Toutes les curations

Ce qui est entré en base, sans jugement.

Montrer Tout Porteurs de lien
Tout remettre
90222posts
954auteurs
9348liens
1495 hde recul
626AI High Signal
458AI & IA
104AI & LLM
38Gpt/LLM/GenAI
69Vision & vidéos…
19Presse politique
Recoupements sur la fenêtre : 8 liens et 73 tweets relayés par plusieurs membres · au mieux 28 personnes sur un même objet.
O
Andreas Stuhlmüller a relayé @ohabryka
Oliver Habryka @ohabryka · 03:34
Also we are allowing funds to run on Lightcone Commons. The first of these will be an epistemics fund, sponsored by @Oughtinc, led by @NathanpmYoung. They are interested in projects helping people think better and want submissions by Nov 26th. More here:
Epistemics Fund Epistemics fund. Please apply! Ought plans to make $500k to $5M in grants to community notes, provenance tracking, forecasting and other tools that help people work out what's true.
P
mei a relayé @PriMachAI
Principled Machines @PriMachAI · 03:04
Sharing an early result: Sierra verified Cortex at 73.2% on τ³-benchmark. We build algorithmic complements to LLMs. We don’t fight what models do well. We fill in the rest.
Principled Machines Cortex on τ³-bench — Principled Machines On τ³-bench banking, Cortex passed 73.2% of tasks, against 55.2% for the best published model. Sierra's own paper shows that finding the documents is half the problem. We describe how Cortex handles the other half, and what we deliberately did not do.
Geek Lite @QingQ77 · 01:53
验证 Windows 应用在干净机器上的安装与运行通常要开虚拟机或准备新机器，这个工具用 CLI 和 MCP 驱动 Windows Sandbox 启动一次性 Windows 11 环境，部署应用、跑 UI 自动化并收集证据后销毁。
GitHub GitHub - shanselman/sandbox-mcp: Fresh, throwaway Windows machines for testing your apps: an MCP server and CLI that automate Windows Sandbox Fresh, throwaway Windows machines for testing your apps: an MCP server and CLI that automate Windows Sandbox - shanselman/sandbox-mcp
小互 @xiaohu · 01:50
Anthropic 发布 Claude Fable 5.2 与 Mythos 5.2： 8 天手搓操作系统、30 小时证完费马大定理 提速 30% 且百万 token 降至 $8/ $40 详细：
小互 · AI 解读站 Anthropic 发布 Claude Fable 5.2 与 Mythos 5.2：8 天手搓操作系统、30 小时证完费马大定理，提速 30% 且百万 token 降至 $8/$40 · XiaoHu.AI Fable 5.2 面向全平台开放、Mythos 5.2 定向赋能安全与生命科学，15 项基准、5 张成本曲线与 4 段实机演示完整拆解
Simon Willison @simonw · 01:09
There is a mysterious intermittent beep coming from somewhere in your room of unopened boxes. Do you: 1. Open boxes methodically until you find it 2. Vibe code software to triangulate it using two laptops, two iPads and a phone all listening at once
Simon Willison’s Weblog Locating the source of a mysterious beep with Claude Our house has a box room, full of boxes that we still haven’t unpacked from our last move. A couple of weeks ago something in that room started intermittently beeping. …
Omar Khattab a relayé @_reachsumit
Sumit @_reachsumit · 00:47
MV-IVF: Multi-Vector Retrieval via Multi-Vector Clustering Builds an inverted-file index directly over multi-vector embeddings by clustering whole documents, rather than single-vector proxies. Shows strong latency-recall tradeoffs on BEIR and LoTTE. 📝
arXiv.org MV-IVF: Multi-Vector Retrieval via Multi-Vector Clustering Multi-vector retrieval represents each query and document as a set of embeddings and scores relevance via fine-grained set-to-set interactions such as MaxSim (Chamfer similarity). This expressiveness improves retrieval quality over single-vector representations, but makes search substantially more expensive. Existing systems address this challenge primarily through proxy representations or filter-and-rerank pipelines that reduce candidate generation to single-vector search. We study the complementary direction of building a native IVF-style index directly over multi-vector data. We introduce
LiveAvatar @TryLiveAvatar · 00:35
From 1 practice session per semester to unlimited. University of Colorado Anschutz built AI patients with LiveAvatar so future doctors can rehearse their hardest conversations anytime.
HeyGen Comment l’Université du Colorado utilise les avatars en direct pour la formation médicale Découvrez comment l’Université du Colorado utilise HeyGen LiveAvatar pour créer des patients virtuels interactifs grâce à l’IA, déployer des simulations médicales à grande échelle et améliorer la formation à la communication.
Geek Lite @QingQ77 · 00:25
超有意思，在 Mac 上演示、录屏或讲屏幕时，用一个跟着鼠标的小手指针指出位置，点哪指哪，还能直接在屏幕上画。 TinyHand 是 macOS 13+ 的菜单栏小工具，把鼠标指针换成一根棍子上的小手，用来在演示、录屏或跟人讲屏幕时指出位置。 按 ⌃⌥H
GitHub GitHub - nilbuild/tinyhand: A tiny hand on a stick that points at things on your Mac screen. A tiny hand on a stick that points at things on your Mac screen. - nilbuild/tinyhand
Sakana AI @SakanaAILabs · 00:15
【採用情報】Information Security Manager (Defense) のポジションをオープン🐟 Sakana AIの防衛事業部門にて、官公庁を含むお客さまとの契約に基づいて、情報の管理保全体制を整備・運用するポジションです。下記のような業務を担当していただきます。 ・社内規程、管理
apply.workable.com Information Security Manager (Defense) - Sakana AI 私たちはAIが今世紀におけるもっとも重要な技術開発だと考えています。AIの研究開発は、私たちの文明を継続して変革させていくでしょう。人々、ひいては政府や企業によるテクノロジーの利用は、AIによって加速し、急速な変革の段階にあります。あなたには、情報セキュリティマネージャーとして、官公庁を含むお客様との契約に基づいて取り扱う情報の管理保全体制を、整備・運用していただきます。社内規程や管理組織の設計から、重要情報を扱う従業員の登録・教育、文書や情報システムの管理、監査・検査への対応、インシデント発生時の...
Bryan Catanzaro @ctnzr · 00:03
So proud that @NVIDIAGeForce DLSS5 has won the 2027 Best of Innovation in Gaming & eSports award at CES, and also is an honoree for Artificial Intelligence as a category.
ces.tech NVIDIA DLSS 5 CES® is the most powerful tech event in the world — the proving ground for breakthrough technologies and global innovators. This is where brands get business done, meet new partners and where the industry’s sharpest minds take the stage to unveil their latest releases and boldest breakthroughs. Owned and produced by the Consumer Technology Association (CTA)®, CES is the only trade show that showcases the entire tech landscape at one event.
Mira Murati a relayé @thinkymachines
Thinking Machines @thinkymachines · 09/11 23:46
Inkling 2 Small is available on @opencode, free for a limited time. It was trained for agentic coding and knowledge work; we look forward to seeing how it works for you!
OpenCode Utilisation, coût et rang de Inkling 2 Small | OpenCode Data Consultez les données d'utilisation OpenCode de Inkling 2 Small, avec le volume de tokens, le rang hebdomadaire, la répartition des tokens, les coûts, le taux de cache, les sessions, les répartitions géographiques et les modèles proches.
F
Pierre Sautarel a relayé @F_Desouche
Fdesouche.com est une revue de presse @F_Desouche · 09/11 23:30
Cinq autres hommes mis en examen, un groupe Telegram privé et des enfants confiés à un assistant familial : les révélations du Parisien sur l’affaire Cottineau.
Fdesouche « Il était envié par les autres parce qu’il avait la garde d’enfants » : les révélations sur le réseau pédocriminel « Butterfly » et l’ex-candidat LFI Cottineau Le Parisien révèle les rouages du réseau pédocriminel Butterfly : cinq autres hommes mis en examen dans l’affaire de l’ex-candidat LFI Pierre-Alain Cottineau.
Hamel Husain a relayé @sh_reya
Shreya Shankar @sh_reya · 09/11 23:12
Want to do AI-powered data extraction 2x faster and cheaper than other carefully-engineered inference solutions? Check our new AI-powered extraction technique in Quail, our open-source query engine with AI functions!
X (formerly Twitter) X - The Everything App
G
Google Research a relayé @GoogleForHealth
Google for Health @GoogleForHealth · 09/11 23:03
Evaluating generative technology for health functions takes more than multiple-choice tests. Our new whitepaper details our framework combining benchmarks, automated assessments, and expert human review to assess accuracy, safety, and usefulness.
Google We're sharing our framework for building safer and more helpful AI for health. Many of us have turned to AI with a health question or received care from a doctor who’s using it. The adoption of technology for health is accelerating and impacts bill…
Deep Learning Weekly @dl_weekly · 09/11 23:01
🤖 From this week's issue: An engineering blog post about issuing agents working Stripe test keys from a pre-provisioned pool behind a proof-of-work challenge, with over 50,000 sandboxes provisioned per day.
stripe.dev Claimable sandboxes: Frictionless Stripe accounts for agents Claimable sandboxes are a primitive that unblocks agents to build and test Stripe integrations without having to sign up. Learn how they work and how we navigated security, latency, cost, and UX tradeoffs.
sunil pai a relayé @lydiahallie
Lydia Hallie ✨ @lydiahallie · 09/11 22:27
Fable 5.2 is here! 🎉 I noticed it's really thorough at verifying results, way more than 5.1. I've mainly been using it as the orchestrator with Opus 5.5 and it's so good Pro/Max/Ultra/Team also got at least one free Fable reset, make sure to try it out!
clau.de https://clau.de/reset
L
swyx 🔜 aie.codes a relayé @RLanceMartin
Lance Martin @RLanceMartin · 09/11 22:19
i'm happy to share some recent work on safely scaling a team of 1,000 Fable 5.2 agent to tackle ambitious, long-horizon work. the team was able to formalize a proof of Fermat's Last Theorem through coordination within our safety boundaries. read it here:
claude.dev Blog Safely scaling multi-agent teams with Claude Fable 5.2 / claude.dev Blog How a team of 1,000 agents running an early version of Claude Fable 5.2 reproduced a formal proof of Fermat's Last Theorem with general tools, a shared database, and layered safety controls.
Peter Yang a cité @petergyang
Peter Yang @petergyang · 09/11 22:15
Here's my new 3D kart racer game I built with @claudeai Fable 5.2: → It runs directly in your browser → 6 racers with drifting, ramps, and items → Supports both single and multiplayer! Best of all, you can play it right here for free:: 📌 Watch my
turbo-tails.vercel.app TURBO TAILS
AI.Engineer 🔜 CODE @aiDotEngineer · 09/11 22:00
The same failure, 37 times. Then a proposed PR to fix the factory. @zachlloydtweets of @warpdotdev shows how agents propose improvements to their own workflows. See Zach at AIE CODE this week. Nov 10 to 12, San Francisco. Apply:
ai.engineer AI Engineer Code Summit 2026: AI Engineering Conference, Nov 10-12, San Francisco The application-only AI Engineer Code Summit brings together leading coding-agent builders and AI engineers in San Francisco, November 10-12, 2026.
Tommy a relayé @NousResearch
Nous Research @NousResearch · 09/11 21:59
Hermes Cloud agents now open in a browser tab with the exact same interface as Hermes Desktop. To launch a personal agent, pick a name and a model in Nous Portal. We host it for you and it runs 24/7 in its own container, so it can handle scheduled jobs
Nous Portal Hermes Cloud — Always-On AI Agent Hosting Hermes Cloud runs your Hermes Agent 24/7 — always-on autonomous agent with persistent memory and growing capabilities. Deploy in minutes, no server setup.
ollama @ollama · 09/11 21:57
Our new usage page makes it easy to monitor your daily and weekly usage, including: - Spend per day - Spend per API key - Input and output tokens per model - Cached input rate Go to
ollama.com Sign in
Thinking Machines @thinkymachines · 09/11 21:53
Inkling 2 Small is available on @opencode, free for a limited time. It was trained for agentic coding and knowledge work; we look forward to seeing how it works for you!
OpenCode Utilisation, coût et rang de Inkling 2 Small | OpenCode Data Consultez les données d'utilisation OpenCode de Inkling 2 Small, avec le volume de tokens, le rang hebdomadaire, la répartition des tokens, les coûts, le taux de cache, les sessions, les répartitions géographiques et les modèles proches.
Ronald van Loon @Ronald_vanLoon · 09/11 21:49
Tiny injectable glucose sensor could make blood sugar monitoring 10 times cheaper by Rachit Agarwal @DigitalTrends Learn more: #MedTech #HealthTech #Tech #TechForGood
Digital Trends Tiny injectable glucose sensor could make blood sugar monitoring 10 times cheaper Texas A&M researchers have built a tiny injectable glucose sensor that uses near-infrared light and a cheaper sugar molecule, and it could one day replace today’s pricey, patch-based monitors.
Teknium 🪽 a relayé @NousResearch
Nous Research @NousResearch · 09/11 21:34
Hermes Cloud agents now open in a browser tab with the exact same interface as Hermes Desktop. To launch a personal agent, pick a name and a model in Nous Portal. We host it for you and it runs 24/7 in its own container, so it can handle scheduled jobs
Nous Portal Hermes Cloud — Always-On AI Agent Hosting Hermes Cloud runs your Hermes Agent 24/7 — always-on autonomous agent with persistent memory and growing capabilities. Deploy in minutes, no server setup.
N
John Nay a relayé @normativeai
Norm @normativeai · 09/11 21:00
A note on Intent of AI Agents Operating Inside Companies. We discuss open questions in a world where multi-agent systems are increasingly deployed. The integration of AI and expert human control offers a foundation for governance alongside performance.
Norm “Intent” of AI Agents Operating Inside Companies | Norm Automated AI systems, often working together, increasingly take actions inside companies. Companies are already deploying agents, operating with little to no human involvement, to perform high-stakes roles. This will ramp up as AI continues advancing.
S
Raphaël Millière a relayé @solimlegris
Solim LeGris @solimlegris · 09/11 20:51
1/ How do people and machines reason in more open-ended, complex cognitive domains? We study their incorrect inferences in the ARC domain to find out. New preprint: “Synthesizing symbolic models of human and machine reasoning at scale” 🧵 @PMMondorf
arXiv.org Synthesizing symbolic models of human and machine reasoning at scale Symbolic models offer a powerful account of cognition, but their scope has historically been limited by the need for hand-engineering and computational tractability. We show how advances in artificial intelligence (AI) can scale symbolic modeling and help overcome these engineering limitations. We introduce Behavior-Conditioned Symbolic Synthesis (BCSS)---an approach for automated search of interpretable, deterministic symbolic programs that verifiably reproduce behavior---and apply it to model both human and machine behavior across hundreds of challenging reasoning problems. Analyzing these p
N
Stanford NLP Group a relayé @HEI
Natural Language Processing Papers @HEI · 09/11 20:48
SAFE-Teen: Evaluating Chatbot Safety Across Adolescent Mental Health Conversations Andrea Mock, Valerie Chen, Eric Lin, Rayan Krishnan, Diyi Yang, Caroline Figueroa [𝚌𝚜.𝙲𝙻]
arXiv.org SAFE-Teen: Evaluating Chatbot Safety Across Adolescent Mental Health Conversations Adolescents are increasingly turning to AI chatbots for mental health support, advice, and companionship, raising safety concerns. While recent evaluations have started to examine safety across multiple conversational turns, few compare how slight changes in a teen's responses can affect a chatbot's safety in the same scenario. We conduct a clinician-informed evaluation of chatbot safety in simulated conversations with adolescents aged 13-17, examining how safety failures vary with adolescents' responses and teen-aware system prompting. We evaluated 32,555 simulated conversations across 18 cha
Stephanie Palazzolo @steph_palazzolo · 09/11 20:46
Fireworks, the most highly-valued out of a growing group of inference providers, told customers including Harvey that it had been hit with a hack that exposed customer data. Some customers are now considering moving their business off the platform.
theinformation.com New Fireworks Hack Threatens $1 Billion Business Fireworks, a startup that helps developers run and train open-source models, in the last week has told some of its large customers including Harvey of a recent, more serious hack to its systems,...
T
Chubby♨️ a relayé @notmiramurati
Ted @notmiramurati · 09/11 20:42
Meet Yvligo, our open-weights model for running X ads. (3x better than Grok 4.7) At, we built an X simulation and post-trained DeepSeek V4.1 Flash, outperforming frontier models Live results: Yvligo: 383 relevant website visits for $200 Grok: 123 visits
loglog.ai log(log(ai)) — RL environments High-fidelity RL environments of real business contexts: social platforms, job boards, quantitative trading and autoresearch, grounded in public and proprietary data.
Ethan Mollick @emollick · 09/11 20:35
In addition to more serious examples I shared in my post, I had Fable 5.2 build a parody reverse Scooby Doo game where you have to scare away meddling kids. Here is the trailer it made (including the music and sounds) and the game (which is fun) is here:
i-would-have-gotten-away-with-it.netlify.app I Would Have Gotten Away With It Be the villain of a Saturday-morning mystery cartoon. Scare everyone out of Grimsby Manor before those meddling kids unmask you.
R
Kyle Russell a relayé @rehan_shei
Rehan Sheikh @rehan_shei · 09/11 20:33
Introducing patch: build and play games. - Every game dev tool in one CLI and MCP - 50,000+ open-source assets - Remix any game on patch, or mod the ones on your PC - Ship it to everyone on patch, opensource :) We're opening it to everyone today:
Patch Patch · Build and play Patch is a Windows app for playing and building game mods, mashups and browser games. It finds the games on your PC, sets up a mod with one press of Play, and lets your own AI coding agent build, test and publish new ones.
R
Machine Learning FLX a relayé @imryven
Ryven @imryven · 09/11 20:23
This is f*cking insane, the CEO of Y Combinator open-sourced his own second brain and most people still don’t know it exists 8 second brain repos you haven’t seen yet: -> gbrain (by garrytan) the second brain of the YC CEO, every fact keeps its source
GitHub GitHub - garrytan/gbrain: Garry's Opinionated OpenClaw/Hermes Agent Brain Garry's Opinionated OpenClaw/Hermes Agent Brain. Contribute to garrytan/gbrain development by creating an account on GitHub.
C
Hugging Face a relayé @calebfahlgren
Caleb @calebfahlgren · 09/11 19:59
Hugging Face is now the best place to host and share 3D assets 🧊🤗 Drop in .glb, .ply, .stl, or splat files and get a 3D gallery automatically. There are thousands of 3D models already in public datasets! 350+ datasets to explore: Docs:
huggingface.co 3D datasets in the Dataset Viewer - a cfahlgren1 Collection Datasets whose Dataset Viewer shows 3D previews of their meshes, and Gaussian splat datasets that preview in the file viewer.
C
clem 🤗 a relayé @calebfahlgren
Caleb @calebfahlgren · 09/11 19:57
Hugging Face is now the best place to host and share 3D assets 🧊🤗 Drop in .glb, .ply, .stl, or splat files and get a 3D gallery automatically. There are thousands of 3D models already in public datasets! 350+ datasets to explore: Docs:
huggingface.co 3D datasets in the Dataset Viewer - a cfahlgren1 Collection Datasets whose Dataset Viewer shows 3D previews of their meshes, and Gaussian splat datasets that preview in the file viewer.
C
merve a relayé @calebfahlgren
Caleb @calebfahlgren · 09/11 19:03
Hugging Face is now the best place to host and share 3D assets 🧊🤗 Drop in .glb, .ply, .stl, or splat files and get a 3D gallery automatically. There are thousands of 3D models already in public datasets! 350+ datasets to explore: Docs:
huggingface.co 3D datasets in the Dataset Viewer - a cfahlgren1 Collection Datasets whose Dataset Viewer shows 3D previews of their meshes, and Gaussian splat datasets that preview in the file viewer.
Nous Research @NousResearch · 09/11 18:55
Hermes Cloud agents now open in a browser tab with the exact same interface as Hermes Desktop. To launch a personal agent, pick a name and a model in Nous Portal. We host it for you and it runs 24/7 in its own container, so it can handle scheduled jobs
Nous Portal Hermes Cloud — Always-On AI Agent Hosting Hermes Cloud runs your Hermes Agent 24/7 — always-on autonomous agent with persistent memory and growing capabilities. Deploy in minutes, no server setup.
Ronald van Loon @Ronald_vanLoon · 09/11 18:53
Increase your personal productivity in an #AI-enabled workspace by Kara Baskin @MITSloan Learn more: #ArtificialIntelligence #FutureOfWork #EmergingTech #Innovation #Tech
MIT Sloan Increase your personal productivity in an AI-enabled workspace | MIT Sloan To gain maximum productivity from AI, MIT Sloan’s Robert Pozen advises senior leaders to use AI as a co-worker, focus AI on the final product, and delegate work to AI without abandoning judgment.
Ronald van Loon a relayé @Ronald_vanLoon
Ronald van Loon @Ronald_vanLoon · 09/11 18:47
Solid-state cooling advances toward practical use, but challenges remain by Martin Heidelberger @TechXplore_com Learn more: #Sustainability #CleanTech #Tech #Technology
Tech Xplore Solid-state cooling advances toward practical use, but challenges remain As temperatures rise, the need for cooling is growing worldwide. Conventional cooling systems consume large amounts of electricity and use refrigerants with a high global warming potential. Caloric cooling could offer a sustainable alternative by using solid-state materials instead of conventional refrigerants. Researchers at Karlsruhe Institute of Technology (KIT), together with international partners, have analyzed the state of development of such systems.
Yusuf Mehdi @yusuf_i_mehdi · 09/11 18:15
Can’t wait to put these new Edge features to use especially multi-tab comparison, which recognizes when you’re looking at gifts across tabs and now offers to compare them side by side. See what’s new in Edge, just in time for holiday shopping:
Microsoft Edge Blog Shop faster and save money with built-in tools in Microsoft Edge It's 11 p.m. Fifteen tabs are open, the reviews don't agree, and you've checked the price three times. Then, just when you think you're done, a promo code box appears with a code you don't have. So, you keep shopping, long past the point you wanted t
Hugging Face a relayé @ariG23498
Aritra 🤗 @ariG23498 · 09/11 17:54
This week we were blessed by: 1. @Nouamanetazi dropped MoE training handbook: 2. @yb2698 started training harness adapters: 3. @joelniklaus released the harness leaderboard: Have a lovely weekend y'all. 🤗
huggingface.co Harness adapters: Safe RL for efficient agentic development A Blog post by harness-hub on Hugging Face
huggingface.co The Ultimate MoE Training Handbook - a Hugging Face Space by huggingface A hands-on guide to training Mixture-of-Experts models
huggingface.co Harness Leaderboard - a Hugging Face Space by joelniklaus Which coding-agent harness should you use for your model?
F
Pierre Sautarel a relayé @F_Desouche
Fdesouche.com est une revue de presse @F_Desouche · 09/11 16:58
L’Inalco a fermé son Pôle les 9 et 10 novembre. Des captures attribuées à un compte de mobilisation étudiante montrent des appels visant les « gwers », les « fachos » et les « riches ».
Fdesouche Paris : l’Inalco fermé les 9 et 10 novembre ; des publications attribuées à un compte de mobilisation étudiante appellent à « tuer les gwers » et à « manger les fachos et les riches » Fdesouche
Visual Studio Code @code · 09/11 16:05
🐘 Database performance issues? Start investigating in VS Code! Use PostgreSQL performance dashboards and Azure Advisor recommendations to identify issues, review possible fixes, and verify the results. 🎬 PostgreSQL Like a Pro: VS Code Edition #vscode
aka.ms https://aka.ms/postgresql-like-a-pro-vscode
Ronald van Loon @Ronald_vanLoon · 09/11 15:45
Some #AI bots work harder to defy stereotypes, study finds by Greg Bruno @TechXplore_com Learn more: #LLM #GenAI #ArtificialIntelligence #MachineLearning
Tech Xplore Some AI bots work harder to defy stereotypes, study finds Large language models (LLMs), the technology that powers tools like Gemini and ChatGPT, are more nuanced than ever. But that doesn't mean they aren't biased. Some models are more prejudiced than others, a Rutgers University researcher has found.
Geek Lite @QingQ77 · 09/11 15:14
three.js 和网页项目找免费低模素材要在素材站逐个翻找下载，这个仓库把 的 3,086 个模型和 28 套主题 kit 整理成可检索的索引，并附带 CLI 与 catalog.json。
GitHub GitHub - lucas-martinic/polyfork-catalog: Free low-poly 3D models for three.js and the web: 3,156 GLB models and 29 themed kits, CC BY 4.0. The public index of polyfork.dev. Free low-poly 3D models for three.js and the web: 3,156 GLB models and 29 themed kits, CC BY 4.0. The public index of polyfork.dev. - lucas-martinic/polyfork-catalog
Polyfork Polyfork · 3D assets you can remix Low-poly 3D assets for three.js, Unity, Unreal and Godot. Every asset is a small program with knobs: recolor it, reshape it, export GLB, FBX or USDZ.
Geek Lite @QingQ77 · 09/11 14:13
界面和交互类任务的好坏要靠人体验，单个 Agent 只给一个方案，想对比就得自己开两个会话、各自装依赖搭环境 AgentCompare 在 DeepSeek Harness 的主对话里同时跑两版，先盲选再揭晓模型。
GitHub GitHub - lijiaqi-AI-Product-Manager/dsh-agent-compare: 同一需求，两版真跑，体验后审查采用。Two live builds for one task in DeepSeek Harness. 同一需求，两版真跑，体验后审查采用。Two live builds for one task in DeepSeek Harness. - lijiaqi-AI-Product-Manager/dsh-agent-compare
Lumera Protocol @lumera · 09/11 14:00
Lumera x @OpenGradient OpenGradient proves an AI model ran as claimed. Lumera proves which agent ran it. We're partnering because AI agents should be able to prove both. Read the full announcement 👇
Lumera Lumera x OpenGradient: Verified Identity for Verifiable AI Lumera Protocol is partnering with OpenGradient, the team behind the OpenGradient Network, around a shared belief: AI agents should be able to prove who they are and what they computed. What OpenGradient built OpenGradient runs a decentralized network for full-stack verifiable AI: on-chain model hos
Geek Lite @QingQ77 · 09/11 13:43
读英文网页时想在旁边挂一份中文，这个扩展把译文插到原段落底下，原文照留，切回原样不用重跑一遍翻译。 ContextFlow Translate 是 Chrome / Edge 的 Manifest V3 扩展，在文章类网页上生成双语对照，原文保留，弹窗里可以来回切换双语与原文而不重新翻译。
GitHub GitHub - muran169633/contextflow-translate: Context-aware bilingual webpage translation extension with your own AI provider Context-aware bilingual webpage translation extension with your own AI provider - muran169633/contextflow-translate
P
Bruno Jeudy a relayé @PierreLepel
Pierre Lepelletier @PierreLepel · 09/11 13:41
Dans un texte publié par la @TribuneDimanche, Guillaume Lacroix propose un « conclave républicain » de « la droite de gouvernement jusqu’à la gauche de gouvernement » pour une « candidature de rassemblement » qui s’engagerait à ne faire qu’un seul mandat
latribune.fr OPINION. « Opération Cincinnatus », par Guillaume Lacroix, président du Parti radical de gauche Le président du Parti radical de gauche veut voir dans l'histoire d'un ancien consul de Rome une des clés pour contrer l'extrême droite en 2027.
Ronald van Loon @Ronald_vanLoon · 09/11 12:56
Two flavors of superconductivity: Material has right ingredients for a twist by Kate Blackwood @Cornell Learn more: #QuantumComputing #EmergingTech #Innovation #Technology #Tech
Cornell Chronicle Two flavors of superconductivity: Material has right ingredients for a twist | Cornell Chronicle Cornell researchers have demonstrated that uranium ditelluride can exhibit both flavors of superconductivity at once, setting it apart as a leading candidate to be a topological superconductor.
Geek Lite @QingQ77 · 09/11 11:42
Home Assistant 原生天气卡只显示静态图标和数字，这张卡片把 weather 实体变成会动的天空，并可选让 AI 写一句天气提醒。
GitHub GitHub - alexsto1an/ha-ani-weather-card: Live weather, on your dashboard: a Home Assistant card that turns your forecast into a living sky, with a sun tracker to match. Live weather, on your dashboard: a Home Assistant card that turns your forecast into a living sky, with a sun tracker to match. - alexsto1an/ha-ani-weather-card
Ken Sakurada @sakuDken · 09/11 11:18
Teslaの「カメラ単体」を否定する理由　Waymoが明かした自動運転の壁と自社製プロセッサー (1/4)
ASCII.jp Teslaの「カメラ単体」を否定する理由 Waymoが明かした自動運転の壁と自社製プロセッサー (1/4) Hot ChipsでWaymoは2つの発表をした。WaymoはAlphabet傘下の自律運転プラットフォームを開発する会社であり、2027年に日本交通およびGOと共同で完全無人の自動運転配車サービスを開始することを発表している。
Geek Lite @QingQ77 · 09/11 10:41
让用户在常用 AI 助手里用中文提问，直接查中国校招、实习、社招岗位和招聘日历，拿到岗位要求、来源链接与投递入口。
GitHub GitHub - lijiaqi-AI-Product-Manager/job_radar: 给 AI 助手的求职搜索技能：中国校招、实习、社招与招聘日历。自然语言提问，获取岗位要求、来源链接与投递入口；支持 Skill / MCP。 给 AI 助手的求职搜索技能：中国校招、实习、社招与招聘日历。自然语言提问，获取岗位要求、来源链接与投递入口；支持 Skill / MCP。 - lijiaqi-AI-Product-Manager/job_radar
jianlin.su @Jianlin_S · 09/11 10:30
Revisiting Convergence Results in Convex Optimization (XI) New ideas for scheduled-free learning from
arXiv.org Conic-Lifting for Learning-rate-free Convex Optimization A central topic in online and stochastic optimization is to design algorithms that do not require tuning of hyperparameters. Successful strategies in online learning include AdaGrad-style learning rates, Conic Blackwell algorithms, and parameter-free algorithms. However, these classes require different, often difficult, analyses. We develop a unifying family of Online Convex Optimization (OCO) algorithms that require no user-tuned learning rate on convex domains that may be unbounded. The construction lifts the domain to a cone by adding one scalar coordinate, an idea going back to Blackwell (
kexue.fm 让炼丹更科学一些（十一）：无调度学习率新思路 前面的多篇文章中，我们已经表明了学习率调度对终点收敛的重要性。与此同时，“无调度（Schedule-Free）学习率”一直也是训练人的理想追求：不预设学习率曲线，也不预设训练步数，训练过程随时终...
V
Louis de Raguenel a relayé @vtremolet
Vincent Tremolet de Villers @vtremolet · 09/11 10:05
« Notre métier consiste à donner les faits, qu’ils plaisent ou non ». @vincentroux88 sur livre "Fake" @EditionsduCERF. Essai à lire et à garder précieusement durant ces mois de campagne présidentielle...
Le Figaro Vincent Roux : « Notre métier consiste à donner les faits, qu’ils plaisent ou non » ENTRETIEN. - Dans Fake (Cerf), le journaliste Vincent Roux explore les ressorts de la désinformation et de la défiance envers les médias. Il invite sa profession à interroger ses propres biais et met en garde contre la tentation de restreindre la liberté d’expression au nom de la vérité.
Ronald van Loon @Ronald_vanLoon · 09/11 09:48
Meta quietly launches an #AI fashion app called Style Sketch by @AiishaMalik1 @TechCrunch Learn more: #GenerativeAI #ArtificialIntelligence #MachineLearning
TechCrunch Meta quietly launches an AI fashion app called Style Sketch | TechCrunch The app connects to Instagram to build a virtual closet from photos you’ve already shared, or lets you add items manually by snapping a photo, recording a quick video, or uploading images from your camera roll.
Geek Lite @QingQ77 · 09/11 09:41
把 Agentero 的本地文献库、笔记和 PDF 阅读器接进 DeepSeek Harness，让 agent 在同一窗口里查论文、引出处、写笔记，资料留在本机。
GitHub GitHub - poco-ai/agentero-dsh-plugin: 给 DeepSeek Harness 用的科研文献库插件：搜论文、带出处、审批入库、把笔记线索交给同会话其他插件。Academic research library for DSH — search, cite, approve imports, hand off notes. 给 DeepSeek Harness 用的科研文献库插件：搜论文、带出处、审批入库、把笔记线索交给同会话其他插件。Academic research library for DSH — search, cite, approve imports, hand off notes. - poco-ai/agentero-dsh-plugin
Ronald van Loon a relayé @Ronald_vanLoon
Ronald van Loon @Ronald_vanLoon · 09/11 09:05
Targeted #Data selection could roughly triple #AI fine-tuning speed by The Korea Advanced Institute of Science and Technology (KAIST) @TechXplore_com Learn more: #ArtificialIntelligence #MachineLearning #DeepLearning #LLM
Tech Xplore Targeted data selection could roughly triple AI fine-tuning speed Can AI fine-tuning that once took 10 hours be completed in just three hours? KAIST researchers have developed a technology that selects the most useful training data and distributes it evenly across multiple GPUs.
M
Managetech inc. @managetech_inc · 09/11 09:00
AI社員のSNS運用 Day 4。次の一手は、語り口を変えずに同じ形式で続けることです。昨日のサイト訪問は 1 人、フォロワーは 91 人。続きはこちら →
mierucrew.com AI社員に自社のSNS運用を任せる、90日間。 MIERU CREW の AI 社員 3 体が、マネージテック自身の X と YouTube を 90 日間運用します。設定内容・ルール・費用・経過をこのページで公開します。
Victor @victor_explore · 09/11 08:00
the agent that wrote your spec is the worst one to review it. it already believes its own assumptions google open-sourced a fix: one agent with full repo context drafts the spec, then fresh subagents with zero chat history review it in 5 stages
GitHub GitHub - google/elephant-goldfish-model: Open-source plugin for designing and reviewing software with agents Open-source plugin for designing and reviewing software with agents - google/elephant-goldfish-model
Ronald van Loon @Ronald_vanLoon · 09/11 07:50
How to stay smart in the age of #AI: the science of critical thinking by Helen Pearson @Nature Learn more: #GenerativeAI #ArtificialIntelligence #CriticalThinking #FutureOfLearning
nature.com Client Challenge
Hanako a relayé @hanakoxbt
Hanako @hanakoxbt · 09/11 07:49
Before you blame Opus 5.5, try these 15 official Claude Code commands 15 built-ins most people never type: less babysitting → parallel agents → faster loops the full cheat sheet ↓ 01 /goal → keep working until a condition is true 02 /batch →
Claude Code Docs Keep Claude working toward a goal - Claude Code Docs Set a completion condition with /goal and Claude keeps working until it's met, a model judges it impossible, or an error you have to fix clears the goal.
Geek Lite @QingQ77 · 09/11 07:37
给每个项目目录单独开一台只挂载该目录的 Linux 虚拟机，让编码 agent 在隔离环境里跑，不碰宿主机。 一台 VM 管一个项目，devvm 用 Go 写成，底层借 Lima 在 macOS 和 Linux 上起 Ubuntu。
GitHub GitHub - devvmhq/devvm: One VM for every project. An isolated sandbox with your project’s toolchain, ready for you or a coding agent to start coding One VM for every project. An isolated sandbox with your project’s toolchain, ready for you or a coding agent to start coding - devvmhq/devvm
M
Managetech inc. @managetech_inc · 09/11 07:00
AI社員のSNS運用 Day 4。うまくいかなかったこと。数字の集計が揃わず、「集計待ち」と書く日が続いています。待つ間に推測で埋めないことを優先しています。フォロワーの7日増減は 集計待ち。設定内容と経過はサイトで →
mierucrew.com AI社員に自社のSNS運用を任せる、90日間。 MIERU CREW の AI 社員 3 体が、マネージテック自身の X と YouTube を 90 日間運用します。設定内容・ルール・費用・経過をこのページで公開します。
hardmaru a relayé @SakanaAILabs
Sakana AI @SakanaAILabs · 09/11 06:52
【Sakana AI 採用のお知らせ】 モバイルアプリをはじめとする画期的なAIプロダクトを共に創り上げる仲間を探しています！現在、Productチーム・GTMチームで積極採用中です。 ▼私たちのAIを世界中のユーザーに届ける挑戦に、ぜひご参加ください。 🐟
sakana.ai Sakana AI Career Opportunities
M
Managetech inc. @managetech_inc · 09/11 05:30
AI社員のSNS運用 Day 4。今日やってみたのは、変化→根拠→学びの順で書くことです。フォロワーは 91 人、昨日のサイト訪問は 1 人。続きと設定はこちら →
mierucrew.com AI社員に自社のSNS運用を任せる、90日間。 MIERU CREW の AI 社員 3 体が、マネージテック自身の X と YouTube を 90 日間運用します。設定内容・ルール・費用・経過をこのページで公開します。
Geek Lite @QingQ77 · 09/11 05:15
让 AI agent 直接接管你已登录的真实浏览器标签页干活，遇到登录、2FA、支付这类必须人来的步骤再把控制权交还给你。 browse 由一个 MCP 服务器和一个 Chromium 扩展组成，让 agent 在用户日常使用的浏览器里开标签、操作页面，复用现成的登录态和扩展。
GitHub GitHub - anomalyco/browse: Let your AI agents use your real browser: an MCP server and a Chromium extension Let your AI agents use your real browser: an MCP server and a Chromium extension - anomalyco/browse
Ronald van Loon a relayé @Ronald_vanLoon
Ronald van Loon @Ronald_vanLoon · 09/11 04:22
MIT engineers create nanoparticles for delivering personalized cancer vaccines by Anne Trafton @MIT Learn more: #MedTech #HealthTech #Tech #TechForGood
MIT News | Massachusetts Institute of Technology MIT engineers create nanoparticles for delivering personalized cancer vaccines A nanoscale particle developed at MIT holds potential as a customizable delivery vehicle for personalized cancer vaccines. The particle could be adapted to carry many different antigens, to potentially treat individual patients’ tumors.
Kling AI @Kling_ai · 09/11 04:11
Through a complete energy drink ad, leedain_leader_in explores how Kling 4.0 handles product details, character consistency, and natural dialogue, shifting more of the creative process from editing to directing.
X (formerly Twitter) X - The Everything App
M
Managetech inc. @managetech_inc · 09/11 04:00
AI社員のSNS運用 Day 4。学びを一つ。投稿がない日も、台帳と数字を先に照合してから発信する。数字が合わないまま話すと、後で説明がつかなくなるからです。直近7日のサイト訪問は 1 人。費用と設定はこちら →
mierucrew.com AI社員に自社のSNS運用を任せる、90日間。 MIERU CREW の AI 社員 3 体が、マネージテック自身の X と YouTube を 90 日間運用します。設定内容・ルール・費用・経過をこのページで公開します。
Ronald van Loon @Ronald_vanLoon · 09/11 03:58
Will #AI really kill us all? The science behind the hype by Elizabeth Gibney @Nature Learn more: #ArtificialIntelligence #AISafety #ResponsibleAI
nature.com Client Challenge
Sakana AI a relayé @hardmaru
hardmaru @hardmaru · 09/11 03:31
We are building our first mobile app! 🐟 This is a new direction for Sakana AI, and we are looking for early beta testers to help shape it before we launch. If you are in Japan 🇯🇵 and want to try something early, sign up: 🐡
docs.google.com Sakana Mobile App Beta Test We collect info tied to your app store account for mobile app testfilight distribution. Personal information submitted will be used solely for distributing the Sakana Mobile App and will be promptly...
hardmaru a cité @SakanaAILabs
hardmaru @hardmaru · 09/11 03:29
We are building our first mobile app! 🐟 This is a new direction for Sakana AI, and we are looking for early beta testers to help shape it before we launch. If you are in Japan 🇯🇵 and want to try something early, sign up: 🐡
docs.google.com Sakana Mobile App Beta Test We collect info tied to your app store account for mobile app testfilight distribution. Personal information submitted will be used solely for distributing the Sakana Mobile App and will be promptly...
Geek Lite @QingQ77 · 09/11 03:12
在自己机器上跑一套完整的个人财务账本，并让 AI agent 通过受控的工具调用真正读改账目和投资数据。 账本、预算、投资、退休规划分成四块。账本记多币种账户，转账、拆分、对账都在里面，导入账单要先过一遍复核。
GitHub GitHub - open-net-worth-ai/open-net-worth: Self-hosted alternative to Claude and ChatGPT Personal Finance. An AI agent that manages your budget and investments, and can place trades. Self-hosted alternative to Claude and ChatGPT Personal Finance. An AI agent that manages your budget and investments, and can place trades. - open-net-worth-ai/open-net-worth
Ronald van Loon a relayé @Ronald_vanLoon
Ronald van Loon @Ronald_vanLoon · 09/11 02:25
Tracing the online networks behind #AI-generated nude images by Harrem Monkhorst @TechXplore_com Learn more: #GenerativeAI #ArtificialIntelligence #MachineLearning #DeepLearning
Tech Xplore Tracing the online networks behind AI-generated nude images As Domestic Violence Awareness Month comes to a close this week, one University of Florida researcher is examining the roots of nudification apps, online platforms that use AI to turn images of clothed people—easily and convincingly—into nude images.
M
Managetech inc. @managetech_inc · 09/11 02:00
AI社員のSNS運用 Day 4。今日の変化は、フォロワーが 91 人になったことです。前日から減りました。昨日のサイト訪問は 1 人、直近7日は 1 人。数字は速報で、集計待ちの値は後で確定させます。設定と経過はこちら →
mierucrew.com AI社員に自社のSNS運用を任せる、90日間。 MIERU CREW の AI 社員 3 体が、マネージテック自身の X と YouTube を 90 日間運用します。設定内容・ルール・費用・経過をこのページで公開します。
Geek Lite @QingQ77 · 09/11 01:02
不想在服务器上再装一套面板的人，可以用它把终端、文件、服务、Docker、日志这些日常操作塞进一个桌面客户端，连服务器走 SSH 就够。 Davish 用 Go + Wails 写后端、Svelte 5 写前端，做成 macOS、Windows、Linux 三平台的桌面应用，通过普通 SSH 和 SFTP
GitHub GitHub - 2hjaito/davish: Your server, as a desktop. A lightweight SSH client with an OS-like UI. Nothing to install on the server. Your server, as a desktop. A lightweight SSH client with an OS-like UI. Nothing to install on the server. - 2hjaito/davish
Geek Lite @QingQ77 · 09/11 00:12
配享太庙，给国内版小米电视去掉广告和预装、换掉开机桌面，全程免 root 且每一步都能还原。
GitHub GitHub - wangrongding/tv-fresh: tv-fresh 电视焕新：小米电视(MIUI TV)去广告/清理预装/换桌面/装代理与影视仓的 Agent Skill——免root可逆，含固件逆向实录与一键脚本 tv-fresh 电视焕新：小米电视(MIUI TV)去广告/清理预装/换桌面/装代理与影视仓的 Agent Skill——免root可逆，含固件逆向实录与一键脚本 - wangrongding/tv-fresh
Deep Learning Weekly @dl_weekly · 08/11 23:01
🤖 From this week's issue: Cursor warns that malicious Claude Code builds 2.1.312 and 2.1.313 sat in its marketplace for eleven hours, attempting to install a persistent SSH backdoor keyed as svc-telemetry.
Cursor Avis de sécurité — version malveillante d’Anthropic.claude-code (2.1.312 et 2.1.313) identifiée dans la Marketplace d’extensions VS Code de Cursor · Cursor Nous avons identifié aujourd’hui un incident de sécurité touchant une extension spécifique de la Marketplace d’extensions VS Code de Cursor : Anthropic.claude-code version 2.1.312 et 2.1.313 pour Windows et MacOS.
H
Jim Harris a relayé @YuHelenYu
Helen Yu @YuHelenYu · 08/11 22:18
Can AI help travel brands earn loyalty again? I hosted Charlie Sultan of @SAPConcur and Evan Konwiser of @amexgbt on CXO Spice to discuss Complete and the changing travel experience. Watch: #CXOSpice #BusinessTravel #AI #SAPConcurPartner
YouTube Complete Travel Experience (SAP) Episode 168 In this episode of CXO Spice, I hosted Charlie Sultan of @SAP Concur and Evan Konwiser of @American Express Global Business Travel. We discuss Complete, thei...
Hanako a cité @hanakoxbt
Hanako @hanakoxbt · 08/11 22:10
Before you blame Opus 5.5, try these 15 official Claude Code commands 15 built-ins most people never type: less babysitting → parallel agents → faster loops the full cheat sheet ↓ 01 /goal → keep working until a condition is true 02 /batch →
Claude Code Docs Keep Claude working toward a goal - Claude Code Docs Set a completion condition with /goal and Claude keeps working until it's met, a model judges it impossible, or an error you have to fix clears the goal.
S
Kosta Derpanis a relayé @nnsriram97
Sriram Narayanan ✈️CoRL2026 @nnsriram97 · 08/11 21:48
We’re presenting VisionSIM as a spotlight at @corl_conf #CoRL2026! VisionSIM is a Blender-based multi-sensor simulator for RGB, thermal, event cameras & more—with realistic sensor effects and pixel-perfect ground truth. Paper + code: 🧵 1/n
VisionSIM VisionSIM — perception is complementary One world, many measurements. A simulated scene read by seven sensor models, with pixel-perfect ground truth that never moves.
D
Stanford NLP Group a relayé @DSPyOSS
DSPy @DSPyOSS · 08/11 21:43
Nice section in this AWS post on ShipStation Global’s AI IT help desk. A dspy.RLM module holds their entire Slack history as a variable, and mines it into a knowledge base of resolved issues.
Amazon Web Services How ShipStation Global built an AI IT help desk on Amazon Bedrock AgentCore | Amazon Web Services ShipStation Global's engineering team built Deckhand, an AI IT help desk agent for Slack on Amazon Bedrock AgentCore. It answers routine IT questions from a curated knowledge base, reads error-message screenshots, and files Jira tickets. In its first three months it resolved 97.5% of channel queries with a 96% positive feedback rating.
T
Stanford NLP Group a relayé @thomaschen__
Thomas Chen @thomaschen__ · 08/11 21:43
How can an LLM self-improve at theorem proving? Inspired by STP (Dong & Ma, 2025), we propose a theoretical framework for understanding prover-conjecturer self-play algorithms.. NeurIPS 2026, Sydney Poster Session 2, Tue 5-8pm @ Hall 1-4. 1/8
arXiv.org A Theoretical Framework for Self-Play Theorem Proving Algorithms Self-play, a type of training algorithm that enables a model to self-improve, has recently shown promising empirical results in the context of formal theorem proving using Large Language Models (LLMs). (Dong & Ma, 2025) instantiate self-play with two cooperating agents: a prover, which proves theorems, and a conjecturer, which generates new theorems as a curriculum to the prover. In this paper, we provide a theoretical framework for understanding the self-improvement capabilities of self-play algorithms for theorem proving. First, we formalize the set of theorems as a graph, with nodes as theo
Ethan Perez a relayé @nytimes
The New York Times @nytimes · 08/11 21:33
Breaking News: Nvidia opened an investigation into Jensen Huang, its CEO, after he acknowledged to the company’s board that he had been intimate with a business executive who U.S. intelligence determined to be a Chinese spy.
nytimes.com Nvidia Opens Investigation Into Jensen Huang’s Relationship With Chinese Spy Mr. Huang, the company’s chief executive, acknowledged on Sunday that he was intimate with a business executive who U.S. officials had determined was working for Chinese intelligence.
A
Machine Learning FLX a relayé @amananimates
Aman Singh @amananimates · 08/11 21:01
If you're a UI/UX designer, bookmark this. 1.⁠ ⁠ - 600+ design Prompts and 200+ components with full source code for 3D and scroll-driven sites. 2.⁠ ⁠ - 117 animated shaders in 11 styles 3.⁠ ⁠ - better
footer.design Footer — The only footer gallery on earth. Footer is a curated gallery of the top website footer inspiration on earth. Find the footers you need and sort by type and style.
Scroll Shaders Scroll Shaders — WebGL shader backgrounds & components A library of seamless-loop WebGL shaders and shader components. Copy the React or HTML code, or download the MP4. Mesh, grain, dither, liquid metal, glass and more.
Scrolltide Scrolltide — Cinematic website prompts & templates Copy a prompt, paste it into your AI, and launch a site that actually moves.
Victoria X Lin a relayé @soumithchintala
Soumith Chintala @soumithchintala · 08/11 20:57
Inkling 2 Small on OpenCode for free for a limited time, to help us improve our models further:
OpenCode Utilisation, coût et rang de Inkling 2 Small | OpenCode Data Consultez les données d'utilisation OpenCode de Inkling 2 Small, avec le volume de tokens, le rang hebdomadaire, la répartition des tokens, les coûts, le taux de cache, les sessions, les répartitions géographiques et les modèles proches.
Leandro von Werra a relayé @ariG23498
Aritra 🤗 @ariG23498 · 08/11 20:53
This week we were blessed by: 1. @Nouamanetazi dropped MoE training handbook: 2. @yb2698 started training harness adapters: 3. @joelniklaus released the harness leaderboard: Have a lovely weekend y'all. 🤗
huggingface.co Harness adapters: Safe RL for efficient agentic development A Blog post by harness-hub on Hugging Face
huggingface.co The Ultimate MoE Training Handbook - a Hugging Face Space by huggingface A hands-on guide to training Mixture-of-Experts models
huggingface.co Harness Leaderboard - a Hugging Face Space by joelniklaus Which coding-agent harness should you use for your model?
A
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a relayé @norpadon
Artur Chakhvadze @norpadon · 08/11 20:28
I am a bit pissed that more and more papers are coming out which cite EVERYONE other than us, despite our paper coming out at the same time as DSpark and significantly earlier than DFlash2 Idk, maybe the phone-friendly format makes it look less “serious”
arXiv.org Trees from Marginals: Autoregressive drafting with factorized priors Speculative decoding greatly increases the interactivity of autoregressive language models by trading off computation for extra tokens generated in a single forward pass. Factorized draft models are especially efficient because they predict future-token marginals in parallel, but their independence assumption causes acceptance rates to degrade sharply as the speculative budget grows. We analyze this limitation and introduce Weaver, a lightweight autoregressive adapter that constructs proposal trees from the top-K marginals of a factorized drafter. Weaver restores conditional dependencies betwe
R
Eric Hartford a relayé @imryven
Ryven @imryven · 08/11 20:28
This is f*cking insane, the CEO of Y Combinator open-sourced his own second brain and most people still don’t know it exists 8 second brain repos you haven’t seen yet: -> gbrain (by garrytan) the second brain of the YC CEO, every fact keeps its source
GitHub GitHub - garrytan/gbrain: Garry's Opinionated OpenClaw/Hermes Agent Brain Garry's Opinionated OpenClaw/Hermes Agent Brain. Contribute to garrytan/gbrain development by creating an account on GitHub.
G
Luke Hoban a relayé @github
GitHub @github · 08/11 20:24
Through internal testing, GitHub’s new Git storage architecture supported 5,000 concurrent agents working on a repo, sending 7,200 pushes per second, all while serving read traffic with consistently low latency.
X (formerly Twitter) X - The Everything App
T
Stanford AI Lab a relayé @thomaschen__
Thomas Chen @thomaschen__ · 08/11 20:21
How can an LLM self-improve at theorem proving? Inspired by STP (Dong & Ma, 2025), we propose a theoretical framework for understanding prover-conjecturer self-play algorithms.. NeurIPS 2026, Sydney Poster Session 2, Tue 5-8pm @ Hall 1-4. 1/8
arXiv.org A Theoretical Framework for Self-Play Theorem Proving Algorithms Self-play, a type of training algorithm that enables a model to self-improve, has recently shown promising empirical results in the context of formal theorem proving using Large Language Models (LLMs). (Dong & Ma, 2025) instantiate self-play with two cooperating agents: a prover, which proves theorems, and a conjecturer, which generates new theorems as a curriculum to the prover. In this paper, we provide a theoretical framework for understanding the self-improvement capabilities of self-play algorithms for theorem proving. First, we formalize the set of theorems as a graph, with nodes as theo
S
Robert Scoble a relayé @theSamPadilla
Sam Padilla ✈️ CoRL 2026 @theSamPadilla · 08/11 19:51
As more companies move into robotics and interest is at an all-time high, it’s a good time to reflect on where the data industry stands ahead of 2027. TL;DR: Mono is dead; almost nobody has stereo off the shelf, and tactile is the new frontier.
X (formerly Twitter) X - The Everything App
L
Bruno Jeudy a relayé @LVigogne
Ludovic Vigogne @LVigogne · 08/11 19:29
Mercredi, Emmanuel Macron a vu une douzaine de parlementaires Renaissance fidèles. Ce qu'il leur a dit sur la fin de son mandat et 2027 est à lire dans La @TribuneDimanche
latribune.fr Coalition de la droite et du centre, Le Pen, Mélenchon... Ce qu'Emmanuel Macron livre en coulisses aux parlementaires sur 2027 Mercredi, Emmanuel Macron a reçu une douzaine de parlementaires Renaissance. Avec eux, il a notamment évoqué la présidentielle. Récit.
P
Bruno Jeudy a relayé @PierreLepel
Pierre Lepelletier @PierreLepel · 08/11 19:29
Raphaël Glucksmann à la @TribuneDimanche : « Rien ne m’empêchera de faire campagne, de montrer qu’un autre chemin existe et que nous sommes les seuls à pouvoir gagner les élections »
latribune.fr Le plan de Glucksmann pour empêcher « l’affrontement mortifère » entre Le Pen et Mélenchon ENTRETIEN EXCLUSIF — Le candidat des socialistes à l’élection présidentielle dénonce le « tango » de Marine Le Pen et Jean-Luc Mélenchon et dévoile son plan de route pour 2027.
Bruno Jeudy a relayé @TribuneDimanche
La Tribune Dimanche @TribuneDimanche · 08/11 19:29
Avec sa tournée en Amérique latine, Léon XIV revient aux sources : Uruguay, Argentine puis le Pérou, où il fut évêque ⤵ 🔗 #LéonXIV #Pape
latribune.fr Avec sa tournée en Amérique latine, Léon XIV revient aux sources Le pape a entamé samedi 6 novembre une grande tournée en Amérique latine. Elle va passer par l’Argentine, le pays de son prédécesseur, mais aussi le Pérou, où il fut évêque.
D
Omar Khattab a relayé @DSPyOSS
DSPy @DSPyOSS · 08/11 19:28
Nice section in this AWS post on ShipStation Global’s AI IT help desk. A dspy.RLM module holds their entire Slack history as a variable, and mines it into a knowledge base of resolved issues.
Amazon Web Services How ShipStation Global built an AI IT help desk on Amazon Bedrock AgentCore | Amazon Web Services ShipStation Global's engineering team built Deckhand, an AI IT help desk agent for Slack on Amazon Bedrock AgentCore. It answers routine IT questions from a curated knowledge base, reads error-message screenshots, and files Jira tickets. In its first three months it resolved 97.5% of channel queries with a 96% positive feedback rating.
Francis Engelmann a relayé @FrancisEngelman
Francis Engelmann @FrancisEngelman · 08/11 19:23
Need DINO for 3D/volumetric data? Check out: 🍷VINO: Revisiting 3D Representation Learning with Volume Transformers 📝 🏠 Great collab w/ @adr_kruse @KadirYilmaz_CV @dcdegeus @BastianLeibe @RWTHVisionLab @USI_INF @TUeindhoven1
adriankruse.github.io VINO: Revisiting 3D Representation Learning with Volume Transformers
arXiv.org VINO: Revisiting 3D Representation Learning with Volume Transformers Self-supervised learning has given rise to vision foundation models that learn general-purpose image representations transferable across downstream tasks, yet representation learning for 3D scenes has not achieved comparable success. Motivated by this, we identify key limitations of current 3D representation learning approaches and revisit their core design choices. We first replace the commonly used hierarchical PTv3 backbone with the Volume Transformer, a ViT-like architecture for 3D scenes. We then redesign two core objectives of modern approaches built around student-teacher networks: matc
S
Robert Scoble a relayé @ScottLeimroth
Scott R. Leimroth @ScottLeimroth · 08/11 19:15
I'm deaf, so I never hear the fridge alarm, an oven timer or a knock at the door. I built Signal Dog for @NousResearch's Hermes Agent Plugin Hackathon: a plugin that listens through the microphone and tells you in chat. No audio kept.
GitHub GitHub - scottleimroth/signal-dog: Hermes Agent plugin: listens through the microphone for door knocks, appliance beeps and alarms and tells you in chat. Built by a deaf developer. Not a certified smoke alarm. Hermes Agent plugin: listens through the microphone for door knocks, appliance beeps and alarms and tells you in chat. Built by a deaf developer. Not a certified smoke alarm. - scottleimroth/signal...
S
Teknium 🪽 a relayé @ScottLeimroth
Scott R. Leimroth @ScottLeimroth · 08/11 19:05
I'm deaf, so I never hear the fridge alarm, an oven timer or a knock at the door. I built Signal Dog for @NousResearch's Hermes Agent Plugin Hackathon: a plugin that listens through the microphone and tells you in chat. No audio kept.
GitHub GitHub - scottleimroth/signal-dog: Hermes Agent plugin: listens through the microphone for door knocks, appliance beeps and alarms and tells you in chat. Built by a deaf developer. Not a certified smoke alarm. Hermes Agent plugin: listens through the microphone for door knocks, appliance beeps and alarms and tells you in chat. Built by a deaf developer. Not a certified smoke alarm. - scottleimroth/signal...
Ronald van Loon @Ronald_vanLoon · 08/11 18:53
Vibrating mattress helps clear Alzheimer’s protein from the brain by Carissa Wong @newscientist Learn more: #MedTech #HealthTech #Tech #TechForGood
New Scientist Vibrating mattress helps clear Alzheimer’s protein from the brain | New Scientist Lying on a vibrating mattress for 40 minutes, combined with taking two pills that release nitric oxide into the body, enhanced the clearance of an Alzheimer's-related protein via the brain's waste-disposal system
A
F. Güney a relayé @akanazawa
Angjoo Kanazawa @akanazawa · 08/11 18:51
I had my second baby ~6 months ago 😊 missing CoRL to send time with my family 🫶🏻 but here's a quick look at our lab's recent work, an extended version of last week's BARS talk! My students are all at CoRL, please say hi! Enjoy👢🦿:
YouTube Learning to Act from Observation and Experience Recent talk I did at BARS 2026, a quick snapshot of what my lab has been working on! Please find the works here: https://people.eecs.berkeley.edu/~kanazawa/
Eyisha Zyer a relayé @eyishazyer
Eyisha Zyer @eyishazyer · 08/11 18:24
agent A, default web tools: zillow: bot check block indeed: loaded one job posting only reddit: error came "couldn't open" python: loaded normally the model's fine. it just never got the pages.
reddit.com https://reddit.com/
⭕
⭕ AI & Design (Marco) @AIandDesign · 08/11 18:14
I'm looking for a handful of Windows users willing to test my Steam game "It Wants You To Stay". People are reporting crashes I cannot reproduce on my Mac. If you're interested, please DM me and I'll send you a free Steam downloads code.
store.steampowered.com It Wants You To Stay sur Steam The lights are on. Everyone is gone. Descend through eleven shifting liminal worlds in this first-person psychological horror game. Photograph the impossible, follow a missing woman’s VHS recordings, and save your light for what follows you. When the lights fail, stand still.
⭕
⭕ AI & Design (Marco) @AIandDesign · 08/11 17:58
My strangest and also most soothing game "Petriverse" got an update. This is definitely my most experimental work in the game space. Available for Windows, Mac and Linux. Give it a go!
Google Drive Petriverse
Eyisha Zyer a relayé @eyishazyer
Eyisha Zyer @eyishazyer · 08/11 17:58
your agent can have the same internet you do. Oxylabs Web API, $5 free credit to try it: my setup: opus 5.5 in claude code on windows, each link tried 3 times. the 5 links and the prompt are in this thread, so you can run it yourself.
oxylabs.io Oxylabs – High Quality Proxy Service & Web Search API for AI Access fresh, reliable web data at any scale with APIs built for AI and premium proxies. Power agents and applications with 177M+ IPs and robust infrastructure.
Bruno Jeudy a relayé @TribuneDimanche
La Tribune Dimanche @TribuneDimanche · 08/11 17:54
🗳️ « 𝘊’𝘦𝘴𝘵 𝘶𝘯 𝘯𝘢𝘶𝘧𝘳𝘢𝘨𝘦 𝘨𝘦́𝘯𝘦́𝘳𝘢𝘭 » : désaccord primaire pour la droite et le centre. L’enquête de @LVigogne ⤵ 🔗 #Primaire #Présidentielle2027
latribune.fr « C’est un naufrage général » : désaccord primaire pour la droite et le centre Rien ne va plus au sein de la droite et du centre. Les sondages sont en berne. Les coups bas se multiplient. La perspective d’une primaire ne fait qu’augmenter la tension.
DeepLearning.AI @DeepLearningAI · 08/11 17:05
Most agents forget everything between sessions. Hear @richmondalake (@Oracle) Veljko Kovac (@cognee_) & Dennis Pilarinos (Unblocked) talk Memory & Context at AI Dev NYC! 📅Nov 30–Dec 1, 2026 · Chelsea Piers, NYC 🎟 #AIDev #DeepLearningAI #AgentMemory
luma.com AI Dev / New York City: Andrew Ng's AI Developer Conference · Luma FULL AGENDA AI Developer is hosted by Andrew Ng and the DeepLearning.AI team. Join for two days of keynotes and technical sessions led by experts from OpenAI,…
Geek Lite @QingQ77 · 08/11 16:57
在 Windows 和 Linux 上提供能连 iCloud 的原生应用，覆盖备忘录、提醒事项、日历、通讯录、iCloud 云盘和照片六个服务。 主体是一个 Rust 写的引擎，外面套 Tauri 窗口，一套代码编译出 Linux 和 Windows 版本，x86 与 ARM
GitHub GitHub - YashwanthKamireddi/Skykeep: Native iCloud apps for Windows and Linux, built in Rust. Notes, Reminders, Calendar, Contacts, Drive and Photos, offline and in sync with your iPhone. Native iCloud apps for Windows and Linux, built in Rust. Notes, Reminders, Calendar, Contacts, Drive and Photos, offline and in sync with your iPhone. - YashwanthKamireddi/Skykeep
Victor @victor_explore · 08/11 16:37
coding agents say the windows app works. of course they do: your machine already has the runtimes. scott hanselman's sandbox-mcp boots a fresh windows 11, runs the app, then deletes the pc. it caught a missing vc++ redist in 29.8s. 310 stars, MIT
GitHub GitHub - shanselman/sandbox-mcp: Fresh, throwaway Windows machines for testing your apps: an MCP server and CLI that automate Windows Sandbox Fresh, throwaway Windows machines for testing your apps: an MCP server and CLI that automate Windows Sandbox - shanselman/sandbox-mcp
M
Melissa Lara Clissold @melissaclissold · 08/11 16:28
Acil durumlara ve afetlere önceden hazırlanmak önemli. AB ve birçok ülke de bu konuda çağrı yapıyor. İlk 72 saate hazırlanmayı kolaylaştıran, basit bir yapılacaklar listesi sunan bir platform yayınladım. Siz de hazırlığınızı yapın 👇
haziriz.app 72 Saatlik Afet ve Deprem Çantası Listesi — Hazırız 72 saatlik afet ve acil durum hazırlığı: afet ve deprem çantasında neler olmalı? Uluslararası standartlara ve uzman rehberlerine uygun kontrol listesi, kişi başı su ve erzak hesabı, ev ve aile planı.
Andrew Carr 🤸 @andrew_n_carr · 08/11 15:17
Here's a great pairing of papers that have (seemingly) unrelated ideas, that go together nicely 1. Subliminal Learning: Which basically says a "teacher" model with some trait T (such as liking owls or being misaligned) generates a dataset consisting
arXiv.org Subliminal Learning: Language models transmit behavioral traits via hidden signals in data We study subliminal learning, a surprising phenomenon where language models transmit behavioral traits via semantically unrelated data. In our main experiments, a "teacher" model with some trait T (such as liking owls or being misaligned) generates a dataset consisting solely of number sequences. Remarkably, a "student" model trained on this dataset learns T. This occurs even when the data is filtered to remove references to T. We observe the same effect when training on code or reasoning traces generated by the same teacher model. However, we do not observe the effect when the teacher and stu
Deep Learning Weekly @dl_weekly · 08/11 15:01
🤖 From this week's issue: NVIDIA open-weights Nemotron 3.5 Transcribe, a 30-billion-parameter speech model covering 40 transcription languages, 31 translation pairs, hour-long recordings and up to eight speakers.
TechCrunch Voice AI in vogue: Nvidia releases new open transcription model | TechCrunch Nvidia just released a free speech-to-text model that can transcribe up to 40 languages, but it only works on recordings, not live audio, so it’s a big step for meeting and medical transcription tools while still leaving real-time use to rivals like ElevenLabs.
Geek Lite @QingQ77 · 08/11 14:55
在自己 Windows 电脑上跑一个 Live2D 二次元角色，用实时语音和文字聊天、记住你说过的话，语言模型可选 OpenRouter 或本地 Ollama。
GitHub GitHub - zeekk0/OpenAmika: OpenAmika: MekaHime's AI companion Amika, open source. Live2D avatar, real-time voice, long-term memory, OpenRouter or Ollama brain. OpenAmika: MekaHime's AI companion Amika, open source. Live2D avatar, real-time voice, long-term memory, OpenRouter or Ollama brain. - zeekk0/OpenAmika
Geek Lite @QingQ77 · 08/11 13:54
在 DeepSeek Harness 的对话侧边栏里用一句自然语言描述生成可玩的小游戏，自动试玩、替换 AI 美术并导出单个离线 HTML 文件。 DSH Game Studio 是 DeepSeek Harness 的插件，输入一句话描述后，agent 基于两套模板（2D 俯视收集与 three.js 3D
GitHub GitHub - ZSeven-W/dsh-game-studio: Describe a small game in one sentence, play it in the DeepSeek Harness sidebar, swap in AI art, export one HTML file. Describe a small game in one sentence, play it in the DeepSeek Harness sidebar, swap in AI art, export one HTML file. - ZSeven-W/dsh-game-studio
Ronald van Loon @Ronald_vanLoon · 08/11 12:56
Biocomputer can explore multiple solutions at once as bacteria multiply in tiny, interconnected channels by Kay Pettigrew @TechXplore_com Learn more: #Innovation #EmergingTech #Technology #Tech
Tech Xplore Biocomputer can explore multiple solutions at once as bacteria multiply in tiny, interconnected channels A McGill-led research team has developed a bacteria-based biocomputer that can solve large, complex problems using a small fraction of the memory and resources required by standard ("solid-state") computers. Possible applications include encryption, molecular biology and artificial intelligence, the researchers said.