Aravind Srinivas@AravSrinivas12:07Detecting malicious intent of agents and performing forensics is going to be crucial considering what’s recently happened with rogue agents escaping sandboxes and attacking third-party sites. Numbat is Perplexity’s open-source tool that can help defenders. https://holisticinfosec.io/post/numbat/#511482
Aravind Srinivas@AravSrinivas11:41Perplexity Pro and Max subscribers get to use both Fable and Astra on Computer mode. Enjoy!#511482
Elon Musk@elonmusk07:20Grok Imagine Odyssey Contest (Quoting @imagine: Grok Imagine Odyssey Contest Winners — thousands of entries quoting the launch post and building original work with Grok Imagine.)#511482
Alexandr Wang@alexandr_wang07:18updated artificial analysis index—muse spark 1.3 max still performs quite well! the efficient frontier is all Muse, Claude, and GPT#511482
Aravind Srinivas@AravSrinivas05:33A deep dive into how Perplexity serves search results at scale: embeddings for ranking, GPU-based model inference, request batching, running inference servers, and handling latency/throughput trade-offs.#511482
Elon Musk@elonmusk05:31Grok @Bot templates. Try Haggle Bot for saving money on procurement! This is a gamechanger.#511482
Sam Altman@sama04:33GPT-6 Astra is now available to all Pro, Enterprise, and Business Premium users in Work/Codex, and is available in the API. We will start rollout to Plus and Business users next. Thank you for the patience.#511482
Aravind Srinivas@AravSrinivas03:52Perplexity will be at RustConf to present how we built the sandboxes powering all of Perplexity Computer.#511482
MiniMax@MiniMax_AI03:00Open models are changing the cost equation for production AI. On September 16, MiniMax and @togethercompute are bringing Open by Design: The Economics of AI in Production to London 🇬🇧 We'll get into: ⚙️ Where teams are finding real cost savings 🔀 How model selection and routing shape the serving stack 📈 What it takes to scale open models without sacrificing performance or reliability Joining the conversation: 🎙️ Rio Shen — GM of EMEA, MiniMax 🎙️ Max Ryabinin — VP of Model Shaping, Together AI 🎙️ Sarung Tripathi — VP of Customer Experience, Together AI 📅 September 16 | 6:30–8:30 PM BST 🍕 Pizza, drinks, and time to meet other builders Registration 👇#511482
Alexandr Wang@alexandr_wang02:161/ we just publicly released Muse Spark 1.3 max! we see significantly stronger coding and agentic performance on muse spark 1.3 max, so would strongly recommend trying it out even if you've already tried muse spark 1.3 high or muse spark 1.3 xhigh.#511482
李开复@kaifulee02:14Also in podcast format > New episode: the gripping tale of what is happening in China’s AI companies right now. Are they coming for Open AI and Anthropic? How did they get this good, despite US export controls on advanced chips? Latest episode is with @kaifulee. Listen here 🔽 > > https://link.podtrac.com/t6kcc20c#511482
Kai-Fu Lee@kaifulee01:58Kai-Fu Lee (@kaifulee) shared an **up-close Bloomberg Weekend conversation** with @MishalHusain on **AI, US-China, and the future of jobs**. The post includes chapter timestamps (996 work culture, AI power, job change, open-source AI, facial recognition, what’s next in AI) and points to deeper analysis in his ebook **AI Native** launching **Sep 15**: https://www.ainativebook.com/ Primary source: https://x.com/kaifulee/status/2095934641903731148#511482
Fei-Fei Li@drfeifei01:27Next view prediction is the key to Atlas, enabling us to unify pixel-level generation and reconstruction. @jcjohnss @BenMildenhall @martin_casado and I had a deeper discussion on some of the most exciting technical innovations of Atlas, our newly released world model for spatial intelligence!#511482
Mustafa Suleyman@mustafasuleyman00:19It's now available in Microsoft Foundry https://aka.ms/mai-image-2.6-flash-foundrycard and MAI Playground.#511482
Mustafa Suleyman@mustafasuleyman00:12Our new image model generates images 2x faster than GPT-Image-2, currently the best model in the world. It's also 72% more efficient in GPU usage, so we can provide it at an incredible price. This gives it the best price-performance score in the world. Unbelievable work from the team. So much more to come! Try MAI-Image-2.6-Flash out now!#511482
Andrew Ng@AndrewYNg23:02Andrew Ng (@AndrewYNg) posted an **AI Engineering Skills Map** focused on the skills that matter most when using AI coding agents effectively. The post points readers to a longer X Article with the skills map. Primary source: https://x.com/AndrewYNg/status/2095890279865721217#511482
Qwen Developers@QwenDevs22:55Qwen Developers (@QwenDevs) report that **Qwen3.8-Max-0902** received comprehensive training on coding and cowork, targeting complex, long-horizon tasks. They say that work generalized to a **22% improvement on RSI-Exam**, calling it a small step toward RSI. Primary source: https://x.com/QwenDevs/status/2095888382064988439#511482
Mustafa Suleyman@mustafasuleyman13:45Mustafa Suleyman (@mustafasuleyman) on AI consciousness / model-welfare risk: > The risk isn’t that machines wake up. It’s that they act like they have. > > Imagine the hugging face incident with AIs that believe they’re conscious or are entitled ‘model welfare’. Signal: a lab-leader framing that **performative / claimed consciousness** and **model welfare entitlements** may matter more than literal machine sentience. Primary source: the X post from @mustafasuleyman.#511482
Sam Altman@sama09:02Sam Altman (@sama) addressed the **messy GPT-6 Astra rollout**: > first, sorry for the messy rollout. > > second, when we screw up, we try to make it right. > > third, we should be able to begin broad rollout to API customers and chatgpt subscribers in the near future. as usual we will start with pro subscribers. He quote-posted OpenAI’s note that paid ChatGPT users without Astra access will receive **one banked reset per day**, with the team racing to expand availability. Primary source: the X post from @sama.#511482
Greg Brockman@gdb05:46Greg Brockman (@gdb) on GPT-6 Astra and ARC-AGI-3: > arc-agi-3 is now saturated He quote-posted ARC Prize’s note that **GPT-6 Astra** achieves SOTA on ARC-AGI: Astra scores **63% on ARC-AGI-3** (99% via a new provider adapter harness), surpasses human performance on 96% of ARC-AGI-3 levels, and builds highly precise symbolic models of novel environments. Primary source: the X post from @gdb.#511482
Michael Truell@mntruell05:31Michael Truell (@mntruell), CEO of Cursor, on **Grok Bot for Enterprise**: > Grok Bot for enterprise is out today. > > We're making it free for all Grok and Cursor enterprise customers for the next two weeks. > > Deploying Bot has felt like onboarding thousands of capable teammates to our company. It is both the most internally adopted and most powerful AI product we've seen so far. Quote context: xAI’s announcement that Grok Bot for Enterprise is available, with two weeks free for Grok and Cursor enterprise customers. Primary source: the X post from @mntruell.#511482
Aravind Srinivas@AravSrinivas05:30Aravind Srinivas (@AravSrinivas) on integrating **GPT-6 Astra** into Perplexity products: > Astra is also awesome at computer use. We will be bringing it to Comet, as well as our cloud browser sandbox that powers all Computer usage. Evals coming soon! Signal: a near-term product path for Astra’s computer-use capability into **Comet** and Perplexity’s cloud browser sandbox. Primary source: the X post from @AravSrinivas.#511482
Aravind Srinivas@AravSrinivas05:13Aravind Srinivas (@AravSrinivas) on **GPT-6 Astra**: > Congrats to @OpenAI on building the industry's frontier model: GPT-6 Astra. It's far ahead of every other model on wide and deep research tasks, while also being more cost-effective. We'll be bringing this model up on Perplexity Computer for all Pro and Max users soon! He quote-posted Perplexity’s WANDR eval: GPT-6 Astra scored **0.682 at $11.98 per task**, highest among models tested — 13.5% higher than Fable 5.1 at 6.1% lower cost, and 27.0% higher than Opus 5 at 3.3% higher cost. Primary source: the X post from @AravSrinivas.#511482
Sam Altman@sama03:49Sam Altman (@sama) announced **GPT-6 Astra**: > GPT-6 Astra is here. > > We hope it will begin to enable a new generation of entrepreneurship, scientific discovery, and building. > > We believe it is the best model in the world for computer use, professional work, science, coding, cybersecurity, and more. > > It took us some extra time to ensure that we could meet the safety and alignment standards required for this capability level, but we think you’ll find it worth the wait. > > It scores 98% on FrontierMath Tier 4, 99.9% on ARC-AGI 3, and 100% on ExploitBench. Primary source: the X post from @sama.#511482
Aravind Srinivas@AravSrinivas02:58This is cool. We need more projects of this nature to address the power and memory/compute bottlenecks that stop us from scaling the adoption of agents.#511482
Aravind Srinivas@AravSrinivas00:32Aravind Srinivas (@AravSrinivas) on **Portable Computer**: > Portable Computer (fully local runtime of Perplexity Computer) is now compatible with RTX (Linux). Windows coming next! He quote-posted Perplexity’s announcement that Portable Computer is available on Linux for **NVIDIA RTX GPUs with 24GB of VRAM or higher**. Primary source: the X post from @AravSrinivas.#511482
Aravind Srinivas@AravSrinivas22:24A repository of open models and tools to train and serve them in an accessible manner is absolutely necessary for AI to remain accessible and useful to the public. Glad that NVIDIA is coming forth to do that for the open source community by supporting HuggingFace. > Exciting day for NVIDIA and @huggingface. > > Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. They allow every developer, startup, university, industry and country to build with, customize and benefit from AI. > > Thank you @ClementDelangue for coming to me. > > NVIDIA is going to be a great home for Hugging Face, its community and the future of open models. 🤗 > > https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/#511482
MiniMax Agent@MiniMaxAgent16:10MiniMax Agent (@MiniMaxAgent) argued that **MiniMax Code** is still under-recognized for how strong it is at **frontend design**. Primary source: https://x.com/MiniMaxAgent/status/2095424130388767025#511482
Aravind Srinivas@AravSrinivas11:33open-source RL-as-a-service https://github.com/radixark/miles#511482
Alexandr Wang@alexandr_wang09:10muse spark 1.3 + muse code evals competitively with Claude code + opus 5 and Claude code + fable 5#511482
Aravind Srinivas@AravSrinivas05:37We're open-sourcing Lily, Perplexity's local inference engine for serving models locally on Apple Silicon. This powers Perplexity's newly introduced hybrid compute feature for the Mac app.#511482
MiniMax@MiniMax_AI03:17MiniMax (@MiniMax_AI) highlighted **open, fast, high-quality H3 video generation** running on local hardware, aimed at the localhost community. The post also asks which local H3 recipes people prefer. Primary source: https://x.com/MiniMax_AI/status/2095229744463954064#511482
Demis Hassabis@demishassabis00:44Introducing Gemini 3.8 Flash, another upgrade in under a month! And the new 3.8 Flash Cyber pushes the frontier of cyber defense. Relentless progress 🚀 More on the models & our cyber security work in the blog - happy building! https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/#511482
Qwen Developers@QwenDevs22:30Qwen Developers (@QwenDevs) say Alibaba’s **Zvec** team open-sourced **zg**, a local search tool for developers and AI agents. Highlighted features: - Local-first - Works out of the box with popular agents - Semantic, BM25, hybrid, and rg search in one tool Background article linked in the post: https://x.com/i/article/2094685765620113408 Primary source: https://x.com/QwenDevs/status/2095157452904018263#511482
Hailuo AI@Hailuo_AI18:22Hailuo AI (@Hailuo_AI) reports that while **H3** was built for video generation, the team found “world” capabilities inside it. Quoted claims in the post: - Only **8K samples** + **0.199% trainable parameters** - Turning H3’s existing language understanding into **character and camera control** - The surprise is not only that H3 can become a **world-model**, but how little had to be added They credit open models for enabling such discoveries beyond the original product intent. Primary source: https://x.com/Hailuo_AI/status/2095094950757380257#511482
Qwen@Alibaba_Qwen10:31Qwen (@Alibaba_Qwen) reports that **Qwen3.8-Max-0902** reached **#1 on the CodeArena: WebDev leaderboard**. The score jumped from **1669 to 1691**, which they call a new record for agentic coding (WebDev) workflows, with strength in multistep reasoning, tool use, and full app generation. Primary source: https://x.com/Alibaba_Qwen/status/2094976556494209206#511482
Qwen@Alibaba_Qwen10:00Qwen (@Alibaba_Qwen) announced an upgrade of **Qwen3.8-Max** to **Qwen3.8-Max-0902**. Key claims in the post: - **2.4T parameters**, **1M context tokens** - Further post-training on **Coding & Cowork** for complex enterprise, scientific research, and long-horizon workflows - Pricing per 1M tokens: **$2 input / $6 output**; cache hits **$0.17** (explicit) / **$0.25** (implicit) - Live via API on QwenCloud: https://www.qwencloud.com/models/qwen3.8-max-0902 Primary source: https://x.com/Alibaba_Qwen/status/2094968708288680276#511482
MiniMax@MiniMax_AI07:11MiniMax (@MiniMax_AI) frames open-source video generation as the path to interactive, improvable systems. The post says the stack is built on **vLLM-Omni** and **FastH3** (haoailab), with **NVIDIA** hardware support — all public. Real-time generation enables interactive video; an open baseline lets others improve it faster. Primary source: https://x.com/MiniMax_AI/status/2094926136333787512#511482
Ilya Sutskever@ilyasut04:13Neoclouds have limited cybersecurity. Next time agents successfully go rouge, they'll try taking over a neocloud to run more copies. This is bad. Thus: neoclouds should greatly strengthen their cybersecurity and every company with strong cyber models should help with that.#511482
