.AIresearchPAVE: Predictive Alignment and Value-Guided Evolution for World-Action Policies
arXiv:2608.30378v3 Announce Type: replace-cross Abstract: Direct vision-language-action policies generate continuous robot actions efficiently, but standard behavior cloning leaves two complementary gaps: their represent…
- arXiv cs.AIresearchTabScope: Question-Adaptive Scope Selection for Table Question Answering
arXiv:2609.03395v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have shown strong performance on table question answering, yet their accuracy often degrades as table size increases. We find that th…
- arXiv cs.AIresearchVLA-Precision: Asymmetric Co-Bootstrapping for Efficient Real-World Online RL of Vision-Language-Action Models
arXiv:2609.04355v3 Announce Type: replace-cross Abstract: Pretrained vision-language-action (VLA) models enable broad manipulation but remain unreliable in tasks demanding precision and repeatability. Applying real-world…
- arXiv cs.AIresearchFine PT-PT Web: A High-Quality 41 Billion Tokens Data Collection of the European Portuguese Web
arXiv:2609.07699v3 Announce Type: replace-cross Abstract: Curating Web corpora for regional language variants like European Portuguese (PT-PT) is heavily bottlenecked by dialectal overlap (mainly with PT-BR) and data pro…
- arXiv cs.AIresearchDo New Attention Mechanisms Actually Fix Attention Sinks at Million-Token Context?
arXiv:2609.08574v2 Announce Type: replace-cross Abstract: Long context language models now advertise windows of one million tokens, but two habits limit how much of that window is used. Attention heads with nothing usefu…
- arXiv cs.AIresearchHigh-probability guarantees for linear accessibility in feature superposition
arXiv:2609.09556v3 Announce Type: replace-cross Abstract: Neural networks can leverage feature superposition to encode more concepts than dimensions, but cross-feature interference constrains the linear accessibility of…
- arXiv cs.AIresearchData-Free On-Policy Distillation: How Far Can We Go Without External Data?
arXiv:2609.14193v4 Announce Type: replace-cross Abstract: On-policy distillation (OPD) is increasingly applied to frontier foundation model post-training. Prior work in this area has largely focused on algorithmic advanc…
- arXiv cs.AIresearchA primer on evaluation methods for large language models in healthcare
arXiv:2609.14819v3 Announce Type: replace-cross Abstract: Large language models (LLMs) have a growing range of applications in medicine, and their evaluation is critical for ensuring they provide benefit and not harm. Th…
- TechCrunch AIresearchWorld model companies are keeping a lot of secrets
Everyone in the world-models space is sitting on a pile of cash and a ton of buzz, but good luck getting anyone — from the founders to their own data suppliers — to tell you what they're actually building.
- LessWrongresearchEvaluating task vectors, unlearning and inoculation
TL; DR In the previous post I introduced some ideas and similarities between unlearning and inoculation, as well as a distinction between learned and human-written adapters. This post serves as a short empirical evaluati…
- AI Stack Exchangeresearchreal vs synthetic vs semi-synthetic image classification model training
I am currently using working on training image model for classifying real, synthetic, semi-synthetic images. I am using transfer learning on dinov3-vit-large model. I have data corpus about 330k images. In the phase 1 ex…
- Guardian AIresearchThe Guardian view on AI v mathematicians: humans are still vital to the field, but tech firms refuse to see that | Editorial
OpenAI claims its agents have solved the Navier-Stokes problem. But questions remain about how useful and independent its work really is On 8 September, OpenAI claimed that its AI agents had solved the Navier-Stokes prob…
- LessWrongresearchLabs could soon start automated research into architectures driven by no-CoT perfomance
No-CoT performance has been a decent proxy for tracking the g-factor intelligence of base models (see work on latent multi-hop reasoning by Ryan Greenblatt ). I think that this capability is very beneficial for reasoning…
- The DecoderresearchSimulated students that make realistic mistakes help AI tutors learn faster
Microsoft and the University of Illinois built StudentSim to replicate individual students from limited data and give AI tutors fast, low-cost feedback. In tests covering 60 students across chess, English, and math, it o…
- LessWrongresearchFailure of the coding theorem for randomized stopping machines
Epistemic Status and Contributions. This post explains a technical separation result in algorithmic information theory which was derived during Mikhail Mironov's Summer 2026 PIBBSS fellowship . The result contributes to…
- The DecoderresearchAI conference ICLR is drowning in abstracts, with roughly 50,000 submissions before the deadline
ICLR 2027 has already pulled in about 50,000 abstracts, up from 19,500 at ICLR 2026. The flood is driven by the AI hype, corporate pay tied to publication records, and above all, AI making it faster to crank out papers.…
- WIRED AIresearchMathematicians Hate AI. They Can’t Quit It
Powerful AI models have created an existential risk to the field, but researchers can’t stop relying on them because they’re too useful.
- LessWrongresearchKoopman Theory and Metaethics
tl;dr: The constructivist view of natural selection as the main source of our moral views can be expressed in a surprisingly tractable mathematical form. Measurements of the universe can be rolled forwards in time using…
- BBC TechnologyresearchTech Now
Adrienne Murray follows NASA scientists in Iceland researching a rare total solar eclipse.
- TechCrunch AIresearchWorld model companies are keeping a lot of secrets
Everyone in the world-models space is sitting on a pile of cash and a ton of buzz, but good luck getting anyone — from the founders to their own data suppliers — to tell you what they're actually building.
- SiliconANGLE AIresearchAnthropic opens AI-powered biology research lab
Anthropic PBC has opened a wet lab, a facility dedicated to biology research, in the San Francisco Bay Area. Reuters reported today that the company will use robots to automate certain scientific tasks at the hub. The ro…
- IEEE Spectrum AIresearchParallel Reads and Write Optimization for Large-Scale Data Replication
This White Paper gives data engineers and architects a practical overview of how parallel partitioned reads, write-path optimization, and cloud-native bulk loading reduce large-table replication times, and why replicatio…
- LessWrongresearchPretraining data, not verifiability, is why LLMs are especially good at math (and coding)
Follow-up to: “LLMs are (still) mostly powered by imitative learning, not RL” [Edited a bit since publishing, see changelog at the bottom.] A common take I’ve been hearing is: “LLMs are especially good at math (compared…
- LessWrongresearchThe J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
Welcome back to the Digital Minds Newsletter, your curated guide to the latest developments in AI consciousness, digital minds, and AI moral status. If you enjoy this newsletter, please consider sharing it with others wh…
- Google AI BlogresearchNew experts join Google’s AI & Economy team
We are expanding our AI & Economy team with world-class academic advisors, fellows, and core internal researchers.
- LessWrongresearchCollective Epistemics: Napkin Math on Independent Errors
Part of a larger series I want to put together on some of the basic equations and models of collective epistemics from the more mathy side of the social sciences. LLM Status: Pictures + Picture descriptions are LLM-assis…
- AI Stack ExchangeresearchCan the hidden-state dimension of a selective state-space model be interpreted as an effective tensor-network bond dimension?
I am trying to understand whether there is a precise tensor-network interpretation of the state dimension in modern selective state-space models such as Mamba/SSD. Consider a selective state-space update $$ h_t = A(x_t)h…
- AI Stack ExchangeresearchHow do Energy Based Models solve Multiple possible outputs given one input
I've been looking into Energy Based Models recently which Yann LeCun has been strongly advocating for. One problem that he lists with probabilistic based models is that in the case when there are multiple possible output…
- Google ResearchresearchNew research shows how AMIE, our medical AI, could help manage health conditions.
Research-heavy updates from Google Research.
- Google ResearchresearchThree new satellites join the fight against wildfires.
Research-heavy updates from Google Research.
- arXiv cs.AIresearchRegularized Emphatic Temporal-Difference Learning: Stability under Constant Stepsizes
arXiv:2609.19170v1 Announce Type: new Abstract: Emphatic temporal-difference learning (ETD) stabilizes the expected off-policy TD update and changes its projection geometry, but neither property determines constant-steps…
- arXiv cs.AIresearchBioPhys-Bridge: A Benchmark for Interdisciplinary Scientific Reasoning in Physics-Grounded Biological Research
arXiv:2609.19180v1 Announce Type: new Abstract: Language models face unique challenges in analyzing interdisciplinary scientific research literature. In biophysics research, faithful answers require grounding observed da…
- arXiv cs.AIresearchWhat Do We Expect from LLMs? Mapping the Design of LLM Benchmarks
arXiv:2609.19182v1 Announce Type: new Abstract: Benchmarks are central to how progress in large language models (LLMs) is assessed and communicated. Yet model rankings alone reveal little about how evaluation requirement…
- arXiv cs.AIresearchWhat Do Current Systematic Generalization Tasks Miss? A Reasoning-Centered Analysis
arXiv:2609.19212v2 Announce Type: replace Abstract: Systematic generalization, the ability to solve novel problems by recombining known atomic elements, is central to human intelligence but difficult to study rigorously…
- arXiv cs.AIresearchThe syntax and semantics of goals
arXiv:2609.19448v1 Announce Type: new Abstract: In both cognitive science and computer science, goals are conceptualized as cognitive states that flexibly combine with world knowledge to organize and specify purposeful b…
- arXiv cs.AIresearchCompositional Reasoning in Language Models under Reinforcement Learning Post-Training
arXiv:2609.19465v1 Announce Type: new Abstract: Compositional reasoning is critical for real-world problem solving: since training data is necessarily limited, models must generalize by composing learned skills in new wa…
- arXiv cs.AIresearchQVAC Genesis III: A Large-Scale, High-Quality Open Synthetic STEM Corpus for Efficient Language Model Pre-Training
arXiv:2609.19513v2 Announce Type: replace Abstract: High-quality pre-training data is a critical bottleneck for educational and STEM-specific language models targeting edge AI and on-device deployment where token budgets…
- arXiv cs.AIresearchLLM-as-an-Improver: Turning Verification into Better Candidates
arXiv:2609.19515v1 Announce Type: new Abstract: Verifier-based selection improves LLM performance by generating multiple candidate solutions and using a verifier to select the most promising one. However, existing method…
- arXiv cs.AIresearchContinual Enterprise World Model Discovery in Dynamic Systems
arXiv:2609.19551v1 Announce Type: new Abstract: In an enterprise system, updating one field can set another, create a record, or start an approval. These effects are produced by business rules that are not built into the…
- arXiv cs.AIresearchWhen2Think: Learning When and How Much to Reason
arXiv:2609.19671v2 Announce Type: replace Abstract: Large Reasoning Models (LRMs) often overthink easy problems and underthink hard ones, leading to inefficient computation allocation. Existing methods regulate generated…
- arXiv cs.AIresearchTorchCraft: Unified binder design by inverting an all-atom structure predictor
arXiv:2609.19770v1 Announce Type: new Abstract: All-atom structure predictors model diverse molecular interactions, but using their learned structural priors for binder design remains challenging. Here we present TorchCr…
- arXiv cs.AIresearchIntegrating knowledge from case reports: a medical ontology based multimodal information system with structured summary
arXiv:2609.19775v1 Announce Type: new Abstract: Published medical case reports serve as a crucial medical information carrier, documenting discoveries in rare diseases, diagnostic methods, and innovative treatments. Desp…
- arXiv cs.AIresearchSteering Equilibrium Selection in Regularized Self-Play via the Reference Policy
arXiv:2609.19820v1 Announce Type: new Abstract: Regularized self-play -- the family behind DeepNash's Stratego play -- drives a two-player zero-sum policy to a Nash equilibrium by best-responding to a slowly moving, entr…
- arXiv cs.AIresearchMetaRTL: Meta-path Attention Enhanced Relational Table Learning
arXiv:2609.19832v1 Announce Type: new Abstract: Relational table learning has gained increasing attention with the widespread use of relational databases. Existing methods typically rely on deep GNN or HGNN stacks, leadi…
- arXiv cs.AIresearchConstraint-Safe Graph-Context Scoring for Stable Point-Feature Labels Under Text-Width and Accessibility-Inspired Profiles
arXiv:2609.19848v1 Announce Type: new Abstract: Point-feature label placement on interactive maps must reconcile geometric validity, display yield, local placement utility, and stability across camera motion. Accessibili…
- arXiv cs.AIresearchReproducibility is not construct validity: LLM measurement of institutionally situated communication
arXiv:2609.19866v1 Announce Type: new Abstract: High annotation reproducibility does not necessarily imply that an LLM-inferred measure captures the construct it is intended to measure. We test this distinction using a d…
- arXiv cs.AIresearchPhysical knowledge on historical data matters more than enforcing physical constraints on the forecast
arXiv:2609.19871v1 Announce Type: new Abstract: Time series forecasting has seen signicant advancements with the emergence of new deep learning models. However, forecasting time series in applications involving physical…
- arXiv cs.AIresearchFrom "Who Is This User?" to "What Does This Purchase Mean?": A Deployed Pipeline for Semantic User Profiling at Bank Scale
arXiv:2609.19928v1 Announce Type: new Abstract: Per-user LLM inference on transaction histories binds the inference budget linearly to user count, which becomes prohibitive at applied scale. We re-cast attribute inferenc…
- arXiv cs.AIresearchBeyond Depth Truncation: Controlled Evaluation of Depth Utilization in Recursive Language Models
arXiv:2609.19934v1 Announce Type: new Abstract: Depth-recurrent language models iteratively apply a small layer stack, decoupling per-token compute from distinct parameter count. To determine whether such a model genuine…