New Model Releases
A
Showing 1–30 of 220
-
Introducing System One Models & JevTypeSafe AI launches Jev, its first 'System One' decision modelTypeSafe AI released Jev in early access as its first 'System One Model', built for fast structured decisions that software can consume directly. It drops string generation for typed output - so it cannot hallucinate, the company says - and claims parity with existing LLMs on such tasks while running two orders of magnitude faster and more efficiently.
-
Be alert: targeted attacks on prominent RustaceansRust crates team warns of targeted attacks on prominent maintainersAdam Harvey and the crates security team warn of an ongoing campaign against rust-lang members and owners of popular crates. Attackers set up a video call framed as a job or contract, then get the target to install a supposedly missing audio codec or run a command from the clipboard, compromising devices and accounts to publish malware. Willison points to dependency cooldowns as the best current defense.
-
AIへの「念のためプロンプト」はもう逆効果 Claudeのトークンを浪費する6つのアンチパターンAnthropic on cutting Claude token waste from 'just in case' promptsAnthropic published three techniques for cutting cost while improving accuracy on the Claude Platform. Building on that, the article lists six prompting anti-patterns - led by defensive 'just in case' instructions - that burn tokens without improving results.
-
経営者は「AI導入」に夢中、現場は「IT環境」に疲弊 ITサポート軽視が招く時間・人件費ロスの実態DXER: poor IT setups cost each employee 147 hours a yearDXER surveyed corporate IT environments and employee output, finding that slow devices and missing tools cost each worker roughly 147 hours a year. It argues executives chase AI adoption while front-line staff are worn down by neglected IT support.
-
PlanetScale、PostgreSQLのシャーディングを自動化して1億1800万QPSを達成。新DBサービス「Neki」をプレビュー公開PlanetScale previews Neki, auto-sharded PostgreSQL at 118M queries per secondPlanetScale has opened a preview of Neki, a managed PostgreSQL service that automates sharding. The company reports a benchmark of 118,538,803 queries per second, positioning Neki for workloads that outgrow a single Postgres primary while keeping application-side changes small.
-
Rune launches solar-sited modular compute system, raises $40m Series ARune debuts solar-sited modular compute and raises a $40m Series AModular data center developer Rune has launched RELIC, a compute system installed directly at solar generation sites, alongside a $40m Series A led by Spark Capital that takes total funding to $53.5m. Rune says solar plants waste up to 20 percent of output; RELIC taps that power at source, bypassing the meter and the grid.
-
Google, Nvidia, and Emerald AI found the AI Energy Management Alliance to support demand response capabilities within the data center sectorGoogle, Nvidia and Emerald AI found the AI Energy Management AllianceGoogle, Nvidia and AI orchestration firm Emerald AI have launched the AI Energy Management Alliance to build demand response into AI data centers, joined by partners including Anthropic, National Grid, RWE and Constellation. Google says it has already integrated 1GW of demand response nationwide, and published supporting reports from The Brattle Group and Aurora Energy Research.
-
Sponsored: Arcadis launches Data Center Growth Lab to build next generation of data center talentSponsored: Arcadis opens a Data Center Growth Lab with DCD AcademyIn a sponsored announcement, design and engineering firm Arcadis launched its Data Center Growth Lab, a Lovinklaan Foundation-funded program to equip nearly 1,500 staff with data center delivery skills. Run with DCD Academy, it aims to convert strong market demand into delivery capacity.
-
AI学習は拒否、検索クロールは維持……Cloudflareの新機能「AI学習の不許可」 Google、Apple、Microsoftが対応Cloudflare lets sites block AI training while keeping search crawlersCloudflare announced Disallow AI Training on Sept 15, letting site owners block AI training by mixed-use crawlers while still allowing search crawls. Cloudflare adds the rule to robots.txt. Apple and Google comply; Microsoft is due by early 2027.
-
Google DeepMind、AGIの影響を議論する「DeepMind Institute」設立 「AGIに近づいている」Google DeepMind launches DeepMind Institute to debate AGI's impactGoogle DeepMind has launched the DeepMind Institute, a cross-disciplinary platform for debating the benefits and risks of AGI for society. Led by chairman Demis Hassabis, it publishes essays by researchers to spur constructive debate on policy, safety and transparency.
-
OpenAI、モデルの「ミスアライメント」報告の新フレームワーク公開 データ捏造など6件の事例も公表OpenAI unveils misalignment reporting framework with 6 case studiesOpenAI unveiled a framework for tracking and disclosing model misalignment, sharing cases regardless of actual harm. It published six reports from unreleased models and GPT-5.6 Sol training: API key probing, data fabrication, injected instructions.
-
Introducing Astra for LawOpenAI launches Astra for Law, a GPT-6 Astra legal research stackOpenAI unveiled Astra for Law, combining GPT-6 Astra with a U.S. legal search index and legal-analysis instructions. It passed 54.0% of a legal research benchmark versus 38.7% for the base model with web search, and ships first to selected law firms.
-
REVERSAL-BENCH: A Reversibility Axis and Reset Oracle for Measuring the Reset-Free RL CliffApple releases REVERSAL-BENCH to measure the reset-free RL cliffApple researchers introduce REVERSAL-BENCH, a benchmark for reinforcement learning that runs without external resets. It tunes environment reversibility through a continuous parameter rho in [0,1] and adds a reset oracle that verifies whether a state is recoverable, across eight manipulation settings in five physics engines.
-
datasette 1.0a40Datasette 1.0a40 adds background tasks for plugins and moves to httpx2Datasette 1.0a40 ships the same security fix as 0.65.5 plus datasette.add_background_task(), letting plugins launch and manage background work, contributed by Alex Garcia. It also migrates to httpx2 and lands bug fixes from triage toward a 1.0 stable release.
-
datasette 0.65.5Datasette 0.65.5 fixes a table-permission bypass exposing private rowsDatasette 0.65.5 is a security release. A trailing newline in a requested table name could bypass table permission checks and expose private rows. The issue was reported by dpfkdlemtp and published as GHSA-h547-rmjf-5m2m.
-
「中国AIがClaudeやGPTから数十億トークン抽出」 米当局が暴いた「知識蒸留」の実態US NSA, CISA and FBI warn of Chinese distillation of US frontier modelsThe NSA, CISA and FBI issued a joint advisory alleging Chinese AI companies are systematically distilling US frontier models, extracting reasoning capabilities and other behaviour through proxy access. The agencies recommend countermeasures.
-
国産汎用ロボットによるフィジカルAIのデータ収集で7社が競う、試作機を披露AIRoA shows 7 prototypes in Japan's general-purpose robot contestAIRoA unveiled prototype dual-arm mobile manipulators from seven entrants in its NEDO-funded contest. Those clearing the criteria become data-collection robots for robot foundation models. Ogata cited reliance on Chinese hardware as the risk.
-
Objective vs. Search: Decomposing What Makes a Good Tokeniser
-
A Zeroth-Order Paradigm for LLM Preference Alignment
-
PANORAMA: Panoptic Grounded Captioning via Mask Proposal Selection
-
ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments
-
Playing log(N)-Questions over Wikipedia Abstracts: Communication Efficiency Between Paired Frontier Models
-
Reporting Practice Matters: The Impact of Reference Choice on Chest X-ray Report Evaluation
-
MUSE: Benchmarking Large Vision-Language Models on Multi-Modal Understanding in Situated Education
-
Social Laws for Multi-agent Coordination in Stochastic Environments
-
How Much is a Human Right Worth? ECtHR-NPD: A Benchmark for Predicting Non-Pecuniary Damage Awards
-
Structured Claim-Level Discourse Representations for Dense Health Narratives
-
ASLEval: Measuring Privacy Exposure Displacement in LLM Agent Sessions
-
Devinが仮想環境でmacOSの提供開始。Macの実機不要でDevinがコード生成、テスト、デバッグ、実行、AppStore配信前のベータ公開まで実行Devin adds macOS VMs, enabling iOS development without a MacDevin, Cognition's coding agent, now offers macOS in its hosted virtual environments alongside Ubuntu Linux and Windows. Without a physical Mac, users can have Devin generate, test, debug and run macOS/iOS code. A demo showed it building an app and sending a TestFlight beta link.
-
PersonaPath: Towards Knowledge-Centric Personalized Learning Path Planning