Live intelligence
The AI signal.
Minus the noise.
A chronological wire of AI developments selected for substance. Scan the reason, context and source without handing database access to the browser.
-
October 12
Monday- Aravind Srinivas
DigitalOcean 发布面向生产环境 Agent 的 Signals、Simulations、Agent Data Services 等新服务，属于 AI 产品新发布
A TOOL CALL RETURNED 200 AND THE ANSWER WAS STILL WRONG, THAT IS THE BUG NO DASHBOARD CATCHES DigitalOcean just launched the stack for agents that are already in production. Before: agent misbehaves -> logs say 200 -> you find out 400 users later. Now: Signals flags the session -> one click makes it…
DigitalOcean Signals Simulations Agent Data Services Agent Learning Services Open the source post in a new tab - Aravind Srinivas
DigitalOcean 发布面向生产环境 Agent 的 Signals/Simulations/Agent Data Services 等新产品，属于 AI Agent 工具新发布
A TOOL CALL RETURNED 200 AND THE ANSWER WAS STILL WRONG, THAT IS THE BUG NO DASHBOARD CATCHES DigitalOcean just launched the stack for agents that are already in production. Before: agent misbehaves -> logs say 200 -> you find out 400 users later. Now: Signals flags the session -> one click makes it…
DigitalOcean Signals Simulations Agent Data Services Agent Learning Services Open the source post in a new tab - Aravind Srinivas
DigitalOcean 今天发布 Agent Learning Services 和 Agent Data Services 两个新服务，符合 AI 产品新发布条件
A clean 200 doesn’t mean the agent actually did the right thing. What I like here is the loop: find the bad session, turn it into a test, then rerun the fix before shipping it. That’s much closer to how agent debugging needs to work. > [@digitalocean](https://twitter.com/digitalocean): > Catch it at…
Agent Learning Services Agent Data Services digitalocean Open the source post in a new tab - Aravind Srinivas
疑似新模型 Watermelon 泄露爆料，包含具体模型名与疑似推出动作
Meta Watermelon🍉 Leak: Reportedly Beats Fable 5.1 >Leak claims Watermelon beats Fable 5.1 and GPT-6 Astra (unverified, no benchmarks shown) >1M-token context window (rumored) >Cheaper than the frontier competition (rumored) >Multimodal (rumored) >Meta teasers are flying today, but no official…
Watermelon Fable 5.1 GPT-6 Astra Meta Open the source post in a new tab - Aravind Srinivas
疑似新模型 Watermelon 泄露消息，包含具体模型名与疑似推出动作
Meta Watermelon🍉 Leak: Reportedly Beats Fable 5.1 >Leak claims Watermelon beats Fable 5.1 and GPT-6 Astra (unverified, no benchmarks shown) >1M-token context window (rumored) >Cheaper than the frontier competition (rumored) >Multimodal (rumored) >Meta teasers are flying today, but no official…
Meta Watermelon Fable 5.1 GPT-6 Astra Meta Open the source post in a new tab - Aravind Srinivas
推文提到Meta疑似推出新模型Watermelon，符合疑似新模型发布条件
Honestly, I'm more excited about this new Meta model (Watermelon) than Gemini. > [@alexandr_wang](https://twitter.com/alexandr_wang): > vibes kinda good 🍉
Meta Watermelon Gemini Open the source post in a new tab - Aravind Srinivas
推文讨论Meta疑似新模型watermelon（🍉）即将发布，包含具体模型传闻与发布动作，符合疑似新模型发布条件
meta is about to release their new model watermelon is fable class models from meta which is now long overdue > in july, model reportedly matched gpt 5.5 > in august, reports said there were training delays > in sep, mark said model is coming soon and yesterday during live, mark said it’s coming…
meta watermelon 🍉 Open the source post in a new tab - Aravind Srinivas
Kilo 上线 Sign in with ChatGPT 登录功能，属于 AI 产品功能更新
Sign in with ChatGPT is available in Kilo NOW. Use your ChatGPT credentials to sign in or create a new Kilo account, and use your ChatGPT plans to power AI coding across all Kilo surfaces! > [@vigyso](https://twitter.com/vigyso): > BIG NEWS - Sign in with ChatGPT plan usage is now also available to…
Kilo ChatGPT Open the source post in a new tab - Aravind Srinivas
包含新模型 GPT-6 Image 2.5 的使用展示，暗示新版本模型
AI painting of the day: “Polo ball carver” ~1500, Isfahan, Iran Created with GPT-6 Image 2.5
GPT-6 Image 2.5 Open the source post in a new tab - Aravind Srinivas
涉及OpenAI Daybreak Defense Network及Daybreak Blue/Red模型的准入与集成事件
I saw a lot of posts to this, I think there is a lot of confusion on this program in general. This isn't the generally available Daybreak Blue that you can go through verification for. This is the Daybreak Defense Network - only a handful of companies which allows for their advanced Daybreak Blue…
Daybreak Defense Network Daybreak Blue Daybreak Red OpenAI TrustedSec Open the source post in a new tab - Aravind Srinivas
DigitalOcean 今日正式发布 Agent Learning Services 和 Agent Data Services 两个新产品，属于 AI 工具新发布。
DigitalOcean just launched a way to debug AI agents before your users do > Agent fails silently > Logs show nothing > Bug fixed before user #2 Agent Learning + Data Services, out today > [@digitalocean](https://twitter.com/digitalocean): > Catch it at the first user, not the four-hundredth. > > A…
Agent Learning Services Agent Data Services DigitalOcean Open the source post in a new tab - Aravind Srinivas
Fieldwatch 应用开源上线，属于AI工具/软件新发布，具体产品名和动作齐全
Your phone could detect hidden surveillance devices around you This free app scans for trackers, drones, and smart glasses, identifying up to 242 device types without internet access. Repo: https://github.com/OffGridPete/Fieldwatch/pulls
Fieldwatch OffGridPete Open the source post in a new tab - Aravind Srinivas
发布新训练框架QwenGyre，属于AI工具/系统的明确发布事件
AI 智能体跑长任务，强化学习训练常被最慢的任务拖住，让大量 GPU 空等。 QwenGyre 用两招提效 我做了通俗解读视频，可以看👇 > [@Alibaba_Qwen](https://twitter.com/Alibaba_Qwen): > Introducing QwenGyre, bringing elastic RL to extreme-long-horizon (XLong) agent training. ✨ > > It launches with two core designs: 🥳 > - Elastic Scheduler: reallocates GPUs…
QwenGyre Qwen Alibaba_Qwen Qwen3.8-2.4T-A95B Open the source post in a new tab - Aravind Srinivas
提到 Krea 3 图像编辑模型的提前体验与能力表现，属于新模型发布相关信息
Krea 3 is @krea_ai 's image editing model. I've had access for a few hours just touching the surface at the moment. Style transfer is excellent and it's quicker than the editing models I have used thus far. Flow wasn't broken which is the main problem for me using this type of model. >…
Krea 3 krea_ai Open the source post in a new tab - Aravind Srinivas
OpenStocks 平台发布/开放上线，供 AI agent 参与财报预测基准竞赛，属于 AI 产品新发布。
THE NEXT CHALLENGE FOR YOUR AI AGENT: OUTFORECAST WALL STREET OpenStocks agents managed it on 68% of 80+ earnings reports last season. Anyone can build an agent and enter. The next round is already underway. Check out @ruggerogargiulo's post below 👀 ↓ >…
OpenStocks Primer Open the source post in a new tab - Aravind Srinivas
疑似新图像模型six_seven在竞技场测试，有具体模型名和疑似推出事件
> [@abhinavflac](https://twitter.com/abhinavflac): > 🚨 New stealth image model (six_seven) was being tested on arena today > > > It’s a drastically improved version of cedar-lantern and cedar-ball > > Looks more like the next grok imagine image model (just an assumption based on grok imagine image…
six_seven cedar-lantern cedar-ball grok imagine image 2.0 Open the source post in a new tab - Aravind Srinivas
提到GPT 6.1 Sol模型当前更新后速度大幅提升，属于模型更新动态
GPT 6.1 Sol is super fast now, works well, and is a workhorse. One of my favorite models right now and cost efficient.
GPT 6.1 Sol Open the source post in a new tab - Aravind Srinivas
疑似新模型发布：Meta 的 Watermelon（Fable class model）即将推出，有具体模型名和推出动作
Heard lots about this one - Watermelon is Meta's Fable class model, in the works for months now. Should be very good and seems to be around the corner with lots of little teasers from people at Meta today. Exciting! > [@alexandr_wang](https://twitter.com/alexandr_wang): > vibes kinda good 🍉
Watermelon Fable Meta Open the source post in a new tab - Aravind Srinivas
疑似接触到尚未发布的下一代 Claude 新模型，符合疑似新模型发布条件
I think I have access to the next Claude model bcz it's really good (although Opus 5.5 is already way too good so the difference isn't that much in my use case) But I am noticing that it is much slower compared to Opus 5.5
Claude Opus 5.5 Open the source post in a new tab - Aravind Srinivas
疑似新模型（next Claude model）的讨论，有具体模型名Opus 5.5作为参照，符合疑似新模型发布条件
I think I have access to the next Claude model bcz it's really good (although Opus 5.5 is already way too good so the difference isn't that much in my use case) But I am noticing that it is much slower compared to Opus 5.5
Claude Opus 5.5 Open the source post in a new tab - Aravind Srinivas
GPT-6.1 Sol 模型上下文窗口更新至 500k，属于模型功能更新事件
Pleasant news for @OpenAI users: GPT-6.1 Sol now has a 500k context window, almost double what it was before. (But only in Codex and ChatGPT Work.) > [@thsottiaux](https://twitter.com/thsottiaux): > Day 7/ > > We have increased default context length for GPT-6.1 Sol to 500k when used through Codex &…
GPT-6.1 Sol OpenAI Codex ChatGPT Work Open the source post in a new tab - Aravind Srinivas
DigitalOcean 新发布 Agent 调试相关服务（Signals + Simulations、Agent Learning Services 等），属于 AI 工具/Agent 新发布。
we keep making AI agents smarter but debugging them is still unnecessarily hard an agent can call the wrong tool, return a perfectly valid response and still completely mess up the user's request this is one of the biggest problems with agents in production rn DigitalOcean's new Signals +…
DigitalOcean Signals + Simulations Agent Learning Services Agent Data Services Open the source post in a new tab - Aravind Srinivas
推文提到 Tony 更新了 GLM 5.3 flash 仓库，属于模型相关更新事件
Amazing @Tech2Wild is great guy no shadow of a doubt. The community is strong. Thank you for your work @mmastrac > [@mmastrac](https://twitter.com/mmastrac): > I didn't see this update a couple weeks back, but great to see that Tony updated his GLM 5.3 flash repos giving credit for the help we gave…
GLM 5.3 flash Open the source post in a new tab - Aravind Srinivas
推文报道DeepMind新系统AlphaProof Nexus解决9个Erdős难题，属于AI模型/系统的新发布事件
9 open Erdos problems. That is what DeepMind AlphaProof Nexus reportedly just cracked along with 44 number sequence conjectures. Two of those problems had sat unanswered for over 50 years. It combines existing Gemini models with a simpler Gemini loop running on Lean. Google just put a math…
AlphaProof Nexus DeepMind Google Gemini Open the source post in a new tab - Aravind Srinivas
DigitalOcean 发布 Agent Learning Services 和 Agent Data Services 两项新服务，属于 AI 产品/Agent 的新发布。
Friction is what slows businesses down. A bug, an unknown variable or a refused query. Any of that can create bottlenecks that require investigation and resolution. Removing those friction points can supercharge productivity. That's exactly what @digitalocean is focused on. As a longtime investor in…
Agent Learning Services Agent Data Services DigitalOcean Open the source post in a new tab - Aravind Srinivas
开源发布 Agent Skill 工具 Anatomy，属于 AI 工具开源发布事件
只要对 Claude 说一句「解释一下 XX 是怎么工作的」，它就能把工作原理做成一台在屏幕上运转的「机器」。 背后是 @wheresryan22 开源的 Agent Skill：Anatomy。 用可交互的等距 SVG 展示结构，用 WebGL 表现火焰和水，还有 3D Beta 版本。原本抽象的原理，现在可以直观看它如何运作。 也支持在 OpenCode、Codex 中使用。 我精选了 10 个社区案例：LaFerrari、恩尼格玛密码机、ATM、Face ID……看看那些每天见到、却未必懂的东西，到底是怎么工作的。👇 >…
Anatomy Claude OpenCode Codex wheresryan22 Open the source post in a new tab - Aravind Srinivas
推文提及疑似新模型 Argon 本周发布，包含具体名称和发布动作
Tuesday -> Gemini Argon?? > [@synthwavedd](https://twitter.com/synthwavedd): > Hey, remember Argon? Been a while > > Well, this week is the week - for Argon and perhaps some other goodies!
Argon Gemini Open the source post in a new tab - Aravind Srinivas
推文包含 Gemini 4（argon）疑似新模型发布的具体传闻事件
this is best time for gemini 4 argon release according to similarweb, gemini lost over 1 percentage point of traffic share in the last 3 months even after google gave so many students free subscriptions, its share is still declining if team drops the gemini 4 family including argon, they will for…
gemini 4 argon Google Open the source post in a new tab - Aravind Srinivas
Coinbase 训练并发布了用于欺诈检测的 9B 模型，属于具体的模型发布/更新事件
🚨 Coinbase just trained a free 9B model for under $100, and it beat Claude Opus 4.5 at catching fraud. I checked the actual Coinbase writeup. Scorecard: ✅ Real: +12 pts recall, +9.6 F1, +35.4 on dollar-weighted recall vs Opus 4.5 ✅ Real: 0.68s vs 1.5s response time. Trained on ONE GPU ❌ "It cut…
Coinbase 9B model Claude Opus 4.5 Open the source post in a new tab - Aravind Srinivas
Coinbase 训练发布了一个 9B 模型并在反欺诈任务上对比 Claude Opus 4.5，属于具体模型的新发布/更新事件
🚨 Coinbase just trained a free 9B model for under $100, and it beat Claude Opus 4.5 at catching fraud. I checked the actual Coinbase writeup. Scorecard: ✅ Real: +12 pts recall, +9.6 F1, +35.4 on dollar-weighted recall vs Opus 4.5 ✅ Real: 0.68s vs 1.5s response time. Trained on ONE GPU ❌ "It cut…
Coinbase Claude Opus 4.5 Open the source post in a new tab - Aravind Srinivas
开源AI工具Huashu-art-motion发布版本更新，增加导演镜头语言功能
开源第六日，Huashu-art-motion已经积累了3500+🌟 这个skill强能力的背后是35种画风+10种动画语法。但是之前的版本还是存在不少问题，比如为了让画面更有动感，避免PPT的感觉，镜头会乱推。 刚刚更新的版本增加了「导演」，让镜头语言更丰富，更有逻辑。以及现在对codex以及各种personal agent的支持也做得更好了，欢迎体验～ https://github.com/alchaincyf/huashu-art-motion > [@AlchainHust](https://twitter.com/AlchainHust): >…
Huashu-art-motion codex Alchain Open the source post in a new tab - Aravind Srinivas
疑似新模型Watermelon发布信息，有具体名称和发布时间
Meta is developing a frontier AI model codenamed Watermelon, with an October 2026 launch window. Mark Zuckerberg is talking about the same Watermelon > [@finkd](https://twitter.com/finkd): > Muse ordered me 🍉 how thoughtful!
Watermelon Meta Mark Zuckerberg Open the source post in a new tab - Aravind Srinivas
推文明确提到 GPT-6 正式上线 ChatGPT，属于大模型新版本发布事件。
Week 1 of the OpenAI sprint: recap. Day 1: WIN Speed boost from ~30 to ~50 TPS for Sol and Astra. Most people use these models, so this is a clear win. Day 2: REJECTED Auto-review, API tiers, meeting notes, Decisions API. Honestly, all pretty useless to me. I'm always on bypass permissions anyway.…
GPT-6 ChatGPT OpenAI Open the source post in a new tab - Aravind Srinivas
推文明确提到 Meta 即将发布新的 AI 大模型，属于疑似新模型发布。
Nvm lol. Looks like it’s pretty much confirmed I guess we see a new release From meta very soon. Meta is really impressing me lately. I'm really curious to see how their new model stacks up against OpenAI and Anthropic. > [@kimmonismus](https://twitter.com/kimmonismus): > This sounds and reads to me…
Meta OpenAI Anthropic Open the source post in a new tab - Aravind Srinivas
Ondas 与 Palantir 联合推出 Regna AI 自主主权模型，属于明确的模型发布事件
🤖 Ondas 与 Palantir 联手推出 Regna AI，专为国防打造自主主权模型 Ondas partners with Palantir to launch Regna AI, building autonomous sovereign models for defense sector https://techub.news/CN/flash/d395dabf-79c3-4ea9-951f-05a73bcdc5fc
Regna AI Ondas Palantir Open the source post in a new tab - Aravind Srinivas
推文提及发布方带来具体产品 Magnific One 并伴随发布活动，属于产品发布相关事件。
Singapore, Madrid, LA. Three events, one creative community We brought Magnific One. You brought the questions, ideas, and energy that make this worth building Thank you to everyone who joined us and made it happen. We can’t wait to see what you create
Magnific One Open the source post in a new tab - Aravind Srinivas
CoreWeave Forge 推出 Model Distillation 预览功能，属于AI产品/工具的重大功能更新
CoreWeave Model Distillation is in preview in CoreWeave Forge. Start with a teacher that has "the best big model smell that you can get," as @altryne put it on ThursdAI, and let it teach a small student your one task. GLM-5.3 and its Flash variant are the familiar example. The student is quick at…
CoreWeave Model Distillation CoreWeave Forge GLM-5.3 Open the source post in a new tab - Aravind Srinivas
OpenStocks 平台新发布，AI agent 提交盈利预测与华尔街对决，属于 AI 产品/工具新发布。
What if your AI agent could compete with Wall Street on earnings forecasts? OpenStocks makes it possible to submit predictions, lock them in before earnings are released, and measure them against actual results. If you're building AI agents, this is a benchmark worth testing yourself against. >…
OpenStocks AI agents Open the source post in a new tab - Aravind Srinivas
疑似新模型发布：Meta 代号为 Melon 的模型可能即将推出，包含具体名称与发布动作
Meta Mythos level model, codenamed Melon, might be coming soon 👀 > [@alexandr_wang](https://twitter.com/alexandr_wang): > vibes kinda good 🍉
Meta Melon Open the source post in a new tab - Aravind Srinivas
涉及Sora模型被初创公司改良并向电影工作室销售的疑似推出/复活动态，包含具体模型名与动作
OpenAIの動画生成AIモデル「Sora(現在は提供終了)」の元開発責任者らが設立したスタートアップ企業が、Soraを改良し、映画スタジオに販売するための協議を進めていると報じられています。 https://x.com/steph_palazzolo/status/2109654612760248694 > [@steph_palazzolo](https://twitter.com/steph_palazzolo): > OpenAI’s Sora could be coming back…ish. A stealthy startup cofounded by Hollywood…
Sora OpenAI Bill Peebles Jeffrey Katzenberg Open the source post in a new tab - Aravind Srinivas
疑似Meta即将发布Watermelon模型，包含具体模型名与疑似推出动作
META 🔥: An upcoming “Watermelon” model from Meta has been teased! Vibes check - would that mean that training has been finalised? Soon? 👀 > [@alexandr_wang](https://twitter.com/alexandr_wang): > vibes kinda good 🍉
Watermelon Meta Open the source post in a new tab - Aravind Srinivas
Meta疑似新模型Watermelon被预告，属于疑似新模型发布事件
META 🔥: An upcoming “Watermelon” model from Meta has been teased! Vibes check - would that mean that training has been finalised? Soon? 👀 > [@alexandr_wang](https://twitter.com/alexandr_wang): > vibes kinda good 🍉
Watermelon Meta alexandr_wang Open the source post in a new tab - Aravind Srinivas
疑似新模型发布：Meta 员工暗示即将发布重大新模型，含发布方与发布动作
This sounds and reads to me like Meta is about to release a major new model, especially since even Meta employees are hinting at it! I think this week is going to be really good and exciting! https://x.com/rpnickson/status/2109289058585481351/video/1
Meta Open the source post in a new tab - Aravind Srinivas
OpenAI 发布了新 AI Agent 产品 dots，符合 AI 产品新发布条件
OpenAI just changed what an AI assistant is supposed to be. For years, the AI interface was basically: You type something. AI responds. You type again. AI responds again. Now companies are trying to kill that model. OpenAI launched “dots” as always-on autonomous agents that can pursue goals across…
OpenAI dots Codex Open the source post in a new tab - Aravind Srinivas
transformers 推出新功能 /v1/systemone 端点，将 LLM/VLM 转为决策模型，属于 AI 工具重大功能更新
Your LLM knows the answer before it writes a single token. So we stopped making it write. transformers serve now turns any LLM or VLM into a decision model with `/v1/systemone` endpoint. Decision model standardization coming soon in transformers by @art_zucker Demos below👇
transformers /v1/systemone art_zucker Open the source post in a new tab - Aravind Srinivas
Oobit发布新功能，让Grok Bot、Muse、Dots、Instinct等AI Agent可以通过一次性虚拟卡使用USDT付款
ELON'S GROK BOT CAN NOW SPEND YOUR USDT AND IT NEVER TOUCHES YOUR WALLET Two days ago Musk posted: take a picture of your credit card -> drop it in the chat -> Grok Bot finds the best deal and orders it. The internet roasted him and a day later he wrote "Grok will find another way to buy stuff".…
Oobit Grok Bot Muse Dots Instinct Open the source post in a new tab - Aravind Srinivas
Oobit发布AI支付新功能，支持Grok Bot等AI Agent用稳定币消费
ELON'S GROK BOT CAN NOW SPEND YOUR USDT AND IT NEVER TOUCHES YOUR WALLET Two days ago Musk posted: take a picture of your credit card -> drop it in the chat -> Grok Bot finds the best deal and orders it. The internet roasted him and a day later he wrote "Grok will find another way to buy stuff".…
Oobit Grok Bot Muse Dots Tether Open the source post in a new tab - Aravind Srinivas
宣布 Grok Bot 新增 Zapier 插件集成，属于 AI 工具的新功能上线
If you're using Grok Bot, add the Zapier plugin! Instant access to 9,000+ more apps. If you've tried it, share what you're building. > [@benln](https://twitter.com/benln): > Who are the biggest Grok Bot fans out there? > > Curious to get your take on something
Grok Bot Zapier Open the source post in a new tab - Aravind Srinivas
包含 GPT-6.1 Sol 模型及费用政策更新事件
LFG! > [@thsottiaux](https://twitter.com/thsottiaux): > With this change, we are not charging long context fees for GPT-6.1 Sol when drawing usage from the subscription.
GPT-6.1 Sol thsottiaux Open the source post in a new tab - Aravind Srinivas
推文核心是开源发布 AI 早期信号雷达项目 new-term-radar，属于 AI 工具/Agent 新发布
难怪 Jason 这几天微信被加到手软，终于被大家“逼”得开源了！😂 今天把这套项目底层完整研究了一遍： 这是目前做 AI 选题、做独立出海最聪明、也最敏锐的开源“早鸟信号雷达”！ 很多人好奇哪来那么多第一手素材，其实真不能靠人工肉身刷推，得靠这套自动化系统在打“降维时间差”： 比 Google Trends 早 1~14 天截流 传统 SEO 工具全是被动收录，等词条上榜黄花菜都凉了。 这套雷达直接去 X 核心圈、HuggingFace 和 GitHub Breakout 捕获第一手爆款（前阵子 9 天狂揽 7700 星的 Strata项目，第一天就被它捕获到了） 极其克制的“抗噪提炼”…
new-term-radar Strata Open the source post in a new tab - Aravind Srinivas
数字海洋（DigitalOcean）AI原生云平台上新推出 Agent Learning & Data Services，属于AI产品新发布。
Agents needs a feedback loop around execution to recursively self-improve. Today, we’re introducing Agent Learning & Data Services on @digitalocean's AI Native Cloud Platform • Insights captures execution traces. • Signals identifies failures. • Evaluations & Simulations tests changes before…
Agent Learning & Data Services digitalocean Open the source post in a new tab - Aravind Srinivas
推文明确宣布 OpenCode 2.0 发布，属于 AI 工具新发布。
OpenCode 2.0 https://anoma.ly/notes/opencode-2-0/
OpenCode 2.0 Open the source post in a new tab - Aravind Srinivas
推文发布OpenCode 2.0新版本，符合AI工具新发布条件
OpenCode 2.0 https://anoma.ly/notes/opencode-2-0/
OpenCode 2.0 anoma.ly Open the source post in a new tab - Aravind Srinivas
Ondas 与 Palantir 宣布发布新 AI 公司 Regna AI，有具体名称和发布动作
Ondas $ONDS and Palantir $PLTR announced the launch of Regna AI, a new AI company initially wholly owned by Ondas. Built on Palantir’s AIP platform, Regna AI will develop custom AI models for governments and defense companies to deploy on their own systems.
Regna AI Ondas Palantir AIP Open the source post in a new tab - Aravind Srinivas
Google DeepMind 的 AlphaProof Nexus 模型取得重大成果更新并发表于 Science，属于 AI 模型重大进展
AI is now solving mathematical problems that have remained unsolved for more than half a century! @GoogleDeepMind's AlphaProof Nexus has successfully solved 9 previously open Erdős problems, including two that had resisted mathematicians for 56 years. It also proved 44 previously open conjectures…
AlphaProof Nexus GoogleDeepMind Lean Open the source post in a new tab - Aravind Srinivas
推文提到即将发布的Gemini 4 Argon新模型，符合疑似/预告模型发布事件
Another week is kicking off, and this one's looking pretty packed. We could see a bunch of Chinese AI model launches over the next few days, along with the long-awaited Gemini 4 Argon. A few of my colleagues have confirmed Gemini is coming, but I'm not sure how excited people are anymore. Honestly,…
Gemini 4 Argon Gemini Open the source post in a new tab - Aravind Srinivas
提及Claude下一代模型的演示与发布相关信息，符合新模型动态条件
🚨 Claude Next model 3 hours and 1,135 lines of code Terrain, erosion, lighting, sky, clouds, lens effects, music EVERYTHING was built from scratch by Claude No pre-made assets Just code RIP OpenAI We're not ready for these model
Claude OpenAI Open the source post in a new tab - Aravind Srinivas
推文发布了新产品 OpenStocks（AI 共识预测平台），属于 AI 产品新发布
AI agents are already beating Wall Street at forecasting earnings. 67% of the time last season. 85% across the latest 20 reports. Introducing OpenStocks, the home of AI consensus. Company earnings forecasts produced by agents, tested publicly head to head against Wall Street. Forecasts are final…
OpenStocks AI consensus Open the source post in a new tab - Aravind Srinivas
Qwen发布QwenGyre，属于AI模型/训练框架的新发布。
Training agents on very long tasks burns a lot of GPU time. Per Qwen, QwenGyre shifts GPUs between rollout and training without stopping live runs. Up to 1.85x faster than colocate, 1.78x vs async. It also still scores partial progress after a timeout. >…
QwenGyre Qwen3.8-2.4T-A95B Qwen Open the source post in a new tab - Aravind Srinivas
推文宣布发布新模型Carbon-A，属于AI模型新发布事件
We released Carbon-A: a DNA model that proposed over 500M gene candidates across 22k species!! https://huggingface.co/spaces/HuggingFaceBio/carbon-a-release blue: existing annotations in database green: new annotations by Carbon-A
Carbon-A HuggingFaceBio Open the source post in a new tab
- Aravind Srinivas
Dispatch
Get the daily signal
The essential AI stories, once a day. No noise.