New Model Releases
A
Showing 181–210 of 276
-
LLMs as Judges and Examinees for Medical Item Difficulty Prediction
-
Anthropic、独自の“創薬活動”に着手 社員1人のバイオ企業を買収、小児難病の治療薬研究を進めるAnthropic starts own drug discovery, buys one-person biotech firmAnthropic announced on Oct 27 that it is pursuing its own drug discovery, targeting rare pediatric conditions such as muscular dystrophy. It acquired a one-employee company that used Claude to run experiments with 11 outside research institutions, and took over that research.
-
AI音声の詐欺電話、6人に1人が応じる可能性 だまされやすい手口とは 米ハーバード大やMetaが4100人で検証Harvard, Meta study: 1 in 6 may fall for AI voice scam callsResearchers from Harvard University, Meta and others published a paper, 'Evaluating AI Models' Capability to Automate Voice Phishing Attacks', testing how easily people are fooled by AI-driven scam calls. In a study of about 4,100 people, roughly one in six may comply, and the work examines which tactics prove most deceptive.
-
Discord、公開サーバの会話をGoogle検索などの対象に プライバシーへの懸念もDiscord to open public server chats to web search; privacy concerns riseDiscord announced that conversations in some public servers will become searchable via web engines such as Google. AI-generated summaries go public from mid-November, and full post text becomes searchable from mid-December. In Japan the feature is on by default, so users must change settings to opt out, amid privacy and AI-training concerns.
-
PFNの国産フラッグシップモデル「PLaMo 3.1 Prime」発表 オンプレでDeep Research対応 API無料プランも開始PFN launches PLaMo 3.1 Prime, a domestic LLM with on-prem Deep ResearchPreferred Networks has released PLaMo 3.1 Prime, its domestically developed flagship generative AI foundation model. It adds Deep Research that can run on-premises and is said to reduce hallucinations. A free API plan is also available for trying it out.
-
AI for Science expands to foundations and research institutionsAnthropic expands AI for Science to 20 foundations and institutionsAnthropic is expanding its AI for Science program with 20 foundations, institutions and research groups worldwide. The program gives free Claude credits to researchers on high-impact projects and has supported hundreds of projects since May 2025.
-
NVIDIAとAMD、MetaとのAI分野での提携拡大を発表 「Muse」の強化や次世代インフラ構築で協力NVIDIA, AMD expand Meta tie-ups with $3B and $1B investmentsNVIDIA and AMD each announced expanded strategic partnerships with Meta in AI, cooperating on development and operation of the Muse AI agent and on next-generation infrastructure. They also pledged investments in Meta ($3 billion from NVIDIA, $1 billion from AMD), as Meta builds out infrastructure with both in-house chips and external supply.
-
Google DeepMind、ワールドモデル「Genie 4」発表 音声や会話もリアルタイム生成、Geminiと連携Google DeepMind unveils Genie 4 world model with audioGoogle DeepMind announced Genie 4, a world model that generates interactive virtual worlds in real time. It improves visual quality and physics and adds simultaneous ambient sound and speech. It launches as a research preview, rolling out to Google AI Ultra subscribers and others.
-
Honeycomb-Surfels: Compact Hexagonal Primitives for Efficient Scene Rendering and Reconstruction
-
ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning
-
When Should We Intervene at Test Time?
-
Heads of Security: Learned Attention Gates Improve Prompt Injection Robustness
-
GEPA-Curator: Curated Learning for Self Improving Agents
-
Can LLMs Uncover Hidden Physical Laws through Free Exploration?
-
Metacognitive Decoding: Efficient Inference for Diffusion Language Agents
-
Rare, Relevant, and Still Wrong: How Memory Fails LLM Group Decisions
-
VideoLVR: Latent Visual Reasoning over Grounded Video Evidence
-
EnvSched: Environment-Aware Scheduling for Rollout-Efficient Agent Reinforcement Learning
-
MoI-Bench: A Benchmark for Image Generation and Editing on Mobile Devices
-
GAS-PFN: Graph-Aware Spatiotemporal Prior-Data Fitted Networks for Zero-Shot Traffic Forecasting
-
Understanding the limits of latent world models
-
Zero-Shot Multimodal Event Spotting in Soccer Broadcasts
-
TeamGym: A Gym for Team Decision-Making by AI Agents with Distributed Information
-
Gym-Anything-RT: Scaling Real-Time Environments for Vision-Language Agents
-
Pneuma: A Large-Scale Real-World Time Series Benchmark for Commercial Aircraft Predictive Maintenance
-
Kinematic Constraint-Referenced Evaluation for Articulated Object Manipulation
-
When, Where, and What to Correct: Human-Guided Adaptive Residual Correction for Robot Policies
-
GroundControl: Satellite-Conditioned Multi-View Diffusion for Reconstruction-Free Novel View Synthesis
-
GTFormer: A Syntax-Free Graph Network Alternative to Transformers for 8192$\times$ Length Extrapolation in Linear Time