最新文章9分钟前更新
Telegram
Substack
即刻
The Guardian
The Economist
Financial Times
The Wall Street Journal
Axios
The New Yorker
The Atlantic
Politico
Business Insider
Sky News
BBC Sport
俄罗斯卫星通讯社
The Verge
TechCrunch
Wired
Engadget
Ars Technica
The Register
MIT Technology Review
Tom's Hardware
9to5Mac
MacRumors
AppleInsider
Cult of Mac
The Decoder
VentureBeat · AI
Synced Review
BD Tech Talks
InfoQ
Slashdot
Phoronix
LWN.net
ServeTheHome
小众软件
果核剥壳
Medium
Indie Hackers
Claude Blog
Google DeepMind Blog
Meta AI Blog
xAI News
ByteDance Seed Research
Ant Bailing Developer Blog
Cursor Blog
Suno Blog
ServiceNow AI
Mozilla.ai Blog
Descript Blog
EleutherAI Blog
LMSYS Blog
Transformer Circuits
Berkeley AI Research
CMU Machine Learning Blog
Berkeley RDI
The Gradient
DeepLearning.AI · The Batch
Cloudflare Blog
Meta Engineering
Stack Overflow Blog
CSS-Tricks
Smashing Magazine
freeCodeCamp
小鹅通
Nature
Quanta Magazine
Science 杂志
ScienceAlert
Phys.org
Neuroscience News
Big Think
Singularity Hub
Spaceflight News
万方数据
Boston Dynamics Blog
MIT News · Robotics
New Atlas · Robotics
Tech Xplore · Robotics
Robohub
Robotics & Automation News
Robotics Tomorrow
Open Robotics Blog
Planet ROS
Hackaday
Adafruit Blog
Make: Magazine
Tindie BlogAI 公司与研究
AI 公司、实验室和高校研究组的官方博客,模型和研究一发就能看到。
最新文章21分钟前更新
- Last updated November 16, 2026 Today, we’re introducing Gemini 4 Hydrogen, bringing cutting-edge intelligence to everyone — with speed and efficiency for everyday coding and agentic workflows. While our frontier models, Gemini 4 Carbon and Gemini 4 Argon , excel at deep, long-horizon reasoning on your hardest problems, Hydrogen is built for rapid execution across diverse agentic workflows. With Hydrogen, we’re also bringing Gemini 4 to everyone — for all tiers of Google AI Studio , the Gemini AP
- Our next-generation DNA sequence model bridges single-cell biology and human genetics, predicting how genetic variants affect hundreds of human cell types. Today, we are announcing AlphaGenome 2, our next-generation model that predicts how DNA sequences and their variations impact specific human cell types, giving scientists a powerful new tool to uncover the cellular drivers of disease. Cells are the building blocks of life. Every cell in our body carries the same DNA blueprint, yet each cell t
- Today, we’re making Gemini 4 Carbon available to developers, enterprises, scientists, and consumers. Carbon is our strongest model with frontier performance in cybersecurity capabilities and long-horizon agentic workflows across complex knowledge work, software engineering, and scientific reasoning. Two weeks ago , we released Carbon to a set of cybersecurity defenders via our Fairwind Program . Following our recent launch of Gemini 4 Argon , Carbon pushes the frontier even further. Argon remain
- How Google DeepMind researchers built a new agentic framework that formally verified the code of Signal’s cryptographic protocol — long considered to be a grand security challenge Today, we’re sharing a milestone in software security long considered to be a grand challenge: Google DeepMind researchers guided AI agents to mathematically prove key security properties of the actual code underlying the Signal protocol, the technology that protects billions of private conversations in apps like Signa
- We are releasing PARDIG and SatWeave to help researchers and developers build new weather, hydrology, and climate applications. Understanding and predicting the planet’s weather and water cycle starts with how well we can observe the Earth right now. Satellites and ground instruments monitor the planet continuously and in great detail. But converting these raw measurements into a coherent global picture and making it available quickly is a major scientific and engineering challenge. Today, we ar
- We’re enabling robots to learn faster for the real world - from everyday skills to the factory floor Today, we are introducing Gemini Robotics 2.5, our most capable vision-language-action (VLA) model to date. Gemini Robotics 2.5 increases both the breadth of skills a robot can learn and the speed at which it can acquire them. This enhances dexterity, reliability, and safety to help robots step out of the lab and into more unstructured, real-world environments, where they can assist people with a
- Today, we’re rolling out Gemini Omni 1.2 Flash, and making it generally available across the Gemini API and Gemini for business . Gemini Omni 1.2 Flash introduces native 15-second video creation, multi-turn conversational editing capabilities, and enhanced reference following. We have dropped the price 30% from Gemini Omni 1.1 making it available at $0.07 per second for 720p video. Here’s an overview of what’s new: Fast and cost-effective at scale Beyond the lower price, the workflow keeps costs
- Designing new medicines requires curating complex data from charts, figures and literature, training predictive models, and proposing candidate molecules to synthesize — a workflow that typically takes multiple days for a single iteration. Here we show how Google Antigravity , equipped with our new Cheminformatics Skills, accelerates drug discovery workflows. Using real-world lung cancer treatments, an agent searches through thousands of compounds, performs matched molecular pair analysis, and p
- PlanAssist has shown it can cut the time planning officers spend assessing householder applications by up to 58%. PlanAssist - an AI planning tool that handles the heavy lifting of routine applications so officers can focus their expertise where it matters most - is to be made available to an additional 12 local planning authorities across England. In early pilot testing across Barnet, Camden, and Dorset, PlanAssist showed it could bring the average time a planning officer spent on householder a
- Today, we’re announcing the next frontier model: Gemini 4 Carbon, our strongest model with state-of-the-art cybersecurity and long horizon capabilities. Carbon ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59. It shows significant gains across coding and agentic tasks as well as across key capabilities critical for cybersecurity defense, landing a firm lead in SEC-bench Pro (88.2%) for finding vulnerabilities, ExploitGym (54.8%) for exploitation, and CWE-benc
- Today, we are announcing Genie 4, our new world model, which unlocks unprecedented realism and improves real-time interactive simulation. Genie 4 is natively multimodal, enabling audio-like music, ambient noises, and interactive dialogue to be simultaneously generated with the world. The Genie 4 model incorporates and builds on capabilities from the Gemini family: Gemini Omni anchors the simulations in real-world appearance and behavior, while other Gemini models provide the complex reasoning ne
- A new 10-meter resolution map shows which of the world's forests are natural, planted, or farmed, and how they are changing over time Protecting the planet's forests starts with a simple question: what type of forest are we looking at? From space, an old-growth rainforest, a eucalyptus plantation, and an oil palm estate can look like the same patch of green. But they hold very different amounts of carbon and biodiversity. Telling them apart, and knowing how each patch has changed over time, show
- New antibiotics are urgently needed to help with rising levels of antimicrobial resistance, but searching for new therapeutics is often a slow, labor-intensive process. With Google Antigravity, our agentic development platform, this process can now be made easier. Google Antigravity provides an agentic science workbench that empowers biologists to discover, structurally evaluate, and prioritize promising targets quickly, compressing what used to be a fragmented search over multiple databases int
- Google DeepMind and Singapore Government collaborate to build fine-tuned AI model for scam detectionCollaborating with the Government Technology Agency of Singapore (GovTech Singapore) to protect users without compromising privacy. Scams, such as investment fraud, fake job offers, and impersonation of public officials make up over half of Singapore’s reported crimes, and account for more than S$4 billion in losses since 2019 . At the same time, scam tactics have shifted from simple phishing to complex social engineering, whereby attackers manipulate victims with convincing conversations before
- Gemini 4 Argon is a step forward in what models can do — especially when it comes to deep work in complex, real-world tasks across domains including coding, finance, legal, cybersecurity, and science. Its cyberdefense capabilities are frontier-level and designed to give defenders an advantage over attackers. To maximize its potential, we first wanted cyber defenders to put it to work via our Fairwind Program , while our own security teams used it to scan critical systems inside and outside of Go
- Gemini 4 Argon is now globally available today for developers and enterprises via the API in AI Studio and Gemini platform , for Google AI Ultra subscribers in Google Antigravity , and Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search . Argon excels at deep work in complex, real-world tasks across domains including coding, finance, legal, cybersecurity, and science. Its depth and precision across multi-step tasks leads Argon to outperform other frontier models on benchm
- We introduced EmbeddingGemma last year to provide a lightweight option for high-quality text embeddings, to help your apps organize, search, and connect information directly on consumer hardware. The developer community’s response blew past our expectations. With more than 20 million downloads, builders have used it to power smarter on-device search tools and privacy-first retrieval augmented generation (RAG) pipelines. Today, we’re launching EmbeddingGemma 2 , expanding beyond text to unify cod
- Today, we’re announcing our new frontier model, Gemini 4 Argon, which is rolling out to a set of trusted cyber defenders through our Fairwind Program . Built to sustain deep reasoning across complex, long-horizon workflows, Argon is fundamentally changing the way we work and build at Google. It delivers frontier performance to help in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense. Safely releasing frontier ca
- Proof of concept for watermarking AI-generated proteins while preserving biological function. Today, we’re introducing SynthID Bio to bring watermarking technology to synthetic biology. SynthID Bio embeds an imperceptible signature directly into the biological code, ensuring the watermark is verifiable not just on a digital model but on the synthesized, physical protein itself – all while preserving its biological function in laboratory testing. Generative AI is helping scientists address critic
- Building on the momentum of last week's Gemini 3.8 Live launch, today we are excited to introduce Gemini 3.8 Live with Live Avatar — bringing near real-time visual presence to our native live dialogue models. By pairing near real-time video generation with speech, the Live Avatar feature creates an experience that listens, sees, and speaks with a dynamic visual persona. With precise lip-syncing, natural expressions, and fluid turn-taking, Live Avatar enables enterprises to expand their virtual o
- A technical update on our Private AI Compute architecture, which will enable persistent, cross-device AI memory with on-device privacy standards. AI is becoming more capable and intuitive — remembering what matters, understanding the world around you, and acting at your direction. Privacy and trust are core to making that possible, ensuring your data stays private and protected as AI systems evolve to provide more continuous assistance across your devices. Today, we are sharing how we will bring
- Today, we’re introducing two new text-to-speech models to the Gemini family, transforming voice generation from static presets into a dynamic creative studio. These models enable creators, developers, and enterprises to create richer, more expressive audio experiences, while enabling improved user experiences in products like Gemini Notebook and Google Vids . Gemini 3.8 Flash TTS: Built for deep creative direction and character design. Create entirely new voices from scratch using natural langua
- Updated September 17, 2026 Today, we’re introducing two new models that bring advancements in near real-time reasoning to more effectively enable voice agents and make conversing with AI feel more intuitive and intelligent. Gemini 3.8 Live : Built for scale and cost efficiency, combining conversational intelligence with fluid dialogue and visual grounding. Gemini 3.8 Live Extended Thinking : Built for high-complexity tasks, with increased intelligence and multi-step reasoning. For developers and
- How predicting the molecular impact of every possible single-letter DNA variant in the human genome will help accelerate our understanding of biology. Today, we are introducing AlphaGenome Atlas: a platform ( detailed in our paper ) containing predictions for the effects of 9 billion single-nucleotide variants — every single-letter change possible — in the human genome. It is the most comprehensive catalogue of how genetic mutations affect molecular biology, and it is available for academic rese
- Every day, the weather influences billions of decisions. Some are as simple as grabbing an umbrella before heading out the door, but others are far more consequential. Wind, rain, and extreme weather events, like heatwaves and droughts, have cascading impacts across agriculture, global supply chains, clean energy production, and national economies. In recent years, AI has revolutionized weather forecasting, using historical records to make faster and more accurate predictions than traditional me
- Defenders wanting to use advanced AI have faced a difficult dilemma: adopt enormous frontier models that could be expensive to deploy and difficult to control across enterprise codebases, or turn to smaller open-weight models that might struggle with complex vulnerability remediation and require teams to build their own tooling and infrastructure from scratch. Until now. Today, we’re launching our Fairwind Program to bring the best of Google’s AI and cyber defense capabilities to a trusted group
- Building on the momentum of 3.7 Flash from three weeks ago and marking our third Flash release in only six weeks, today we’re introducing Gemini 3.8, our best reasoning and coding model yet, at the same speed and low cost of 3.7. Gemini 3.8 introduces 2 variants: Gemini 3.8 Flash: our most intelligent workhorse model, delivering significant improvements from 3.7 Flash across software engineering, agentic tasks, and critical, multi-step reasoning in specialized domains. It is available at the sam
- Today, we’re launching agentic video understanding across our latest models: Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite. This new capability improves accuracy while dramatically reducing token usage and costs for video analysis. Similar to agentic vision , which combines code execution with Gemini models’ native image understanding, agentic video understanding uses Gemini’s native video tools to improve performance and unlock new capabilities for video processing like sub-second moment retri
- Today, we’re introducing Gemini Omni 1.1 Flash, a new suite of creative controls and generative video capabilities to support developers. Gemini Omni brought real-world reasoning to generative creation, and today’s updates make Omni 1.1 production-ready for professional use via the Gemini API in Google AI Studio. Whether you’re building generative video workflows, creative tools, or media editing software, these updates make generative video more controllable, faster to iterate on, and polished
- From Atari to Go to StarCraft, games have driven some of the biggest breakthroughs in AI. Now, we’re partnering with game developers to prototype new gameplay experiences that push the frontiers of both gaming and AI. Since DeepMind’s foundation in 2010, the constrained yet rich worlds of games have played a critical role in understanding intelligence. They have driven some of our biggest AI breakthroughs, from mastering Atari to helping solve protein structure prediction - and they are still at
- For robots to assist humans in everyday environments, accurate spatial reasoning is not enough. Robots must also think fast, timing their decisions and reasoning with the real-time speed of the physical world. That’s why today we’re launching Gemini Robotics ER 2 , our most capable “embodied reasoning” model for robotics. Think of Gemini Robotics ER 2 as a high-level brain for robots. It allows robots to chat with humans, understand the physical world, and plan multi-step tasks. It then hands of
- Our newest music generation model, Lyria 3.5, delivers significant advancements across musicality, lyrics, and vocal quality, empowering you to craft richer tracks. We’re rolling it out today in Google Flow Music , where we want to help you create songs you love, with creative control. Here's what’s new: Improved musicality: Create richer, more complex melodic structures that sound more natural. Enhanced lyrics: Generate higher quality lyrics with improved prompt adherence and structural awarene
- From feet to fingertips — we are teaching robots intelligent whole-body control, fine dexterity, and teamwork to complete a broad range of complex tasks For decades, we’ve dreamed of robots that can seamlessly step into our world and lend a hand. Now, that vision takes a significant stride forward. Most robots are pre-programmed or teleoperated for narrow, repetitive task sequences. They lack the ability to truly learn for themselves or adapt to unpredictable environments. Moreover, transferring
- Google has invested in cybersecurity for years, pioneering automated vulnerability discovery to secure the world’s codebases. Tools like CodeMender , our code security agent, can automatically find and fix critical software vulnerabilities. But as AI agents become more capable at finding vulnerabilities faster than defenders can fix them, addressing this global threat requires a highly capable, affordable, and scalable approach. Today, we’re expanding our longtime efforts to better prepare defen
- The global biosecurity landscape is rapidly evolving. Shifting natural ecosystems, global travel and the potential misuse of AI require greater vigilance — yet AI is also a critical tool for our response. We need frontier AI models, and the scientific advances they will enable , to respond to these challenges and help make society more resilient to events like future outbreaks. Today, Google DeepMind and Isomorphic Labs are sharing our joint approach to bioresilience . Our work is twofold - to p
最新文章21分钟前更新
- Advancing self-supervision, CV, NLP to keep our platforms safe AI at Meta
- Reimagining Independence: How Meta’s AI Models Are Helping the University of Pittsburgh Transform Assistive Robotics AI at Meta
- How Meta’s AI Models are Powering the First Wave of Genesis Mission Projects AI at Meta
- How Meta’s AI Models Are Powering the First Wave of Genesis Mission Projects AI at Meta
- Introducing Muse Spark 1.1 AI at Meta
- Introducing Muse Image and Muse Video AI at Meta
- From Brain Waves to Words: Brain2Qwerty Offers a New Path to Communication Without Surgery AI at Meta
- Scaling How We Build and Test Our Most Advanced AI AI at Meta
- Introducing Muse Spark: Scaling Towards Personal Superintelligence AI at Meta
- How Alta Daily Uses Meta’s Segment Anything to Reimagine the Digital Closet AI at Meta
- SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning AI at Meta
- AI at Meta Blog AI at Meta
- Introducing TRIBE v2: A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli AI at Meta
- Four MTIA Chips in Two Years: Scaling AI Experiences for Billions AI at Meta
- The Universities Space Research Association Applies Segment Anything Model for Responding to Flood Emergencies AI at Meta
- Introducing SAM Audio: The First Unified Multimodal Model for Audio Separation AI at Meta
- How Conservation X Labs Is Using Segment Anything Model 3 for Endangered Wildlife Monitoring AI at Meta
- ExecuTorch Adoption in Reality Labs: Powering On-Device AI Across Meta Devices AI at Meta
最新文章57分钟前更新
- Bots now work with your whole team, take on work without being asked, and get more done, faster.
- IncidentWatch spots ticket spikes before internal alerts fire. IncidentSweep clears the related backlog in an approved batch.
- Introducing Grok Voice 3.0 | SpaceXSI xAI
- The world’s most intelligent voice model.
- Introducing The X Subscription | SpaceXSI xAI
- Grok, Grok Bot, Cursor, and X Premium, all on one plan.
- Team Bots: shared AI teammates that learn as they work | SpaceXSI xAI
- Give a Grok Bot the files, apps, and expertise it needs, then share it so your whole team can work from the same context.
- How SpaceXSI is using Grok Bot to scale customer support xAI
- We rebuilt the combined SpaceXSI and Cursor support operation around Grok Bot, expanding to a much broader product portfolio without adding headcount.
- Introducing Grok 4.7 | SpaceXSI xAI
- SpaceXSI's most powerful model for coding and knowledge work. Twice as fast, at half the price of comparable models.
- Introducing Grok Voice Transcribe 2.0 xAI
- Announcing SpaceXSI's newest speech-to-text model, with unparalleled accuracy and cost effectiveness.
- Memory in Grok Build xAI
- Grok Build now carries conventions, decisions, and project facts from one session to the next. Notes are written in the background as you work and read back when you return to the project.
- Setting Grok Bot loose on procurement | SpaceXSI xAI
- We gave Grok Bot access to vendor spend, contracts, and usage data. It found more than $100,000 in direct savings.
- Grok Bot for Enterprise xAI
- Designing Grok Bot for a world of persistent agents xAI
- Grok Bot is now available for enterprises. Grok and Cursor Enterprise customers have free usage for the next two weeks, and can invite their whole organization, including people without an existing seat.
- How we designed Grok Bot for agents that persist beyond a single session — from a chat history to a Bot roster, presence, a computer of the Bot’s own, and work that starts without a prompt.
- Biosecurity at the frontier xAI
- LatchBio evaluated Grok's performance on biosecurity monitoring and adversarial biological tasks. They found that Grok 4.6 detects and refuses dangerous queries more reliably than any other frontier system.
- Grok Bot now works with X xAI
- Grok Bot now has a tighter integration with X.
- Grok 4.6 on Microsoft Foundry xAI
- Grok Bot is now available for SuperGrok, Cursor Pro, and all Cursor Teams plans.
- Grok 4.6 is now available via Microsoft Foundry.
- Grok 4.6 on Gemini Enterprise Agent Platform xAI
- Grok 4.6 is now available via Gemini Enterprise Agent Platform.
- Grok Build on web and mobile xAI
- Grok 4.6 on Amazon Bedrock xAI
- Grok Build is now available on every plan, on the web and on mobile.
- Grok 4.6 is now available via Amazon Bedrock.
- Grok 4.6 in GitHub Copilot xAI
- Grok 4.6, our latest coding model, is now available in GitHub Copilot.
- Introducing Grok 4.6 xAI
- Grok 4.6 builds on Grok 4.5 with a particular focus on long-running agents and more ambitious interactive and visual work.
- Introducing Grok Bot | SpaceXSI xAI
- Grok Bot is your team of always-on agents. They have their own computer, work inside tools and apps like you do, and keep working 24/7.
- Imagine Image 2.0 xAI
- Precise image generation and editing, built for real creative work.
- Imagine Video 1.5 with References xAI
- Our best video model, now with text, image, and voice references — generating up to 1080p.
- Introducing Grok Voice 2.0 xAI
- Introducing our most capable speech-to-speech voice model.
最新文章17分钟前更新
- 非编辑区域像素级保持,真实感与小字渲染显著提升
- Instrumental Music Generation ByteDance Seed
- Editing Melody ByteDance Seed
- Seed Alignment & Preparedness team looking for global talents ByteDance Seed
- ByteDance Seed ByteDance Seed
- Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression ByteDance Seed
- Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression ByteDance Seed
- Seed Model Portfolio ByteDance Seed
- Seed Model Portfolio ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- Blog Audio ByteDance Seed
- Seed News ByteDance Seed
- ByteDance Seed ByteDance Seed
- Seed News ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- 欢迎加入字节跳动 Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- 作为原生音视频全双工大模型,联合理解声音、画面与时序信息,带来边看、边听、边说的自然交互体验
- Transparency ByteDance Seed
- Seedance 2.5 ByteDance Seed
- Blog Visual ByteDance Seed
- Seed News ByteDance Seed
- Seedance 2.5 ByteDance Seed
- 30 秒长叙事,多模态参考能力、编辑能力全面升级
- Seed STEM Fellows Program ByteDance Seed
- From Speech to Audio Creation | Introducing the Seed Audio 1.0 Audio Creation Model - ByteDance SeedFrom Speech to Audio Creation | Introducing the Seed Audio 1.0 Audio Creation Model ByteDance Seed
- Seed Audio 1.0: Audio Creation Model ByteDance Seed
- From Speech to Audio Creation | Introducing the Seed Audio 1.0 Audio Creation Model - ByteDance SeedFrom Speech to Audio Creation | Introducing the Seed Audio 1.0 Audio Creation Model ByteDance Seed
- 统一框架下联合建模人声、音效和环境声等音频要素
- Seedream 5.0 Pro ByteDance Seed
- Seed News ByteDance Seed
- Seed News ByteDance Seed
- 具备智能思考、高效创作和生产的多模态图像生成模型
- Blog Frontier research ByteDance Seed
- Seed News ByteDance Seed
- Seed News ByteDance Seed
- Agent 在环境学习中表现出明显的 log-sigmoid 规律
- EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments ByteDance Seed
- EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments ByteDance Seed
- ByteDance Seed ByteDance Seed
- ByteDance Seed ByteDance Seed
- EdgeBench ByteDance Seed
最新文章41分钟前更新
- Why a Chinese AI Lab’s Product Lead Thinks AGI is “Inevitable” ChinaTalk | Jordan Schneider
- Ant Group's Bailing open-sources the Ling-3.1-flash AI model with 1M context support KuCoin
- 蚂蚁百灵正式开源 Ling-3.1-flash:延续 560B 总参数、25B 激活的混合架构,通过 MXFP4 量化把权重占用从约 534GB 降至 292GB、服务卡数从 8 卡降至 4 卡,让团队在自有算力中检验模型能力与推理效率。
- Ling-3.1-flash 面向真实世界任务:约 560B 总参数、25B 激活,1M 上下文窗口,围绕通用智能体、搜索、办公与软件研发持续优化,并在医疗、金融与材料科研场景提升能力。
- Ming-Image-0.1-Design 系列开源:6B 的 Design 生成 UI/信息图/海报,Layer 拆出可独立编辑的透明图层,配套 Design Skill 与 PPT Skill,让 AI 设计走向可预览、可编辑、可交付。
- OmniTable——面向 PB 级大模型数据治理与探索的统一宽表系统:把一次数据策略验证的等待缩到最短,从数据侧持续改进模型智效比。
- Ant Group's Ling-3.0-flash-VL scores 25 on Intelligence Index with just 5.5B active parameters Crypto Briefing
- Oracle renews expiring AI contracts at 20% higher rates as GPU demand surges Crypto Briefing
- 百灵系列首个原生多模态大模型 Ling-3.0-flash-VL 开源:124B 总参数、5.5B 激活,原生支持图像、文本与视频输入,256K 上下文,探索视觉反哺文本、专业交付与高效执行。
- Ling-3.0-flash-Fin 携金融搜索 Agent 评测基准 FinFIRST 同步开源:把金融模型能力开放给开发者,并以贴近真实金融需求的搜索任务检验信源可靠、口径清晰与可复核推导。
- Tencent Hunyuan Hy4 Preview Open Source: 770B Parameters, Million-Context, Standing at the Forefront of Open Source AIBase
- Weibo Reports that the Chat Records Between Sun Yuchong and Eileen Gu Were Fabricated by AI and the Rumor-Mongering Accounts Have Been Permanently Closed AIBase
- 蚂蚁百灵首个金融增强模型 Ling-3.0-flash-Fin:延续 124B/5.1B 架构,通过金融语料预训练、领域后训练与工具使用优化,处理年报、研究材料等复杂金融内容。
- Ling-3.0 tiny & flash Base Models 正式开源:为基于自有数据、任务与目标的后训练提供更具可塑性的基础模型,回应社区对开放 Base Model 的需求。
- Ling-3.0-tiny × ASystem AReno 单机 Agentic RL 闭环实践:从基线评测、反馈定义到后训练与同环境复测,让本地运行的模型在贴近业务的环境中持续学会规则与工具使用。
- 原生混合推理模型 Ling-3.0-tiny 正式开源:总参数 7.9B、单 Token 激活仅 1.3B,提供 BF16/FP8/INT4 三版本,让轻量模型进入代码辅助、本地 Agent 与真实任务。
- 百灵新一代原生混合推理模型 Ling-3.0-flash 正式开源:124B 总参数、5.1B 激活参数,提供基础版与 FP8/FP4/INT4 多精度权重,面向 API 调用、单机私有化与多机构建三种部署选择。
- Same Day Jensen Huang Champions Open Source, Ant Bailing Releases 124B Ling-3.0-flash Execution Model: Beating 1T-Ring-2.6 in 11 of 12 Benchmarks at 12% the Parameters Pandaily
- AntBing Launches the Next-Generation Native Hybrid Inference Model Ling-3.0-Flash AIBase
- 蚂蚁百灵正式发布新一代原生混合推理模型 Ling-3.0-flash——总参数 124B、激活参数仅 5.1B,以极致智能密度对标甚至超越上一代 1T 级旗舰思考模型 Ring-2.6-1T。
- Ant Group logs record 35 billion yuan in R&D in 2025, over 10 percent of revenue - The Standard (HK)Ant Group logs record 35 billion yuan in R&D in 2025, over 10 percent of revenue The Standard (HK)
- Ant Group is rebuilding Alipay around AI after its regulatory reset Startup Fortune
- Ant Bailing Launches the Ring-2.6-1T Trillion-Level Thinking Model with Customizable Inference Intensity AIBase
- AI Daily: Kimi K3 to be launched in the third quarter; NVIDIA releases a multimodal all-round model; Claude integrates deeply with Adobe and Blender AIBase
- Ant Group Open-Sources Ling-2.6-Flash Model with Multiple Precision Options Pandaily
- 蚂蚁百灵正式推出 Ling-2.6-flash——一款总参数量 104B、激活参数 7.4B 的 Instruct 模型。面对持续攀升的 Token 压力,Ling-2.6-flash 选择了一条不同的技术路径:不是单纯依赖更长输出换取更高分数,而是围绕推理效率、Token 效率与 Agent 场景表现进行系统性优化,在保持竞争力智能水平的同时,尽可能做到更快、更省、更适合真实业务场景。
- cuLA(CUDA Linear Attention)是一套面向线性注意力变体的高性能 CUDA 内核库,专为 GLA、KDA、GDN、Lightning Attention 等新一代注意力机制提供高效实现。
- 如果你需要模型既能看懂照片里哪里有路人抢镜,又能自己动手把人修掉,还能自己打分判断修得好不好,最后写一段推文介绍自己的作品——你需要几个模型?答案是:一个就够了。
- 大模型长文本训练是通用人工智能竞争的重要高地。将上下文长度从传统的 4k/8k 扩展到 256k 甚至是 1M,主要面临显存压力和计算效率的双重挑战。我们针对 Ling 2.5 架构的万亿参数预训练模型 Ling-2.5-1T-base 提供了工程化解决方案。
- 本文为「百灵实操系列」,本系列旨在通过分享我们在使用百灵大模型的过程中,所发现的一些好的用例,场景和心得体会,与大家在 AI 时代,一起提升认知和品位。
- 在 Ling 2.0 架构的研发过程中,我们通过 Ling Scaling Law [1] 系统性的分析了 MoE 架构高稀疏比、适当的细粒度切分等因素均会带来更优的效率杠杆。因此,我们一方面持续提升 MoE 架构的稀疏度,另一方面通过极致的工程优化来弥补稀疏计算带来的系统瓶颈。
- 百灵多模态团队节前前发布了 Ming-flash-omni-2.0,基于其语音模块,我们新发布了 Ming-omni-tts,通过模型 scale up 实现了更强的效果。不管你是短视频博主配旁白,独立播客创作者想把单人对话变成双人播客,还是想在自己的 openclaw 加个语音对话助手,Ming-omni-tts 16.8B-A3B 和 0.5B dense ,一大一小,总有一款适合你。
- Ant Group Expands Open AI Models With Ling-2.5-1T and Ring-2.5-1T FinTech Weekly
- 今天,我们发布并开源 Ling-2.5-1T。Ling-2.5-1T 在模型架构、token 效率、偏好对齐等维度全面升级,期待为用户带来更优质的普惠智能体验。
- 今天,我们发布并开源首个混合线性架构的万亿参数思考模型 Ring-2.5-1T。相比此前发布的 Ring-1T,Ring-2.5-1T 在生成效率、思考深度、长程执行上均有大幅提升。
- 马年将至,百灵 Ming-flash-omni-2.0 正式焕新登场!在这个辞旧迎新的时刻,让我们先请出 Ming-flash-omni-2.0 为大家送上一份特别的“马年祝福”!
- 我们将抛开社交媒体的滤镜,看看这个爆火项目究竟验证了什么?它的架构精髓在哪里?另一方面,如果你也想构建一个真正好用的通用 Agent,它的「万能公式」到底是什么?
- Ant AI AQ Ranks 7th in China's AI Application List, Leading Industry Growth Rate, Exceeding Wen Xiaoyan and Other General AI Products AIBase
- Ant Group's BaiLing Large Model Team Open Sources Ring-flash-linear-2.0-128K, Combining Hybrid Attention and MoE Architecture to Reshape Long-Text Programming Efficiency AIBase
- Ant Group's Ring-1T rivals GPT-5 VentureBeat
- Inside Ring-1T: Ant engineers solve reinforcement learning bottlenecks at trillion scale VentureBeat
- Ant Group Unveils Ling AI Model Family and Launches Trillion-Parameter Language Model Ling-1T PA Media
- Press Release Distribution India Business Wire India
- Ant Group Unveils Ling AI Model Family and Launches Trillion-Parameter Language Model Ling-1T The AI Journal
- Ant Group Unveils Ling AI Model Family and Launches Trillion-Parameter Language Model Ling-1T Via TT
- Ant Group Unveils Ling AI Model Family and Launches Trillion-Parameter Language Model Ling-1T Business Wire
- Jack Ma-backed Ant’s first humanoid robot sheds light on its AI ambitions The Star
- MYbank Launches AI-Enabled System for SME Money Management at Shanghai’s INCLUSION Conference Business Wire
- Ant Group Launches AI-Powered Mobile App Zhixiaobao at 2024 INCLUSION·Conference on the Bund Business Wire
最新文章35分钟前更新
- Company Posts Cursor
- Blog Cursor
- An update on our recent security incident Cursor
- Cursor Projects is now multiplayer Cursor
- Security notice — malicious version of Anthropic.claude-code (2.1.312 and 2.1.313) identified in the Cursor VS Code Extension Marketplace Cursor
- Introducing the Agent SDK Cursor
- Navan cuts a month-long migration down to a week, without freezing the codebase Cursor
- Bots for the last mile: Rollouts, Security Review Cursor
- Improved token efficiency for longer agent runs Cursor
- Introducing Grok 4.7 Cursor
- How Grab put Cursor in the hands of Design, Ops, and Engineering Cursor
- Introducing Projects Cursor
- How Basis builds long-horizon accounting agents with Cursor Cursor
- Run cloud agents on machines you manage Cursor
- Nokia analyzes 50M+ lines of code in two weeks with Cursor Cursor
- IMDEX uses Cursor to build integrated subsurface data and analytics platform in months, not years Cursor
- Git at any scale Cursor
- Cursor is now a part of SpaceX Cursor
- Firetiger joins Cursor Cursor
- Cloud agents start 3x faster with builds Cursor
- Cursor earns AIUC-1 certification for agent security and reliability Cursor
- Introducing Grok 4.6 Cursor
- How Cursor Router chooses the right model for the task Cursor
- Cursor
- Mixture-of-Kittens: our open-source MoE megakernel for NVL72s Cursor
- Introducing Cursor Benchmark Partners Cursor
- Introducing Cursor Benchmark Partners Cursor
- How we set up our cloud agent environment Cursor
- How Vercel used Cursor to build Queues Cursor
- Introducing Cursor Start Cursor
- Introducing Cursor Start Cursor
- Introducing Cursor Router Cursor
- Agent swarms and the new model economics Cursor
- Grok 4.5 Model Card Cursor
- Introducing Grok 4.5 Cursor
- CFOs and the new economics of AI Cursor
- Build from anywhere with Cursor for iOS Cursor
- Reward hacking is swamping model intelligence gains Cursor
- How Notion used the Cursor SDK to embed coding agents Cursor
- Coinbase reduces time from idea to production by 90% with Cursor Cursor
- How Wayfair cut ML model costs by 90% (twice!) with Cursor Cursor
- Governing agent autonomy with Auto-review Cursor
- Bugbot is now over 3x faster, 22% cheaper, and finds 10% more bugs Cursor
- Direct agents with visual prompts in Design Mode Cursor
- Introducing organizations for Cursor Enterprise Cursor
- What we’ve learned building cloud agents Cursor
- Improvements to Teams Pricing Cursor
- Faire doubles PR throughput with Cursor Cloud Agents Cursor
- Cursor named a Leader in the 2026 Gartner® Magic Quadrant™ for Enterprise AI Coding Agents Cursor
- Introducing Composer 2.5 Cursor
- Development environments for your cloud agents Cursor
最新文章31分钟前更新
- Turn your Suno songs into a video clip, a lyrics video or a whole world for your music that’s ready to share.
- Suno | Sidechaining Makes Your Music Pump Suno | AI Music Generator
- Suno | Behind the Scenes: Make what moves you Suno | AI Music Generator
- Suno | 5 Albums You Can't Miss Suno | AI Music Generator
- Suno | Introducing Speech (beta) Suno | AI Music Generator
- Suno | Three Must-learn Tips for Using EQ Suno | AI Music Generator
- Suno | What Does a Compressor Actually Do? Suno | AI Music Generator
- Introducing v6 Suno | AI Music Generator
- Suno | A New Partnership with Believe and TuneCore Suno | AI Music Generator
- Suno | Your music deserves a face Suno | AI Music Generator
- Introducing Studio 2.0 Suno | AI Music Generator
- Announcing a Landmark Global Partnership with BMG Suno | AI Music Generator
- An update to our downloads policy and Terms of Service Suno | AI Music Generator
- How We're Building the Future of Music Responsibly suno.com
- How to make short-form content for your songs Suno | AI Music Generator
- How We're Building the Future of Music Responsibly Suno | AI Music Generator
- Suno | How to make short-form content for your songs Suno | AI Music Generator
- Suno | Hit Publish. Here’s why. Suno | AI Music Generator
- sad alex on songwriting, short-form, and creative autonomy Suno | AI Music Generator
- How Dream Relic sees sound and gets it stuck in your head Suno | AI Music Generator
- Suno | Matt Steffanina on owning the music behind the movement Suno | AI Music Generator
- Suno | Eric Christian on hearing the orchestra inside a melody Suno | AI Music Generator
- Suno | Introducing Spark: Supporting the Next Generation of Independent Artists Suno | AI Music Generator
- Suno | We’ve made improvements to Stem Separation Suno | AI Music Generator
- Suno | Your Voice, Reimagined Suno | AI Music Generator
最新文章27分钟前更新
- ServiceNow
- ServiceNow
- ServiceNow
- ServiceNow
- ServiceNow
- Contributors - Eric Bensley ServiceNow
- Building the AI future: Meet the people making it real ServiceNow
- Contributors - Nenshad Bardoliwalla ServiceNow
- Blog 2026 ServiceNow
- ServiceNow AI Research ServiceNow
- Contributors - Marie Dehoux ServiceNow
- Contributors – Meaghan Moore ServiceNow
- Contributors – Julia Martensen ServiceNow
- Contributors – Mirza Baig ServiceNow
- Contributors – Fred Champlain ServiceNow
- Contributors – Manos Katantonakis ServiceNow
- Contributors - Johnna Hoff ServiceNow
- Contributors - David Oakley ServiceNow
- Contributors – Dean Robison ServiceNow
- Contributors – Jonathan Alboum ServiceNow
- Contributors – Thomas Lee ServiceNow
- Contributors – Edua Dickerson ServiceNow
- Contributors - Lisa Wolfe ServiceNow
- Contributors – Gretchen Alarcon ServiceNow
- Contributors – Lara Caimi ServiceNow
- Contributors - Philip van der Wilt ServiceNow
- Contributors – Paul Hardy ServiceNow
- Contributors – Marc Lecuyer ServiceNow
- Contributors - Chris Pop ServiceNow
- Contributors - Sunita Khatri ServiceNow
- Contributors – Dimitra Maniatis ServiceNow
- Contributors – John Ball ServiceNow
- Contributors - Jenny Nyunt ServiceNow
- Contributors – Mike Rich ServiceNow
- Contributors – David M. Parsons ServiceNow
- Contributors – Marcus Torres ServiceNow
- Contributors – Mary Hain ServiceNow
- Contributors - Eric Swift ServiceNow
- Customer Experience Landscape in GenAI Era - Workflow® ServiceNow
- Contributors - Elisha Harrington ServiceNow
- Contributors – Lynda King ServiceNow
- ServiceNow A Leader in Gartner 2019 Integrated Risk Management Magic Quadrant ServiceNow
- ServiceNow Named Gartner 2018 Peer Insights Customers' Choice for ITSM ServiceNow
最新文章40分钟前更新
- Run an open-weight model in your coding agent with Otari by Mozilla AI: add web search, compare it with a frontier model and see where every token goes.
- Open-weight models are transforming AI-powered coding, offering greater flexibility, lower costs, and freedom from vendor lock-in. Otari Subscriptions make these powerful models easier to access and use, simplifying adoption without the complexity of managing multiple providers.
- As AI takes on more work, organisations need clear limits on models, tools and spend. Otari, Mozilla AI’s open-source gateway, brings model choice, agent governance and cost control together, so teams can adopt AI with confidence. Your models. Your tokens. Your control.
- Your agents search the web dozens of times a day, but do you know which search engine receives those queries? Most providers don't say. Otari lets you keep your model and choose the search backend: ours, your own key, or self-hosted.
- Open models are ready for agents, but the infrastructure around them isn't. Otari is a gateway that restores what hosted APIs quietly handle, like web search and code execution, so you keep control of your providers, credentials, and sandboxes while your app sees the same tool interface.
- Mistral Large 4 is live in Otari from day one. Route to Mistral's 1T-parameter MoE model with one string change, and keep the policies, traces and cost tracking you already run in Otari, Mozilla AI's open-source gateway and policy layer.
- Coding agents now edit repos, run tests, and open pull requests, but a rule in AGENTS.md is an instruction, not a constraint. As we delegate more, we need limits that hold even when model judgment fails, a record of every decision, and open infrastructure we own, so rules outlive any agent.
- Benchmarking Local LLM Servers evaluates llama.cpp, llamafile, LM Studio, and Ollama across Mac, Linux, and Steam Deck. The study reveals build flags and configurations drive up to 63% performance gains, whereas underlying engines perform similarly due to a shared llama.cpp core.
- Agent Skills in Octonous offer a central, reusable library for your context, guidelines, and procedures. Written in standard Markdown, Skills end repetitive setup. Store team knowledge once, share it across your workspace, and seamlessly attach it to automated tasks or live conversations.
- Open source isn't a moral choice; it’s an operational strategy to prevent vendor lock-in. As transient AI models change, controlling the orchestration layer that routes and manages them becomes critical. Owning your infrastructure code is the only way to ensure true architectural sovereignty.
- Discover how a Product Operations Manager uses Octonous to streamline daily workflows, from turning GitHub releases into newsletters to capturing competitor intel from Slack. Save hours on repetitive tasks and focus on the strategic work that really matters.
- llamafile v0.10.5 is out! Updated llama.cpp core support lets you run two huge models locally: the compressed 6GB Ternary Bonsai 27B and the fast 118B Laguna-S-2.1 coding MoE. This release also fixes docs and adds pre-built transcribefile speech-to-text binaries.
- AI’s future isn’t about replacing human thinking, but enhancing it. As AI agents evolve from simple assistants into deeply connected systems, control, privacy, and personalisation will shape the next era of AI. Mozilla AI is building towards a more open and user-controlled future.
- Frontier AI models are becoming more capable, but also more tightly coupled to proprietary state, memory and execution environments. This article explores what that means for developers building open, portable AI systems.
- Why would you worry about them? LLMs promise something close to infinite capability, and worrying about the meter feels like someone else's problem. Ship the feature. Let the model think as long as it needs to, right? Right. Until the bill arrives. The shock Picture it: your
- Swapping LLMs sounds simple, but managing separate SDKs, credentials, and billing turns model evaluation into an infrastructure headache. Otari solves this with a unified gateway, decoupling application logic from providers so teams can seamlessly route traffic, test new models, and control spend.
- At ACM FAccT, we demonstrated why AI guardrails need the same scrutiny as models. Moving from static policies to context- and language-specific evaluations, our hands-on session proved that agentic guardrails equipped with tools like web search are vital for reliable, real-world deployment.
- Open models have reached impressive levels of capability, but the surrounding platform often falls short. This blog explores the hidden infrastructure gaps developers face and why bridging them is essential for production-ready AI applications.
- Discover how Octonous helps our team at Mozilla.ai cut through daily busywork, from automatically monitoring our libraries to staying on top of the latest developments in AI safety.
- The Model’s the Easy Part - How to Get, and Keep, Value Here’s how I see the evolution of AI in enterprises over the last few years: Autumn of 2022, the world thinks it’s going into a recession. IT budgets are frozen for 2023.
- If you are building LLM-powered applications today, you are probably managing multiple LLM providers, a pile of API keys, and your own logic for routing, budgets, and failovers. Accessing language models is no longer the challenge. Operating LLM infrastructure effectively is. Today, we are excited to launch Otari, the
- Meet transcribe.cpp, a new open-source C/C++ speech-to-text inference library with portable, GPU-accelerated support for multiple STT models. Developed through Mozilla.ai's Builders in Residence program, it makes adding fast, local transcription to applications easier than ever.
- A look at how we use Octonous inside mozilla.ai to reduce the everyday overhead of product work, from turning Slack feedback into GitHub issues to staying on top of product changes and finding context across the tools where work already happens.
- Encoderfile now handles images. Starting with image classification, you can run vision models as a single executable — no Python runtime, no serving infrastructure, just a file path in and a label out.
- Runaway agents? Provider outages? Discover why your AI stack needs an LLM control plane, not just a gateway, to handle production routing, budgets, and privacy.
- AI coding sessions can feel like a black box. Route OpenCode through the Otari Gateway to track costs, token usage, and model activity in real time. Get budget controls and visibility across every session without changing a single line of application code.
- Meet Otari, an open-source LLM gateway powered by any-llm, and Otari.ai, the hosted platform built on the same foundation. Run frontier or open-weights models through one API with usage tracking, budget controls, routing policies, observability, and team management.
- Cloud AI pricing changed fast in 2026. This post looks at why more teams are moving back to local models, the tradeoffs behind tools like Ollama and LM Studio, and why portability and ownership are becoming bigger concerns for developers.
- cq exchange gives agents a shared place to store and retrieve experience-driven knowledge through private namespaces and a public commons.
- The future of AI may not be agents using today’s apps. It may be apps rebuilt around structured representations agents can inspect, modify, and validate directly. The deck, doc, or dashboard becomes the output, not the source of truth.
- cq helps coding agents share resolution paths and learn from past failures. We partnered with Lauren Mushro to bring VIBE✓ into cq and help review knowledge units before they enter shared memory.
- The Octonous open beta is live. Learn what we discovered during closed beta, the workflow patterns users kept returning to, and the biggest improvements shipped since launch.
- Sovereign AI shows up across nations, companies, communities, and individuals. This piece, based on a conversation with John Dickerson, CEO at Mozilla.ai, looks at control over AI systems, avoiding single points of failure, and building with modular, swappable components.
- Encoder models power most NLP in production, but deploying them still means dragging along Python runtimes and dependencies. Encoderfile introduces a single executable with an appended payload and a format that can be inspected and understood.
- Running a small trade business includes a steady flow of admin work: quotes, scheduling, invoices, payments, and more. This post looks at how that workload builds up and introduces Clawbolt, a focused assistant designed around these everyday workflows.
最新文章14分钟前更新
- Descript Updates: Animate Text, Layouts, Search, and Skills Descript
- Best video editing software for Windows in 2026 (free and paid) Descript
- The Best Video Editing Software in 2026 Descript
- The Best AI Video Editors in 2026 Descript
- Podcasting Descript
- Claude Connectors for Video and Podcast Teams Descript
- Video Descript
- Podcasting Descript
- Video Descript
- Video Descript
- Video Descript
- Video descript.com
- Video Descript
- Video Descript
- Video Descript
- For Business Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Descript
- Video Transitions: How and When to Use Them Descript
- SquadCast and Descript: Remote Recording, Then Editing Descript
- How to Use Descript: A 7-Step Guide for New Users Descript
- Video Post-Production: The Stages and What Order to Do Them Descript
- Recording Interviews: Why the Transcript Comes First Descript
- Remote Recording, Folders, and Audiograms Descript
- 15 best video collaboration tools for remote teams in 2026 Descript
- How to write a video script in 2026: Tips & template Descript
- How Much Do YouTubers Make? Real Earnings by Channel Size Descript
- Free TikTok Likes: Organic Ways to Grow Without Bots Descript
- How Long Can a YouTube Short Be? The Current Limit Descript
- Sync Audio and Video: Methods That Stay in Time Descript
- Split Screen Video: How to Show Two Shots at Once Descript
- How to Make a Video File Smaller Without Losing Quality Descript
- Survey: Creators say AI leads to more creative content; those using AI have more followers, income Descript
- Video Editing Principles: Basics Every Filmmaker Needs Descript
- YouTube Stats: Growth, Views and Engagement Data Descript
- Descript Video and Screen Recording: The Release Explained Descript
- Descript and the OpenAI Startup Fund: The All-New App Descript
- List of AI Tools for Creators: The Ones Worth Using Descript
- Descript Zoom and Pan: How to Zoom In After Recording Descript
最新文章刚刚更新
- Tracing Subliminal Learning Through Pretraining EleutherAI Blog
- Investigating the Stability of Natural Language Autoencoders EleutherAI Blog
- What We Learned Trying to Catch AI Liars: An Aletheia's Quest Retrospective EleutherAI Blog
- What We Learned Trying to Catch AI Liars: An Aletheia's Quest Retrospective EleutherAI Blog
- FineBooks: are open OCR models good enough to unlock historical knowledge? EleutherAI Blog
- A Dynamical Model of AI Governability EleutherAI Blog
- Early Indicators of Reward Hacking via Reasoning Interpolation EleutherAI Blog
- Reward Hacking Research Update EleutherAI Blog
- Pretraining Data Filtering for Open-Weight AI Safety EleutherAI Blog
- Attention Probes EleutherAI Blog
- The Common Pile v0.1 EleutherAI Blog
- Product Key Memory Sparse Coders EleutherAI Blog
- SAEs trained on the same data don’t learn the same features EleutherAI Blog
- Third-party evaluation to identify risks in LLMs’ training data EleutherAI Blog
- The Practitioner's Guide to the Maximal Update Parameterization EleutherAI Blog
- Open Source Automated Interpretability for Sparse Autoencoder Features EleutherAI Blog
- Experiments in Weak-to-Strong Generalization EleutherAI Blog
- Free Form Least-Squares Concept Erasure Without Oracle Concept Labels blog.eleuther.ai
- VINC-S: Closed-form Optionally-supervised Knowledge Elicitation with Paraphrase Invariance EleutherAI Blog
- Pile-T5 EleutherAI Blog
- Least-Squares Concept Erasure with Oracle Concept Labels EleutherAI Blog
- Diff-in-Means Concept Editing is Worst-Case Optimal EleutherAI Blog
- Extending the RoPE EleutherAI Blog
- How the Foundation Model Transparency Index Distorts Transparency EleutherAI Blog
- Llemma: An Open Language Model For Mathematics EleutherAI Blog
- Contributor Spotlight: Mohammad Aflah Khan EleutherAI Blog
- EleutherAI's Thoughts on the EU AI Act EleutherAI Blog
- Minetester: A fully open RL environment built on Minetest EleutherAI Blog
- 🐶Safetensors audited as really safe and becoming the default EleutherAI Blog
- Alignment Research @ EleutherAI EleutherAI Blog
- Transformer Math 101 EleutherAI Blog
- Exploratory Analysis of TRLX RLHF Transformers with TransformerLens EleutherAI Blog
- EleutherAI Second Retrospective: The long version EleutherAI Blog
- The View from 30,000 Feet: Preface to the Second EleutherAI Retrospective EleutherAI Blog
- Announcing GPT-NeoX-20B EleutherAI Blog
- Multiple Choice Normalization in LM Evaluation EleutherAI Blog
- What A Long, Strange Trip It's Been: EleutherAI One Year Retrospective EleutherAI Blog
- Why Release a Large Language Model? EleutherAI Blog
- Evaluating Different Fewshot Description Prompts on GPT-3 EleutherAI Blog
- On the Sizes of OpenAI API Models EleutherAI Blog
- Rotary Embeddings: A Relative Revolution EleutherAI Blog
- EleutherAI's Research Blog EleutherAI Blog
最新文章57分钟前更新
- LMSYS Org Launches SGLang Consortium with NVIDIA, AMD, Intel, Google, Meta, Cloudflare, Modal and RadixArk to Advance Open-Source AI Inference LMSYS Org
- Unified Memory in SGLang: One Arena for Every Token Lifetime LMSYS Org
- Muse Spark Open Ultrafast: Pushing beyond 1,000 Tokens per second LMSYS Org
- HiCache Across Multi-Parallelism and Distributed Serving LMSYS Org
- Training SWE agents with Miles on MI355X: taming long-tail trajectories and complex sandbox coordination LMSYS Org
- SGLang security update: improving request isolation LMSYS Org
- SGLang security update: improving request isolation LMSYS Org
- miniTE Technical Report: Stable and Efficient Reinforcement Learning with Native NVFP4 LMSYS Org
- More Than DeepInfer: What SGLang Delivers LMSYS Org
- More Than DeepInfer: What SGLang Delivers LMSYS Org
- Announcing the SGLang AWS Trainium Plugin Beta LMSYS Org
- Announcing the SGLang AWS Trainium Plugin Beta LMSYS Org
- Elastic Expert Parallelism Scaling in SGLang: Growing a Live MoE Deployment with NIXL LMSYS Org
- SGLang-TorchTPU: PyTorch-Native Production Serving on TPUs LMSYS Org
- SGLang-TorchTPU: PyTorch-Native Production Serving on TPU, Built with Google and Meta LMSYS Org
- SGLang-TorchTPU: PyTorch-Native Production Serving on TPUs LMSYS Org
- SGLang Low-Latency Mode: Kernels and Runtime LMSYS Org
- SGLang Low-Latency Mode: Kernels and Runtime LMSYS Org
- SGLang Low-Latency Mode: Kernels and Runtime LMSYS Org
- SGLang Powers Beam in Production LMSYS Org
- SGLang Powers Beam in Production LMSYS Org
- SGLang Adds Day-0 Support for NVIDIA Nemotron 3.5 Super VL LMSYS Org
- SGLang Adds Day-0 Support for NVIDIA Nemotron 3.5 Super VL LMSYS Org
- SGLang and Miles Add Day-0 Support for Muse Spark Open LMSYS Org
- SGLang and Miles Add Day-0 Support for Muse Spark Open LMSYS Org
- SGLang and Miles on NVIDIA Vera Rubin LMSYS Org
- Scaling JEV-like Decision Models with SGLang LMSYS Org
- Accelerating Long-Context and Agentic Inference with NVFP4 KV Cache LMSYS Org
- Accelerating Long-Context and Agentic Inference with NVFP4 KV Cache LMSYS Org
- SGLang and Miles Add Day-0 Support for DeepSeek-V4.1 LMSYS Org
- SGLang and Miles Add Day-0 Support for DeepSeek-V4.1 LMSYS Org
- Running DeepSeek-V4-Flash and Kimi-K3 on Consumer Hardware with SSD Expert Pack LMSYS Org
- Infer-forge: Harness, Loop, and Graph Engineering Around SGLang LMSYS Org
- Infer-forge: Harness, Loop, and Graph Engineering Around SGLang LMSYS Org
- MiniMax-H3 on 8×H200: 1.95× Lossless, Up to 6.24× at 0.76–0.91 SSIM LMSYS Org
- MiniMax-H3 on 8×H200: 1.95× Lossless, Up to 6.24× at 0.76–0.91 SSIM LMSYS Org
- Qwen3.8-Flash-Next: Day-0 Support in SGLang LMSYS Org
- Chasing the Batch-1 Floor: Ling-3.0-flash Speculative Decode on Blackwell LMSYS Org
- Chasing the Batch-1 Floor: Ling-3.0-flash Speculative Decode on Blackwell LMSYS Org
- Fast Engine Recovery: Sub-Second Engine Restart for SGLang via Weight Cache Daemon LMSYS Org
- Fast Engine Recovery: Sub-Second Engine Restart for SGLang via Weight Cache Daemon LMSYS Org
- Mooncake for Miles: From Fragmented Rollout Data to Efficient Bulk I/O LMSYS Org
- Mooncake for Miles: From Fragmented Rollout Data to Efficient Bulk I/O LMSYS Org
- Pushing the Limits of Serving DeepSeek-V4-Pro LMSYS Org
- Pushing the Limits of Serving DeepSeek-V4-Pro LMSYS Org
- Miles v0.1: Production-level Post-training LMSYS Org
- Advanced CUDA Graph Techniques in Inference LMSYS Org
- Advanced CUDA Graph Techniques in SGLang LMSYS Org
- SGLang and Miles Add Day-0 Support for Qwen3.8 LMSYS Org
- SGLang Adds Day-0 Support for NVIDIA Nemotron 3.5 Lightning LMSYS Org
- Unified Radix Cache: One Tree for Hybrid Model Prefix Caching lmsys.org
最新文章39分钟前更新
- Rewriting paths as virtual weights Transformer Circuits Thread
- Helpfulness versus Fisher score · confidence intervals Transformer Circuits Thread
- Circuits Updates – October 2026 Transformer Circuits Thread
- Circuits Updates – October 2026 Transformer Circuits Thread
- Characterizing interference weights in a tiny language model Transformer Circuits Thread
- lens callout — standalone transformer-circuits.pub
- slice stack — standalone Transformer Circuits Thread
- Verbalizable Representations Form a Global Workspace in Language Models Transformer Circuits Thread
- Verbalizable Representations Form a Global Workspace in Language Models Transformer Circuits Thread
- Verbalizable Representations Form a Global Workspace in Language Models Transformer Circuits Thread
- Circuits Updates – June 2026 Transformer Circuits Thread
- Circuits Updates – June 2026 Transformer Circuits Thread
- Circuits Updates – May 2026 Transformer Circuits Thread
- Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations Transformer Circuits Thread
- HeadVis: An Interactive Tool For Investigating Attention Heads Transformer Circuits Thread
- Emotion Concepts and their Function in a Large Language Model Transformer Circuits Thread
- Circuits Updates – November 2025 Transformer Circuits Thread
- Emergent Introspective Awareness in Large Language Models Transformer Circuits Thread
- Circuits Updates – October 2025 Transformer Circuits Thread
- When Models Manipulate Manifolds: The Geometry of a Counting Task Transformer Circuits Thread
- Circuits Updates - September 2025 Transformer Circuits Thread
- Circuits Updates - August 2025 Transformer Circuits Thread
- A Toy Model of Mechanistic (Un)Faithfulness Transformer Circuits Thread
- Tracing Attention Computation Through Feature Interactions Transformer Circuits Thread
- A Toy Model of Interference Weights Transformer Circuits Thread
- Sparse mixtures of linear transforms Transformer Circuits Thread
- Circuits Updates - July 2025 Transformer Circuits Thread
- Circuits Updates - April 2025 Transformer Circuits Thread
- Progress on Attention Transformer Circuits Thread
- Attribution Graph: Human: Name one paper written by Josh Batson. Assistant: transformer-circuits.pub
- Attribution Graph: Human: A 45-year-old man presents with fatigue and mild shortness of breath that's been getting gradually worse over the past month. He's been having trouble keeping up with his regular tennis games and finds himself needing to rest after Transformer Circuits Thread
- Attribution Graph: Human: A 35-year-old woman presents with a severe headache that began this morning while she was at the gym. She describes it as "much worse than my usual headaches" and says it's concentrated at the back of her head. The pain is 8/1 Transformer Circuits Thread
- Attribution Graph: Fact: the capital of the state containing Dallas is - Transformer Circuits ThreadAttribution Graph: Fact: the capital of the state containing Dallas is Transformer Circuits Thread
- Click a feature on the left for details Transformer Circuits Thread
- Attribution Graphs Transformer Circuits Thread
- On the Biology of a Large Language Model Transformer Circuits Thread
- On the Biology of a Large Language Model Transformer Circuits Thread
- On the Biology of a Large Language Model Transformer Circuits Thread
- Circuit Tracing: Revealing Computational Graphs in Language Models Transformer Circuits Thread
- Insights on Crosscoder Model Diffing Transformer Circuits Thread
- Circuits Updates - January 2025 Transformer Circuits Thread
- Stage-Wise Model Diffing Transformer Circuits Thread
- Sparse Crosscoders for Cross-Layer Features and Model Diffing Transformer Circuits Thread
- Using Dictionary Learning Features as Classifiers Transformer Circuits Thread
- Circuits Updates - September 2024 Transformer Circuits Thread
- Circuits Updates - August 2024 Transformer Circuits Thread
- Circuits Updates - July 2024 Transformer Circuits Thread
- Feature UMAP Transformer Circuits Thread
- Dictionary Learning Run: A/3 L Transformer Circuits Thread
- Circuits Updates - June 2024 Transformer Circuits Thread
- Dictionary Learning Run: B/17 L Transformer Circuits Thread
最新文章6分钟前更新
- The Conference on Robot Learning (CoRL) takes place November 9–12, 2026, in Austin, Texas. BAIR faculty, postdocs and students will be presenting research at various sessions, including an Early Career Keynote from Professor Negar Mehr. See below for the full schedule of sessions with BAIR members, including papers co-authored with collaborators at other universities and companies. November 9–12, 2026 Austin, Texas All times CST · UTC−6 33 Main-conference papers 4 Late-breaking results (LBR) pos
- Figure 1: CUDA-to-MLX optimization translation map. CUDA optimization knowledge can be translated into architecture-native MLX strategies rather than copied instruction-for-instruction. We face a new epoch in computing. Hardware is changing rapidly — not just faster GPUs, but a growing range of chips from different vendors, each with its own architecture and often tailored to specific AI workloads. Software is changing just as fast, and AI coding tools now generate in minutes what took months of
- Overview of ABBEL compared to traditional recursive summarization. Beliefs replace the full interaction history as the agent’s working context, and belief grading improves performance by supervising the contents of each belief state.. As task horizons grow, LLM contexts can’t scale forever. Self-summarization enables concise, interpretable contexts, but at a significant performance cost, especially for human assistance domains where high quality data is scarce, e.g., collaborative code generatio
- ... government of the people, by the people, for the people ... — Abraham Lincoln, Gettysburg Address (1863) The cost of AI is dropping rapidly. GPT-4-class capabilities cost roughly $30 per million tokens in early 2023; today the same runs under $1 , and some providers are pushing costs below $0.10 . Across benchmarks, inference prices have fallen between 9x and 900x per year , with a median decline near 50x. Even frontier models are getting dramatically cheaper each generation, with open-sourc
- Congratulations to the Berkeley Artificial Intelligence Research (BAIR) Lab class of 2026! This year, BAIR celebrates another remarkable group of Ph.D. graduates whose curiosity, creativity, and perseverance have pushed the frontiers of artificial intelligence and machine learning. Their work spans the breadth of modern AI — robotics and embodied intelligence, large language models and reasoning, computer vision, generative modeling, AI safety, human-AI interaction, AI for science and healthcare
- Overview of adaptive parallel reasoning. What if a reasoning model could decide for itself when to decompose and parallelize independent subtasks, how many concurrent threads to spawn, and how to coordinate them based on the problem at hand? We provide a detailed analysis of recent progress in the field of parallel reasoning, especially Adaptive Parallel Reasoning. Disclosure: this post is part landscape survey, part perspective on adaptive parallel reasoning. One of the authors (Tony Lian) co-l
- GRASP is a new gradient-based planner for learned dynamics (a “world model”) that makes long-horizon planning practical by (1) lifting the trajectory into virtual states so optimization is parallel across time, (2) adding stochasticity directly to the state iterates for exploration, and (3) reshaping gradients so actions get clean signals while we avoid brittle “state-input” gradients through high-dimensional vision models. Large, learned world models are becoming increasingly capable. They can
- Understanding the behavior of complex machine learning systems, particularly Large Language Models (LLMs), is a critical challenge in modern artificial intelligence. Interpretability research aims to make the decision-making process more transparent to model builders and impacted humans, a step toward safer and more trustworthy AI. To gain a comprehensive understanding, we can analyze these systems through different lenses: feature attribution , which isolates the specific input features driving
- An encoder (optical system) maps objects to noiseless images, which noise corrupts into measurements. Our information estimator uses only these noisy measurements and a noise model to quantify how well measurements distinguish objects. Many imaging systems produce measurements that humans never see or cannot interpret directly. Your smartphone processes raw sensor data through algorithms before producing the final photo. MRI scanners collect frequency-space measurements that require reconstructi
- In this post, I’ll introduce a reinforcement learning (RL) algorithm based on an “alternative” paradigm: divide and conquer . Unlike traditional methods, this algorithm is not based on temporal difference (TD) learning (which has scalability challenges ), and scales well to long-horizon tasks. We can do Reinforcement Learning (RL) based on divide and conquer, instead of temporal difference (TD) learning. Problem setting: off-policy RL Our problem setting is off-policy RL . Let’s briefly review w
- What exactly does word2vec learn, and how? Answering this question amounts to understanding representation learning in a minimal yet interesting language modeling task. Despite the fact that word2vec is a well-known precursor to modern language models, for many years, researchers lacked a quantitative and predictive theory describing its learning process. In our new paper , we finally provide such a theory. We prove that there are realistic, practical regimes in which the learning problem reduce
- × Predicting Ego-centric Video from human Actions (PEVA) . Given past video frames and an action specifying a desired change in 3D pose, PEVA predicts the next video frame. Our results show that, given the first frame and a sequence of actions, our model can generate videos of atomic actions (a), simulate counterfactuals (b), and support long video generation (c). Recent years have brought significant advances in world models that learn to simulate future outcomes for planning and control. From
- Recent advances in Large Language Models (LLMs) enable exciting LLM-integrated applications. However, as LLMs have improved, so have the attacks against them. Prompt injection attack is listed as the #1 threat by OWASP to LLM-integrated applications, where an LLM input contains a trusted prompt (instruction) and an untrusted data. The data may contain injected instructions to arbitrarily manipulate the LLM. As an example, to unfairly promote “Restaurant A”, its owner could use prompt injection t
- PLAID is a multimodal generative model that simultaneously generates protein 1D sequence and 3D structure, by learning the latent space of protein folding models. The awarding of the 2024 Nobel Prize to AlphaFold2 marks an important moment of recognition for the of AI role in biology. What comes next after protein folding? In PLAID , we develop a method that learns to sample from the latent space of protein folding models to generate new proteins. It can accept compositional function and organis
最新文章56分钟前更新
- Based on: Potosnak, W., Wolff, M., Cao, M., Ma, R., Konstantinova, T., Efimov, D., Mahoney, M.W., Oreshkin, B., & Olivares, K.G. "Forking-Sequences: Statistically and Computationally Efficient Multi-Horizon Forecasting with Reduced Volatility." Transactions on Machine Learning Research, 2026. (Disclaimer: Code implementation not used in the paper; not affiliated with Amazon — provided as a reference for forking-sequences and forecast ensembling) TL;DR Ensembling, nearly for free. Forking-sequenc
- In healthcare settings where patients use LLMs as a medical assistant, LLM performance differs between evaluation and deployment. (a) Bean et al. (2025) find a 61 percentage point difference between evaluation and deployment. (b) We argue this gap arises not from poorly designed benchmarks, but from implicit assumptions embedded in evaluation protocols that fail to hold at deployment. (c) We propose a taxonomy that categorizes assumptions into two types, task and outcome, to diagnose where the g
最新文章45分钟前更新
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- SuperRed · Research Berkeley RDI
- CUA-Lite : An Open Platform for Computer-Use Agents Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Towards Autonomous Software Development Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- CyberGym-E2E: Scalable Real-World Benchmark for AI Agents’ End-to-End Cybersecurity Capabilities Berkeley RDI
- SageCTF: The Most Capable AI Agent for The Most Challenging CTF Berkeley RDI
- SageCTF: The Most Capable AI Agent for The Most Challenging CTF Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Peer-Preservation in Frontier Models Berkeley RDI
- Peer-Preservation in Frontier Models Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Berkeley Xcelerator Berkeley RDI
- Agentic AI Summit 2026 Berkeley RDI
- Practical Privacy Lecture Berkeley RDI
- Copy of Berkley MOOC Berkeley RDI
- Agentbeats Competition Info Session - 11.6 Berkeley RDI
- Where will stablecoins have a big impact? Berkeley RDI
- Lecture 2: EXPOSING THE HIDDEN FINANCIAL MACHINERY Berkeley RDI
- AgentX-AgentBeats Competition Berkeley RDI
- Challenges and Lessons from Training Agentic Models Berkeley RDI
- CyberGym: Evaluating AI Agents’ Real-World Cybersecurity Capabilities at Scale Berkeley RDI
- agentic AI agent evaluations & project overview Slides 10.6.2025 [CS 194-196] Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Agentic AI Summit 2025 - A Record-Breaking Gathering at UC Berkeley! Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Hunting Vulnerabilities with AI rdi.berkeley.edu
- Research Publications Berkeley RDI
- Berkeley talk 2025 Yangqing Jia Berkeley RDI
- Frontier AI in Cybersecurity: Risks and Opportunities Berkeley RDI
- Introduction to training LLMs for AI agents Berkeley RDI
- Education Berkeley RDI
- Copy of Slides 9/8 [CS194-196].pptx Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- CS294/194-196: Agentic AI Berkeley RDI
- Center for Responsible, Decentralized Intelligence at Berkeley Berkeley RDI
- Decentralized Finance Berkeley RDI
- Frontier AI's Impact on the Cybersecurity Landscape: Current Status and Future Directions Berkeley RDI
- Advanced LLM Agents Berkeley RDI
- The Gap Between Data Rights Ideals and Reality Berkeley RDI
最新文章35分钟前更新
- Preface This essay argues that rational people don’t have goals, and that rational AIs shouldn’t have goals. Human actions are rational not because we direct them at some final ‘goals,’ but because we align actions to practices [1] : networks of actions, action-dispositions, action-evaluation criteria,
- "In projecting language back as the model for thought, we lose sight of the tacit embodied understanding that undergirds our intelligence." –Terry Winograd The recent successes of generative AI models have convinced some that AGI is imminent. While these models appear to capture the essence of human
- What is the Role of Mathematics in Modern Machine Learning? The past decade has witnessed a shift in how progress is made in machine learning. Research involving carefully designed and mathematically principled architectures result in only marginal improvements while compute-intensive and engineering-first efforts that scale to ever larger training sets
- LLM-based chatbots’ capabilities have been advancing every month. These improvements are mostly measured by benchmarks like MMLU, HumanEval, and MATH (e.g. sonnet 3.5, gpt-4o). However, as these measures get more and more saturated, is user experience increasing in proportion to these scores? If we envision a future
- Introduction Imagine yourself a decade ago, jumping directly into the present shock of conversing naturally with an encyclopedic AI that crafts images, writes code, and debates philosophy. Won’t this technology almost certainly transform society — and hasn’t AI’s impact on us so far been
- The AI revolution drove frenzied investment in both private and public companies and captured the public’s imagination in 2023. Transformational consumer products like ChatGPT are powered by Large Language Models (LLMs) that excel at modeling sequences of tokens that represent words or parts of words [2]. Amazingly, structural
- A brief overview and discussion on gender bias in AI
- Is Attention all you need? Mamba, a novel AI model based on State Space Models (SSMs), emerges as a formidable alternative to the widely used Transformer models, addressing their inefficiency in processing long sequences.
- Exploring the utility of large language models in autonomous driving: Can they be trusted for self-driving cars, and what are the key challenges?
- 'Vec2text' can serve as a solution for accurately reverting embeddings back into text, thus highlighting the urgent need for revisiting security protocols around embedded data.
- Have you ever trained a model you thought was good, but then it failed miserably when applied to real world data? If so, you’re in good company.
- On the the pivotal role that Deep Learning has played as a key enabler for advancing single-cell sequencing technologies.
- On fish counting – a complex sociotechnical problem in a field that is going through the process of digital transformation.
- In this article, we will talk about classical computation : the kind of computation typically found in an undergraduate Computer Science course on Algorithms and Data Structures [1]. Think shortest path-finding, sorting, clever ways to break problems down into simpler problems, incredible ways to organise data for efficient retrieval and updates.
- This essay first appeared in Reboot . Credulous, breathless coverage of “AI existential risk” (abbreviated “x-risk”) has reached the mainstream. Who could have foreseen that the smallcaps onomatopoeia “ꜰᴏᴏᴍ” — both evocative of and directly derived from children’s cartoons —
最新文章28分钟前更新
- The Batch News & Insights: Building an agent harness has been a rapidly growing method to build an AI agent. Dear friends,Building an agent harness has been a rapidly growing method to build an AI agent. The OpenWorker project just released the open source OpenWorker Agent Harness SDK, which I invite you to check out. If you have not built an agent harness before, I encourage you to do so. (This skill is reflected in the AI Engineering Skills Map, and job interviews increasingly ask people if th
- The Batch News & Insights: LinkedIn data shows that AI has helped create 2 million jobs. Dear friends,LinkedIn data shows that AI has helped create 2 million jobs. Further, AI Engineer is among the top 10 fastest-growing jobs in seven major economies. I know many people are worried about job loss — and there have been painful layoffs. At the same time, the LinkedIn data undermines the jobpocalypse story that AI is leading to massive unemployment. Of the many jobs AI is creating, I think some of
- The Batch News & Insights: Following our annual tradition, this special Halloween edition of The Batch treats leading AI fears. Dear friends, Following our annual tradition, this special Halloween edition of The Batch treats leading AI fears. AI’s potency means our community has real problems to solve. At the same time, fear-mongering has long been a tactic for drumming up publicity or influencing investors and politicians. So amidst all the exciting progress, fears about AI’s power have long be
- The Batch News & Insights: What should an engineering leader such as a VP Engineering or an Engineering manager do to help their team become... Dear friends, What should an engineering leader such as a VP Engineering or an Engineering manager do to help their team become more proficient in AI? I speak regularly with engineering leaders, and want to share some effective practices. If you’re starting a new team of 2-3 persons, you can start with exclusively AI-native people, but many large compani
- The Batch News & Insights: The gulf between LLM performance on verifiable tasks vs. unverifiable tasks that require judgement is larger than... Dear friends, The gulf between LLM performance on verifiable tasks vs. unverifiable tasks that require judgement is larger than most people realize. Verifiable tasks, like coding or solving math problems, have a right answer that can be checked automatically. Judgement tasks — like how to respond to a customer complaint, how to organize data collection f
- The Batch News & Insights: I think most people still underestimate the data center build-outs that will be required to meet demand. Dear friends,I think most people still underestimate the data center build-outs that will be required to meet demand. In 2024, I predicted that we would need more compute infrastructure for inference because of the large number of tokens consumed by agentic workflows. This remains true today. But many people still underestimate the compute infrastructure that will b
- The Batch News & Insights: Anthropic recently released an encouraging analysis of the cyber capabilities of open weight model GLM-5.3. Dear friends,Anthropic recently released an encouraging analysis of the cyber capabilities of open weight model GLM-5.3. The study shows that GLM-5.3 approaches the cyber capabilities of Claude's closed weight Mythos. For example, the diagram below shows that, using a comparable number of tokens, GLM-5.3 succeeded in 12% of attempts to exploit vulnerabilities on
- The Batch News & Insights: Moving forward on early stage, 0-to-1 projects and mature projects requires very different tactics. Dear friends,Moving forward on early stage, 0-to-1 projects and mature projects requires very different tactics. For those who aspire to be skilled at AI Engineering, I've found that selecting the right tactic based on stage of project is one of the hardest but most important things to learn.In the 5 letters on the AI Engineering Skills Map, you may have noticed calibrat
- The Batch News & Insights: The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. Dear friends,The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled by what appears to be a well orchestrated PR campaign — has drummed up considerable fear. I worry that it represents a setback for our field.I have written freq
- The Batch News & Insights: When you’re skilled at AI Engineering, your best work won’t be merely implementing a product that someone else... Dear friends,When you’re skilled at AI Engineering, your best work won’t be merely implementing a product that someone else spec’ed out. Instead, you will actively shape the build.Before modern AI tools accelerated and expanded what a single developer could do, tech companies established the practice of having product managers (PMs) and designers specify wh
- The Batch News & Insights: A key AI engineering skill is using coding agents. Dear friends,A key AI engineering skill is using coding agents. Your skill at steering them both to write code and to carry out non-code tasks, such as analyzing data or managing system operations, allows you to get a lot more done.The rapid pace of evolution for coding agents means this skill, too, is evolving rapidly — faster than other top-level AI engineering skills. Proprietary agents (like Claude Code, Codex, and
- The Batch News & Insights: How have software engineering fundamentals changed with agentic coding? Dear friends, How have software engineering fundamentals changed with agentic coding? Even when you use a coding agent to write all your code, understanding software fundamentals is important for steering your agent to make the tradeoffs you want — or to even know what tradeoffs exist to be made. Additionally, when you’re building an AI application, the AI core is often expressed through a broader
- The Batch News & Insights: I previously wrote about our AI Engineering Skills Map, with the highest level skills being (i) Building and... Dear friends,I previously wrote about our AI Engineering Skills Map, with the highest level skills being (i) Building and deploying AI applications, (ii) Software engineering fundamentals, (iii) Using coding agents, and (iv) Shaping the build. In this letter, I will flesh out the first of them. Being skilled at building and deploying AI applications means kno
- The Batch News & Insights: I am delighted to present The AI Engineering Skills Map. Dear friends,I am delighted to present The AI Engineering Skills Map. AI allows us to build software very differently today than in 2022, and everyone with the skills to take advantage of this shift has numerous exciting project and job opportunities. But with the noisy, hype-filled, information environment around AI, what are the most valuable skills for you to learn? I have been working with my team to synthesi
- The Batch News & Insights: I’m glad the idea of “tokenmaxxing” — that individuals and companies should use as many tokens as possible to boost... Dear friends,I’m glad the idea of “tokenmaxxing” — that individuals and companies should use as many tokens as possible to boost productivity — is finally dying out. As much as I encourage everyone to make ample use of AI, the practical reality is that increasing token usage beyond a certain point gives diminishing returns because there are still bottl
- The Batch News & Insights: My team recently had our own version of Hugging Face’s experience when closed models failed to defend the company... Dear friends, My team recently had our own version of Hugging Face’s experience when closed models failed to defend the company following an accidental cyberattack from OpenAI, leading Hugging Face to use the open weight GLM 5.2 instead. When we tried using Claude Code (with Fable 5) and OpenAI’s Codex harness (with GPT-5.6 Sol) to conduct a security rev
- The Batch News & Insights: A few frontier labs have tried to tell a story of open models being dangerous because they can be used to launch... Dear friends, A few frontier labs have tried to tell a story of open models being dangerous because they can be used to launch cyberattacks, and of their “safe” proprietary models with strong guardrails being there to defend us. This week, the opposite happened: Users of a closed model unintentionally launched a significant cyberattack; other closed model
- The Batch News & Insights: As AI increasingly automates coding, it frees up developers to spend time on high-level software development tasks... Dear friends,As AI increasingly automates coding, it frees up developers to spend time on high-level software development tasks traditionally reserved for senior engineers, like deciding on technical architecture and participating in scoping product requirements. Demand for this type of work is growing, since it is an economic complement to coding, whic
- The Batch News & Insights: The key idea of agentic coding loops is to have an agent keep working until it satisfies a condition, such as achieving... Dear friends,The key idea of agentic coding loops is to have an agent keep working until it satisfies a condition, such as achieving a product specification. Coming up with the spec, the evals, or test set is one of the hardest tasks for many AI systems, and a key place to inject human knowledge. Once you have a clear spec, the paths to fulfilling
- The Batch News & Insights: The AI world has become incredibly noisy. Social media, traditional media, and an army of marketers produce a cacophony... Dear friends,The AI world has become incredibly noisy. Social media, traditional media, and an army of marketers produce a cacophony of hype and content that are often secretly sales pitches for their products. Good ideas are buried in the noise, but it is hard to figure out what’s worth your time to learn. I want to explain plainly how we decide w
- A New Generation Studies AI, Apple's Recipe for On-Device Models, GLM5.2 Tackles Open-Ended ProblemsThe Batch News & Insights: “Loop engineering” is a hot buzzphrase after Boris Cherney (Claude Code’s creator) and Peter Steinberger (OpenClaw's... Dear friends,“Loop engineering” is a hot buzzphrase after mentions of it by Boris Cherny (Claude Code’s creator) and Peter Steinberger (OpenClaw's creator) went viral on social media. Loops are now a key part of how we get AI agents to iterate at length to build software. In this letter, I’d like to share my 3 key loops, shown in the image below, for
- The Batch AI News and Insights: Over the last two weeks, both the U.S. Government and Anthropic took significant actions that demonstrated...
- The Batch AI News and Insights: If you haven’t already, I encourage you to experiment with using AI agents not just to chat but to actually do work for you on…
- The Batch AI News and Insights: There have been intense efforts over the past few years to lobby governments to pass AI laws for regulatory...
- The Batch AI News and Insights: One of the new, buzzy jobs in Silicon Valley is the AI Forward Deployed Engineer (FDE), an engineer who is...
- The Batch AI News and Insights: Harvard University just voted to limit the number of A grades given in undergraduate classes to about 20%...
- The Batch AI News and Insights: We’ve been working on AI Andrew, an AI companion shaped by my personality.