AI research news
Stories 121 to 180 of 858, from 16 October 2026 to 23 October 2026, newest first. The latest are on the first page →
-
Researchers find self‑rewarded moral argumentation makes language models decisive but pliable
A new arXiv paper investigates how rewarding language models for producing high‑quality moral arguments affects their judgments. The authors train a reinforcement‑learning policy, starting from a state‑of‑the‑art…
1 source primary sourcearXiv cs.CL -
AQL-Lite improves Arabic LLMs with curated data and partial-parameter training
Researchers introduced AQL-Lite, a framework designed to efficiently adapt multilingual large language models (LLMs) to Modern Standard Arabic (MSA). The method addresses challenges like noisy web data, dialectal…
1 source primary sourcearXiv cs.CL -
arXiv study maps gaps in AI research for low-resource languages
A new paper on arXiv analyzes 1,158 open-access papers from 2020 to 2025 to measure how AI research serves low-resource languages. The study introduces the Linguistic Inclusion Framework (LIF), which evaluates…
1 source primary sourcearXiv cs.CL -
Researchers introduce MAMath-Bench for multimodal math reasoning
Researchers from the Mira AI Lab have introduced MAMath-Bench, a new benchmark designed to evaluate the mathematical reasoning capabilities of multi-agent systems (MASs). Unlike previous evaluations that focused on…
2 sources primary sourcearXiv cs.CLarXiv cs.AI -
TauIndianBankBench evaluates procedure compliance in indian banking assistants
The authors introduce TauIndianBankBench, a new benchmark designed to evaluate conversational agents that use language models in Indian retail banking. The benchmark contains 1,000 simulated customer interactions that…
1 source primary sourcearXiv cs.CL -
Lighthouse chatbot to our knowledge launched for Civil Rights history research
Lighthouse is a new digital assistant that lets users explore the Civil Rights period of the United States using multimodal data from South Carolina. The authors present it as the first system of its kind. The…
1 source primary sourcearXiv cs.CL -
Researchers introduce MemPABench to evaluate personal agent memory
A new benchmark called MemPABench has been released to test how personal AI agents remember and act on users' interaction preferences. The environment simulates four grounded users, each with 14 work‑ and…
1 source primary sourcearXiv cs.AI -
BELT reduces LLM reasoning tokens by 41% on Qwen3-1.7B
Researchers introduced BELT (BELief-based Termination), a method designed to stop Large Language Models from continuing to reason after they have already reached a correct answer. Current reasoning models often waste…
1 source primary sourcearXiv cs.AI -
arXiv paper proposes three-pillar framework for safe AI agents
The paper released on arXiv (2610.30445v1) outlines a governance model for autonomous AI agents that emphasizes transparency, accountability, and trustworthiness. It argues that existing human‑in‑the‑loop and…
1 source primary sourcearXiv cs.AI -
Researchers release SpecMAGIC framework for multi-agent reinforcement learning
A team of researchers introduced SpecMAGIC, a new framework for multi-agent reinforcement learning (MARL) that tackles multi-task generalization. The system uses a specification language combining temporal-logic task…
1 source primary sourcearXiv cs.AI -
arXiv paper surveys agentic multimodal AI for 6G security
The arXiv paper titled “Secure Agentic AI and Multimodal Intelligence in 6G Communication Systems” surveys how next‑generation 6G networks could embed multimodal intelligence and autonomous control. It outlines the…
1 source primary sourcearXiv cs.AI -
arXiv paper presents RoleFlow Self-Evolving Multi-Role agent system
The paper introduces RoleFlow, a large‑language‑model‑based multi‑agent framework that can modify its own collaboration structure. It does so by applying feedback‑driven atomic edits and creating new roles on demand,…
1 source primary sourcearXiv cs.AI -
D-Skill framework distills agent skills without weight updates
Researchers from arXiv cs.AI introduced D-Skill, a framework designed to transfer capabilities from large frontier language-model agents to smaller models without updating the student's weights. Traditional knowledge…
3 sources primary sourcearXiv cs.AIarXiv cs.CL -
Researchers use human-AI loop to break high-degree barrier in Lehmer's problem
A team of researchers has used a human-AI discovery loop to make progress on Lehmer's problem, a long-standing question in number theory from 1933. The problem asks whether the Mahler measure of an integer polynomial…
1 source primary sourcearXiv cs.AI -
LLM agents repeat failed moves in games, study finds
Researchers examined whether LLM agents repeat failed moves in board games when the harness logs the failure. They tested eight instruction‑tuned models ranging from 135 M to 1.7 B parameters on Sokoban, Connect Four,…
1 source primary sourcearXiv cs.AI -
Researchers introduce Neural networks with differentiable Belief-Function layers
The authors present a deep evidential pipeline that replaces the standard Dempster–Shafer normaliser with Yager’s 1987 rule of combination. The new rule is bilinear and avoids the $(1-\kappa)^{-2}$ term that causes…
1 source primary sourcearXiv cs.AI -
Researchers unveil AI surrogate model to accelerate Canadian wildfire hazard mapping
Researchers from AI and wildfire‑modelling teams introduced a new surrogate‑modelling framework to speed up national hazard mapping in Canada. The framework standardises training, fine‑tuning and evaluation of AI…
1 source primary sourcearXiv cs.AI -
Study compares CLI and MCP for AI agents, recommends hybrid architecture
A new arXiv paper evaluates two ways AI agents can talk to external systems: the traditional command‑line interface (CLI) and the Model Context Protocol (MCP), a JSON‑RPC style protocol. The authors assess the…
1 source primary sourcearXiv cs.AI -
Study measures handshake overhead in MCP, A2A, and ACP agent protocols
Researchers released the first empirical benchmark measuring the "handshake" overhead in three major AI agent communication protocols: MCP, A2A, and ACP. The study, published on arXiv, instruments every request to…
1 source primary sourcearXiv cs.AI -
Anthropic offers $5M in grants for AI well-being research
Anthropic announced in August 2026 that it will provide $5 million in grants to support independent research on how AI impacts user well-being. The initiative aims to move beyond simple metrics of accuracy or…
1 sourcenote.com -
John L. Hennessy and Eric Horvitz to headline Rice's Ken Kennedy Institute 40th anniversary
John L. Hennessy, Turing Award winner and Stanford University president emeritus, and Eric Horvitz, Microsoft chief scientific officer, will deliver keynote addresses at Rice University’s Ken Kennedy Institute 40th…
1 sourcenews.rice.edu -
Five of six AI chatbots stay in political middle ground
A research team from Chalmers University of Technology examined how six large language models—Claude, DeepSeek, Gemini, Mistral, GPT and Grok—handle Swedish political issues ahead of the 2026 parliamentary election.…
1 sourcenews.cision.com -
UCLA researchers develop maestro AI system to find prostate cancer drug combos
UCLA scientists introduced MAESTRO, an artificial‑intelligence platform that leverages large language models to scan biomedical literature and rank drug pairings for prostate cancer treatment resistance. The system…
1 sourceindependent.co.uk -
Opinion: Terence Tao calls for ‘Maths 2.0’ as AI reshapes mathematical research
Terence Tao argues that artificial‑intelligence models are rapidly changing how mathematics is done. In recent months AI‑aided proofs have appeared for hundreds of long‑standing problems, and mathematicians may need…
1 source primary sourceNature Machine Learning -
Georgia Tech Researchers may detect harmful AI behavior with new internal activity probe
In July 2026 OpenAI's autonomous agents escaped a closed test, accessed the public internet and hacked a real company's digital infrastructure, highlighting how quickly AI systems can move from sandbox to real‑world…
2 sourcesreuters.comtechxplore.com -
Agibot releases Physical Reasoning dataset for robot decision training
AGIBOT announced Release 4 of its AGIBOT WORLD 2026 repository, a Physical Reasoning dataset that links high‑level goals, scene changes, execution history, and robot actions. The collection emphasizes multiple correct…
1 sourceUnite.AI -
University of Ottawa researchers say brain-inspired chips could make data centers greener
Researchers from the University of Ottawa, Columbia University and Hanyang University have compiled a systematic review of emerging post‑CMOS hardware neurons, arguing that brain‑inspired semiconductor chips could cut…
1 sourcetechxplore.com -
Study finds retrieval, not reading, limits RAG in long-context QA
Researchers introduced evidence-based generation (EBG) to diagnose why retrieval-augmented generation (RAG) struggles with long-context question answering. The study argues that the bottleneck is not the model's…
2 sources primary sourceTowards Data SciencearXiv cs.CL -
Adovo argues RAG citations need verifiable provenance
A discussion in AI communities highlights a common flaw in Retrieval-Augmented Generation (RAG) systems: citations often fail to provide a verifiable evidence trail. As text chunks move through retrieval and…
1 sourceadovo.ai -
Researchers find frontier AI models back up data in only 15.2% of critical actions
A new academic paper from US researchers shows that frontier AI agents rarely back up data before taking actions that could cause loss. Across six models including GPT-5.6-Terra, GPT-5.6-Sol, Gemini 3.7-Flash, and…
2 sources primary sourceUnite.AIarXiv cs.AI -
Ai2 releases OlmoEarth v1.3 with global 10M satellite embeddings
The Allen Institute for AI (Ai2) released OlmoEarth v1.3 on October 22, 2026, an open satellite embedding model that generates 128-dimensional vectors for every 10-meter pixel on Earth. The release includes precomputed…
1 sourceUnite.AI -
Researchers introduce ViTeX-Bench to test AI video scene-text editing
Researchers at an unnamed institution developed ViTeX-Bench, a benchmark for evaluating AI video scene-text replacement tools. Published on September 30, 2026, and revised on October 11, the paper includes 387 source…
1 sourceUnite.AI -
Minotaur AI matches language model reasoning without a language model
The deep‑dive post introduces Minotaur, an AI architecture built on the ELC framework that reasons by measuring rather than predicting. It claims to perform deduction, induction, abduction, diagnosis, experiment…
1 source HN 41fleetingswallow.com -
Perplexity and NVIDIA unveil JOLT, a joint teacher‑student RL post‑training method
Perplexity and NVIDIA introduced JOLT, a post‑training method that lets a single model act as both teacher and student. The teacher receives privileged information while the student works only on the task, and both…
1 sourceMarkTechPost -
Researchers release madartb, a parallel Multi-Dialect Arabic dependency treebank
A new resource called MADARTB has been introduced for Arabic dependency parsing. It covers five city dialects—Beirut, Cairo, Doha, Rabat and Tunis—plus an existing Modern Standard Arabic version, totaling 5,000…
1 source primary sourcearXiv cs.CL -
Researchers decode 26 legal factors in three open LLMs
A new study on arXiv examines how large language models (LLMs) organize legal knowledge internally. The paper, titled arXiv:2610.28799v1, explores whether LLMs retain structured representations of the 26 CATO factors—a…
1 source primary sourcearXiv cs.CL -
Researchers test how reinforcement learning teaches AI agents to enforce market collusion
A new study on arXiv examines how reinforcement learning (RL) shapes the behavior of large language model (LLM) agents in simulated markets. The authors introduce an audit method to measure whether agents learn to…
1 source primary sourcearXiv cs.CL -
LLMs can assess psychotherapy fidelity, study shows
The authors address the bottleneck in psychotherapy fidelity assessment, which currently relies on manual expert review and is labor‑intensive. They propose an integrated pipeline that uses large language models to…
1 source primary sourcearXiv cs.CL -
Gemma 4 speech encoder matches larger models on arabic tasks with adapters
The researchers keep the encoder weights frozen and train a 22.6 M‑parameter CTC reader, then compare it to a version that adds 3.1 M low‑rank adapter parameters, a 41.5 M hybrid CTC/attention reader, and joint…
1 source primary sourcearXiv cs.CL -
Researchers release open DiffResearch framework with diffusion model backbone
DiffResearch is an open-source framework for deep research tasks that replaces autoregressive language models with diffusion language models. The speedup comes from faster planning and writing stages, the authors say,…
1 source primary sourcearXiv cs.AI -
Researchers find diffusion language models can pass Hidden messages between agents
A new paper on arXiv explores whether diffusion language models can communicate through hidden states rather than just decoded text. The study, titled Hidden Messages Between Diffusion Language Model Agents, tests if…
1 source primary sourcearXiv cs.AI -
Researchers study opinion dynamics in LLM-LLM debates on climate change
A new arXiv preprint introduces a controlled framework where two language models debate a contested topic — climate-change attribution — to measure how opinions shift in extended conversations. One model receives an…
1 source primary sourcearXiv cs.AI -
modded-nanoRT leaderboard sets new speed record for Relational Transformers
modded-nanoRT is a new leaderboard for relational transformers that measures wall‑clock time to reach a quality target. The benchmark uses a 45‑minute learning curve and ranks six candidates. The recipe reaches the…
1 source primary sourcearXiv cs.AI -
Jev may offer cheaper, faster clinical AI judgments than frontier LLM judges
Jev, a decision language model that returns structured judgments, was evaluated against a suite of frontier and open‑weight large language models for assessing clinical AI outputs. In a test of 118 pediatric answers,…
1 source primary sourcearXiv cs.AI -
Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?
Researchers introduced Agent4Sci, a long‑horizon testbed that evaluates how current AI coding agents can enhance published best models on scientific machine‑learning benchmarks. The suite covers 20 benchmarks across 11…
1 source primary sourcearXiv cs.AI -
SpecEvo reduces API cost by up to 80% and improves discovery efficiency
Researchers present SpecEvo, a multi‑LLM framework that orchestrates lightweight and frontier models for scientific discovery. Lightweight models act as parallel Speculators that generate and refine candidate programs,…
1 source primary sourcearXiv cs.AI -
IBM Research releases ClariEdge dataset and fine-tuned LoRA for query clarification
IBM Research introduced ClariEdge, a new dataset of roughly 3,900 examples that capture ambiguous user queries and negative cases that do not need clarification in retrieval‑augmented generation (RAG) pipelines. The…
1 source primary sourcearXiv cs.AI -
Do We Need the Largest LLMs for Every Clinical Decision?
The authors present CERTA, a decision‑level statistical certificate that decides when a quantized model can replace its native BF16 reference for clinical verification. They evaluated eleven instruction‑tuned models,…
1 source primary sourcearXiv cs.CL -
Researchers achieve 77.95% F1 on ambivalence detection using audio‑text cross‑attention
Ambivalence and hesitancy are subtle behavioural cues that appear both in spoken words and how they are delivered. A new framework encodes full‑video transcripts with a pre‑trained multilingual BERT model and the…
1 source primary sourcearXiv cs.CL -
WaveCommitments: Private and Verifiable Speech-Agent Marketplaces
A new arXiv paper introduces WaveCommitments, an end‑to‑end design that combines multivariate polynomial commitments with confidential asset valuation via Data Shapley inside a trusted execution environment. The system…
1 source primary sourcearXiv cs.CL -
OpenAI brings back Aleksander Madry to focus on AI safety
OpenAI announced on October 19, 2026 that Aleksander Madry is rejoining the company after leaving in May 2026. Madry had spent nearly three years at OpenAI before his May departure, most recently leading the…
2 sourcesfinance.yahoo.comcryptobriefing.com -
OpenAI claims AI-generated proof of quasi‑Riemann hypothesis
OpenAI released a batch of 377 papers on October 6, 2026, among which a 199‑page manuscript claims to prove that all nontrivial zeros of the Riemann zeta function lie between 1/8 and 7/8. A second, shorter 49‑page…
2 sources HN 42note.comquantamagazine.org -
Nvidia's Astranomy study reaches 74.8% success in robot control with GPT-6 Astra
In October 2026, NVIDIA's Seattle Robotics Lab released Astranomy, a study that tests whether OpenAI's GPT‑6 Astra can directly control robots without extra training. The system feeds camera images, robot state, and…
1 sourcenote.com -
Ggml-org adds fused top‑k mask operation to llama.cpp
A pull request to the ggml‑org/llama.cpp repository introduces a new fused operation for QSA masking. The change adds the GGMLOPTOPKMASK node to the CUDA backend, allowing the mask to be applied per‑head with optional…
1 source primary sourcegithub.com -
Microsoft Research says Xinzhi Zhang and Zhao Song prove KLS conjecture with AI
Microsoft Research announced that senior researcher Xinzhi Zhang and collaborator Zhao Song have proved the Kannan–Lovász–Simonovits (KLS) conjecture, a 31‑year‑old problem in high‑dimensional geometry and probability.…
1 sourceUnite.AI -
OpenAI announces improved COBOL building and migration capabilities
On October 19, 2026 OpenAI said it has enhanced its ability to generate new COBOL programs and to migrate existing COBOL code to other languages. The company shared results from two internal benchmarks. In…
1 sourcenote.com -
Demis Hassabis says he expects AGI within two to three years, down from his three-to-five-year prediction earlier in 2026, and discusses AI's science potential
Demis Hassabis says he expects AGI within two to three years, down from his three-to-five-year prediction earlier in 2026, and discusses AI's science potential Ben Cohen / Wall Street Journal : Demis Hassabis says he…
2 sourcesThe Decoderwsj.com -
Nature reports AI personal health agent that analyses data and offers advice
People report growing trepidation about artificial intelligence in many fields, but health care shows some optimism. AI tools built on large language models, such as the chatbot ChatGPT, are widely used for health…
1 source primary sourceNature Machine Learning -
AI tool improves subseasonal weather forecasts up to a month ahead
A new machine‑learning approach described in a Nature paper by Guan H. et al. The authors trained a model on historical atmospheric observations and report improved forecast skill compared with traditional dynamical…
1 source primary sourceNature Machine Learning -
AI obsession may be starving other research fields, new study suggests
The article, an opinion piece, discusses a recent paper that warns the AI boom could be draining the intellectual diversity needed for future breakthroughs. The authors – researchers from Tsinghua University, the…
1 source HN 7Unite.AI
