.@nytimes this morning https://t.co/sX0sgGAoKB

@itsjustcornbro https://t.co/58s619TyNj

Grok 4.3 beta is natively multimodal, and the front-end capabilities are insane You can literally just upload a screenshot of any website you like, and Grok will instantly write the code to clone it for you with an cool UI You don't even need to write a complex prompt...just upload an image or describe what you want and let it build

@Bencera I have all the AI founders, thousands of them, on lists here: https://t.co/fasUz7PuHq Am one too. Built this: https://t.co/kiuZ7QXLzb

okok we officially have GLM 5.1 running on a 256gb mac studio with hermes agent next is linking it to hermes to see how good it is 🗣️ https://t.co/BQTlLiL3jm

Surge AI just made the Forbes AI 50 list. 99% of the rest of the list raised billions in VC. We got there with $0. We didn't do it by building engagement slop and chasing DAUs. We didn't do it by rewarding sycophancy over truth. The standard Silicon Valley playbook — raise billions, blitzscale, worry about the effects of what you're building later — forces you to cut corners, compromise your principles to hit quarterly targets, and optimize for hype instead of substance. We chose a different path. We did it by doing the most unsexy work in the industry: building the school for AGI. Hiring the world's top doctors, engineers, attorneys, scientists, and writers to teach models how to actually think. Designing the curriculum that determines what intelligence becomes. Grading models on the standard of real work, not vibes. Building the full education — reasoning, wisdom, creativity, and taste — not just the standardized exam. You don't need hyper-growth VCs to build the world-changing things that only you could build. You just need an uncompromising commitment to your principles and work so good that your customers keep coming back. Years ago, we bet that AGI deserves more than a textbook education. We bet that the only way to build true intelligence is to raise it on the best of humanity — on the brilliance, rigor, and taste of the most talented experts in the world. We bet that independence and patience would beat headlines and hype. We bet on our technology and the quality of our product. We bet that researchers would notice and care. You can choose a different path. We're just getting started. https://t.co/RSYMJWEjjk

Grok 4.3 can create Slides https://t.co/8aJ3VeHB2b

Alex Finn, @AlexFinn, says OpenClaw running on the cloud is a scam: https://t.co/mGrteVR3oq While he is right in a perfect world that everyone would have it running on a local computer, there are many who don’t want to do that, or who can’t afford it. Keeping it all running is a pain for some people. And I just bought a new Macintosh. One problem: it won’t arrive for months because Apple is sold out. So in meantime I’m running OpenClaw on the cloud since I don’t want to give it access to my main machine where one security mistake could cost me a lot. You might have the same fears, or be in the same boat. ZooClaw is not another AI assistant. It's your personal team of AI specialists. One entry point, multiple specialists. Tasks are routed to the right agent, each equipped with structured domain knowledge. Speak naturally — the voice is native. Built on OpenClaw, ZooClaw stays synced with the latest models and falls back to top open-source models when needed — so the work never stops. No setup, no deployment, no API keys, no token anxiety. This lets you start automating your business and life without worry, and someday you can join the cool kids running everything locally if you want, or can afford to. Try it at https://t.co/OEApOIxzgv The first 20 users who use code "[ZC-CAACCD]" will get 1,000 free credits.

SpaceX completed two Starlink launches about 19 hours apart, pushing the number of Starlink satellites launched in 2026 past 1,000. Starlink 10-24 from Florida put 29 V2 Mini satellites into orbit with B1080 on its 26th flight. Starlink 17-27 from California added 25 more with B1082 on its 21st flight. Both boosters landed successfully. Both payloads were deployed about an hour after liftoff. 54 satellites added in a single day. Starlink now has more than 10,200 satellites in orbit.

You gotta pump those numbers up. Those are rookie numbers in this racket. https://t.co/f2TaEy8NIo

Hermes Agent @nousresearch Creative Hackathon 正式开启！🎉 总奖金 $25k 🔥 用Hermes Agent 创作视频、图像、音频、3D，创新软件、交互等创意作品 主赛道 $15k + Kimi专赛道 $5k（用kimi模型，额外价值 $5k credits ），各设一二三等奖 @Teknium @NousResearch are there any free credits on Nous portal subscription during this competition? 两周时间，5月3日提交截止！ Builder们快来参赛赚美刀～🚀 #HermesAgent #AIHackathon

Building sick new personality/memory traits for my agent ED on my @NousResearch Macbook Pro Hermes... He's starting to evolve right before my eyes! Dude's a comedian. https://t.co/g831GDx0z1

Submissions due EOD Sunday, May 3rd To enter: tweet a demo video tagging @NousResearch with a short writeup, then drop the link to the post in the "⁠creative-hackathon-submissions" channel on the Nous Research Discord. https://t.co/qxek9haYeb The submissions will be judged by the Nous Research and Kimi team on creativity, usefulness and presentation. To qualify for the Kimi Track, you must prove your use of Kimi models in your submission video. Submissions using Kimi models are eligible for both tracks.

Join the discord to turn in your hackathon submissions! https://t.co/SxWoPjZTVD

Nous Research rolled out the "Tool Gateway" within the Nous Portal, offering a unified ecosystem for AI agent development and deployment. A single paid Nous Portal subscription now natively includes access to over 300 models alongside built-in tooling for web scraping, browser automation, image generation, text-to-speech, and a cloud terminal backend.

Just built this in 5 min with Grok 4.3 (beta) Not bad, huh? https://t.co/tebfZAiLt3

Introducing HermesAgent-20, a new Bench Pack for BenchLocal. 20 scenarios extracted straight from the Hermes Agent source code, run against a REAL Hermes instance. The actual workload you'd put your model through. Why I built BenchLocal in the first place: most benchmarks are too abstract. We use local LLMs for practical work, and finding the right model for YOUR task efficiently is the single most important thing, especially when you're constrained to what fits on your machine. BenchLocal is a framework: providers, models, side-by-side comparison, all in one UI. Bench Packs are the unit of testing: ToolCall-15 and BugFind-15 shipped first, and when I launched the BenchLocal 0.1.0, added StructOutput, ReasonMath, InstructFollow, DataExtract. Now, HermesAgent-20 is the newest. Bench Packs install like VS Code extensions. The SDK is open, write your own, share it, grow the ecosystem. Here's the goal: a community-built, practical evaluation layer for the local LLM space. Early numbers on HermesAgent-20: > GLM 5.1 — 85 > Gemma4 31B — 83 > Qwen3.5 27B — 79 > MiniMax M2.7 — 76 Upgrade to the latest BenchLocal to install HermesAgent-20 (SDK update required).

Hermes 一丢 Agent，全网程序员集体进化了！ Nous Research 扔出 hermes-agent（90k+ stars），核心就一个词：自我进化。它不是玩具，而是带持久记忆、自动提炼技能、跨会话成长的底层骨架。 结果？社区直接把它当 DNA，短短几周卷出 80+ 进化体，生态总星 10 万+。这才是开源的最高境界：一个 Agent，变成全世界的进化树。 我挑了 4 个正在 X 上刷屏的 “Hermes 系进化体”，AI 玩家看了会沉默，爱好者看了会狂喜： 1️⃣ Hermes Ecosystem Map / Hermes Atlas（https://t.co/3wVPdY9cFg + https://t.co/L5UNrkxP1N） Kevin 用 Hermes Agent 自己刮全网，做出 80+ 项目安全审查地图 + RAG 聊天机器人。 还能生成《State of Hermes》报告，社区 PR 狂冲，官方都点赞了。 2️⃣ Hermes-Wiki（https://t.co/gTncSju9b0） 喂它自己源码，它就自动生成并维护完整项目 Wiki。 RAG 哭晕，Agent 自己给自己立传、太 meta 了。 3️⃣ hermes-hud（https://t.co/DivT37Yf9r） TUI 仪表盘进化成浏览器 Web HUD，实时监控 Agent 灵魂状态 + 持久记忆。 生态地图直接收录，玩家直呼终于看到 Agent 在思考。 4️⃣ hermes-control-interface （https://t.co/khcp2KHgly） 自托管 Dashboard，一键管多 Agent、进程、记忆和调度。 “把玩具 Agent 进化成生产基础设施”的代表作。 🟢 为什么 Hermes 系这么爆？ 它从不给你黑箱，它给的是可读、可 hack、可自我改进的底层循环。你 fork 它，不是在抄，而是在和 Agent 一起递归自我改进。 这才是 2026 年开源 Agent 的正确打开方式。

+3 more

@omnivaughn My AI can: https://t.co/kiuZ7QXLzb

Introducing the MLX-Benchmark Suite!! https://t.co/sp4ZMIBxov The first comprehensive benchmark for evaluating LLMs on Apple's MLX framework. 🎯 What is this? MLX Benchmark is a CLI tool and dataset that measures how well large language models understand, write, and debug code for Apple's MLX machine learning framework — covering everything from core array operations to LoRA fine-tuning with mlx-lm, mlx-vlm, and mlx-embeddings. 📊 Dataset https://t.co/5b04a7PKAp - 520 questions across 6 task types: knowledge QA, multiple choice, true/false, fill-in-the-blank, code generation, and debugging - 11 categories spanning the full MLX ecosystem: mlx_core, mlx_nn, mlx_lm, mlx_lm_lora, mlx_vlm, mlx_embeddings, mlx_embeddings_lora, mlx_optimizers, coding, debugging, conceptual - 4 difficulty levels: easy → medium → hard → very-hard - 90+ subcategories covering everything from array_creation to lora_finetuning ✨ Features - 🏃 Multi-provider benchmarking — Ollama, Anthropic, OpenAI, Groq, OpenRouter - ⚖️ LLM-as-judge evaluation — strict scoring with an independent judge model - 🔍 Fine-grained filtering — by type, difficulty, and category - 📝 LaTeX export — --latex generates publication-ready booktabs tables - 📈 PNG chart export — --plot generates grouped bar charts comparing models A detailed paper will be coming as well!!!

Introducing the MLX-Benchmark Suite!! https://t.co/sp4ZMIBxov The first comprehensive benchmark for evaluating LLMs on Apple's MLX framework. 🎯 What is this? MLX Benchmark is a CLI tool and dataset that measures how well large language models understand, write, and debug code for Apple's MLX machine learning framework — covering everything from core array operations to LoRA fine-tuning with mlx-lm, mlx-vlm, and mlx-embeddings. 📊 Dataset https://t.co/5b04a7PKAp - 520 questions across 6 task types: knowledge QA, multiple choice, true/false, fill-in-the-blank, code generation, and debugging - 11 categories spanning the full MLX ecosystem: mlx_core, mlx_nn, mlx_lm, mlx_lm_lora, mlx_vlm, mlx_embeddings, mlx_embeddings_lora, mlx_optimizers, coding, debugging, conceptual - 4 difficulty levels: easy → medium → hard → very-hard - 90+ subcategories covering everything from array_creation to lora_finetuning ✨ Features - 🏃 Multi-provider benchmarking — Ollama, Anthropic, OpenAI, Groq, OpenRouter - ⚖️ LLM-as-judge evaluation — strict scoring with an independent judge model - 🔍 Fine-grained filtering — by type, difficulty, and category - 📝 LaTeX export — --latex generates publication-ready booktabs tables - 📈 PNG chart export — --plot generates grouped bar charts comparing models A detailed paper will be coming as well!!!

Stop what you’re doing. This is worth 18 minutes. https://t.co/033UgU9Gm1

Stop what you’re doing. This is worth 18 minutes. https://t.co/033UgU9Gm1

@togethercompute @realDanFu But looking at the reported metrics, the looped 770M model isn’t really close to the 1.3B model. https://t.co/NkFfNjj8eR

For those who made it this far, use code "HERMESAGENT0010" for $10 off a valid Nous Portal Subscription (new or existing). Code valid for the first 250 users. Subscribe to Nous Portal: https://t.co/R6H1nIpuzO Learn more about Hermes Agent: https://t.co/iHmhvv5McL

// Self-Evolving Agent Protocol // One of the more interesting papers I read this week. (bookmark it if you are an AI dev) The paper introduces Autogenesis, a self-evolving agent protocol where agents identify their own capability gaps, generate candidate improvements, validate them through testing, and integrate what works back into their own operational framework. No retraining, no human patching, just an ongoing loop of assessment, proposal, validation, and integration. Why it's worth reading this paper: Static agents age quickly. As deployment environments change and new tools arrive, the agents that survive will be the ones that can safely rewrite themselves. Autogenesis is part of a growing wave of self-improving agent systems, alongside work like Meta-Harness and the Darwin Gödel Machine line, and it's one of the cleaner protocol-level takes on continual self-improvement so far. Paper: https://t.co/3aj9LLjSbk Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX

This is nuts y'all, we've got new tornado warnings popping up everywhere across Iowa and Minnesota. The storms are moving fast and dropping warnings back to back. We are covering all of this live right now, hop on the stream and stay safe. WATCH LIVE: https://t.co/zA4VX48Zhe https://t.co/7LDc79Xjjd

Proud to announce Spatial Desk for Vision Pro available for pre-order with official release 8th May. After a Vision Pro onboarding session with a client - this is the App I wish was available. It's the 'blank canvas' concept. @AppleEDU @Apple https://t.co/xsywkCLWVw @Scobleizer https://t.co/LGcx2dvH6j

+1 more

The Hermes Agent Creative Hackathon starts now 16 Days, $25k in Prizes Presented by @Kimi_Moonshot & @NousResearch For the tinkerers pushing Hermes Agent into creative domains: video, image, audio, 3D, long-form writing, creative software, interactive media and more. Show us what your Hermes Agent can do. Details Below ↓

@NousResearch @Kimi_Moonshot Hackathon time https://t.co/0W4NUREjQS

I'm a slow learner. And a minor in economics. I interviewed @VitalikButerin back when he was just starting. He developed Ethereum. If I was fast I would have bought some. When I interviewed him an Ethereum cost $2. Today it costs a LOT more. Same with understanding the new economic model that's coming. People in San Francisco say "don't worry about jobs, we are headed toward an age of abundance." Hard to get on that train as I see family and friends struggling economically after getting laid off. But this message from @elonmusk made it all click for me. We are, indeed, headed into a world where AI and robots do most of our current jobs. Anthropic's own engineers believe that AI will be able to do half of all jobs within two years, I heard from an investor who recently gave a speech there and asked them. We all feel it and know it. AI is taking away our jobs. Been seeing many messages along these lines on X's AI community. And normies are hating AI for it. We all hate change, even me. The only real answer is not to slow down, but accelerate through this moment of pain. I recently had dinner with a woman who is running a new construction robot company. Her robot can tie rebar 6x faster than a human. And the humans who do that job? They have the highest injury rate in construction, she told me. Robots don't get hurt. If we accelerate we get to the world where AI agents and robots do all of our shitty jobs. Which, let's be honest, almost all jobs suck. I've had jobs all my life and had some of the best ones that humans can ever have (I got paid to walk around Microsoft with a video camera, remember). Even that job sucked. A robot could do that even better. Here is proof of that. Asked Grok to simulate a conversation on this topic between a simulated Robert Scoble and a simulated Elon Musk: https://t.co/OjU9hJcxXE In 10 years when there are millions of Optimus, or other humanoids, walking around, they will be able to do even our most "fun" jobs. Which will leave us to do new jobs. Like setting up shop in a Holodeck that most people don't yet see. Or building new things for robots to do. Like entertainment most people can't yet see either. Grok lays it out. Acceleration is the right answer. Get us through this period of maximum pain fast so we can get on the other side where there is a lot of fun to be had.