Llm

Personal News I got accepted into the Claude Code hackathon last week. 13,000 people applied, 500 got in. The hackathon just wrapped up today and I built Robo.app . Check it out, and watch the 3 min pitch video ! Here’s what caught my attention from Feb 10-16, 2026: 🚀 OpenAI Hires OpenClaw Founder for “Personal Agents” Peter Steinberger joining OpenAI to build next-gen personal agents - Sam Altman personally welcomed him. OpenClaw hit 100K GitHub stars in 2 days (fastest any repo has ever done that), now at 197K+. Both Meta and OpenAI made offers. The fact they’re building a dedicated personal agents team, not folding him into ChatGPT, signals a new product line. OpenClaw will transition to an independent open-source foundation that OpenAI will continue to support. ( @sama ) 🏋️ Three Major Model Launches in One Week Qwen 3.5-397B released - 397B total params, 17B active via hybrid linear attention + MoE. 87.8 MMLU-Pro, 83.6 LiveCodeBench v6, 76.4 SWE-bench Verified. Decoding 8.6x-19x faster than Qwen3-Max. Apache 2.0, 201 languages. Already on OpenRouter . ( @Ali_TongyiLab ) MiniMax M2.5: Sonnet-level coding at 1/20th Opus cost - 230B MoE, 10B active. 80.2% SWE-Bench Verified, $0.30/M input tokens. OpenHands confirmed it’s the first open model to beat Sonnet on their benchmark. Stock jumped 15.7%. If you’re running coding agents, test this as your workhorse model. ( VentureBeat ) ByteDance Seed 2.0 / Doubao 2.0 - Claims GPT-5.2 and Gemini 3 Pro level at ~1/10th cost. Doubao app has 155M weekly active users. China’s frontier labs aren’t 6 months behind anymore. ( @QuanquanGu ) 🌐 WebMCP: Every Website Becomes an API for Agents Google and Microsoft co-authored a W3C spec called WebMCP - Websites can now declare capabilities as structured tools via navigator.modelContext. Two modes: declarative (HTML forms self-describe) and imperative (JS exposes custom tools). 89% token efficiency improvement over screenshot-based agents. Available in Chrome 146 Canary behind a flag. ( @aakashgupta ) Pietro Schirano built a DoorDash agent demo - Adds items to cart and checks out without ever parsing UI. This is what “agents that talk to apps” looks like. ( @skirano ) Links: Chrome blog | W3C spec draft ...

Here’s what caught my attention from the past 48 hours (Feb 6-7, 2026): 🤖 Opus 4.6: The Agent Model Arrives Claude Opus 4.6 launches with agent teams, fast mode, and 1M context - SOTA on agentic coding benchmarks. Can plan before acting, catch its own mistakes, and coordinate parallel agents. This is the model designed for tasks that span hours, not minutes. ( @claudeai ) Agent teams are experimental but available now - Set CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS and spawn multiple agents that coordinate via shared task lists and a mailbox system. I’ve been running these all weekend. ( @bcherny ) Anthropic built a C compiler with agent teams - They mostly walked away. Two weeks later it compiled the Linux kernel. ( @AnthropicAI ) Fast mode: 2.5x output speed - Same model, faster inference. $30/150 MTok (50% discount through Feb 16). Toggle with /fast in Claude Code. Heads up: switching mid-conversation costs more because you pay fast mode price for the full context. ( @claudeai ) 🏎️ GPT-5.3-Codex Drops 57% SWE-Bench Pro, 76% TerminalBench, 64% OSWorld - 40% faster and <50% token usage vs 5.2-Codex. Mid-task steerability means you can course-correct while it’s working. The model was used to build itself. ( @sama ) Codex crosses 1M users - Dedicated macOS app, integrated into Xcode 26.3, rate limits doubled for paid plans through April. ( @sama ) 🐝 Agent Swarms Go Production Reverse-engineering Claude Code to save $200/day - Stan Girard figured out Claude Code’s filesystem-based agent protocol and built a REST API + TypeScript SDK around it. Uses his $200/month Max subscription instead of burning $200/day on API calls. Anthropic hasn’t commented. ( @_StanGirard ) Local agent swarms on a single GPU, no internet - Qwen3-Coder-Next on one GB10 GPU. 100 tokens/sec generation, 17,871 tokens/sec read, 256k context. No cloud, no quota anxiety. ( @iotcoi ) Anthropic PMs don’t write PRDs anymore - They use Claude Code to build the first version, dogfood it for weeks, then ship. This workflow took them from $1B to $9B run-rate in 12 months. ( @aakashgupta ) 💰 The SaaS Reckoning Goldman Sachs: SaaS has a 5-year shelf life - Application software will hit $780B by 2030 but agents will capture 60%+ of the economics. Current SaaS multiples are wrong. ( @aakashgupta ) OpenAI launches Frontier: AI coworkers for enterprise - Agents that connect CRM, data, ticketing, and internal tools. Sam’s framing: “People will manage teams of agents to do very complex things.” ( @sama ) “Ask what your token budget will be” - Ethan Mollick says if you’re considering a job offer, ask about your token budget. AI access is now a comp consideration. ( @emollick ) 🌟 The Tweet That Says It All Tobi Lütke endorses Pi over every funded AI coding tool - Pi has four tools: Read, Write, Edit, Bash. No MCP, no sub-agents, no plan mode. Can write plugins for itself as you use it. Shopify’s CEO calling a solo dev’s project “the most interesting agent harness” is a statement about simplicity as strategy. ( @tobi ) 📋 Honorable Mentions Qwen3-Coder-Next - First open-weight coding agent model, 80B params, trained on 800K verifiable tasks Cloudflare R2 Local Uploads - Up to 75% faster uploads, <100ms reported Cloudflare Queues on free plan - 10k queue ops/day included in Workers free tier X API pay-per-use - New Console, XDK, and MCP for indie builders. I set this up and it’s working great. Xcode 26.3 gets Codex + Claude SDK - Apple sees agentic dev workflows as table stakes ESLint v10 - Flat config only, eslintrc is gone Postgres 18 OAuth 2.0 - OAuth for database auth ElevenLabs raises $500M at $11B - Voice AI is a category now Excalidraw MCP App - Natural language diagrams in Claude chat via MCP Apps protocol dbt Agent Skills - 9 auto-loading skills, key insight: MCP = tools, Skills = knowledge Nvidia Parakeet - 83ms transcription on M3, 26x faster than Lightning Whisper Chatterbox Turbo - Open source voice cloning in 5 seconds, <150ms latency Anthropic Startup Program - Up to $25K in credits, no VC required Ethereum ERC-8004 - Agent payment standard by Ethereum Foundation, MetaMask, Google, Coinbase Waymo World Model - Photorealistic simulation for rare edge cases, built on Genie 3 Claude Code /insights - Analyzes your past month of usage and gives workflow suggestions OpenRouter free router - Routes to free LLMs matched to your request awesome-llm-apps - 92.6k stars, copy-paste code for 9 categories of LLM apps This curated summary covers 182 liked and bookmarked tweets from Feb 6-7, 2026. The meta-story: we’re moving from “AI as assistant” to “AI as autonomous workforce” faster than most realize. ...

I was paying $240/year to host this blog. Now I pay $0. Nexcess was a solid WordPress host, but overkill for a casual wordpress blog. I’d been meaning to migrate to a static site for years, and I finally pointed Claude Code at the task. What Happened I described what I wanted. Opus 4.5 did the rest: Converted 60+ posts to Hugo markdown Preserved all URLs for SEO Set up the theme and config Configured Cloudflare Pages deployment My actual input time: ~5 minutes. The AI handled documentation, code, and debugging. ...

It continues to be quite overwhelming to keep following all the updates in AI. It’s both exciting and stressful because just when you think you’re going to have a quiet week, all the major AI providers release significant updates that rock the entire world that you previously knew. Here are the biggest AI updates that have happened in the last few weeks, with sources so you can learn more about each innovation: ...

Update 3/31/2025: I have made changes to my MCP setup since posting this. I open-sourced the changes here . Anthropic launched the open-source MCP (Model Context Protocol) on 11/25/24 and it’s a total game-changer! This kicked off a fundamental shift in how technology connects – similar to how we evolved from physical network connections to APIs, we’re now moving to automatic connections between systems using AI. Do you prefer to listen instead of read like me? Lucky for you I used NotebookLM to create a podcast out of this post. Listen below: Listen to this post Your browser does not support the audio element. ...

Prefer to listen to a podcast about this post? Listen to an AI-created podcast of this post here: Listen to this post Your browser does not support the audio element. Ever tried finding decent coloring pages for your little ones on Google? If you have, you were probably disappointed. The free options out there are often shockingly poor quality. But fear not! As AI image generation keeps improving by the day, I figured now’s the perfect time to dive in and create some awesome coloring pages from scratch with AI via Midjourney . ...

Forget tedious editing – imagine transforming meeting notes, group chats, or email threads into engaging podcasts in minutes. Do you prefer to listen instead of read like me? Lucky for you I used NotebookLM to create a podcast out of this post. Listen below. Listen to this post Your browser does not support the audio element. What is Google NoteBookLM? Launched in 2023, NoteBookLM is an AI-powered tool that lets you upload any document, transcript, PDF, or even a website URL. The AI processes the content and within minutes, delivers a podcast with two AI hosts discussing the material. ...

I used the LLM fusion technique to create conversation-starting questions for kids in different age groups. Below is a table of the questions, and I then show the prompts I used to generate them. Questions for your kids: AgesQuestion3-4If you could turn into any animal at school today, what would you be and what adventures would you have?3-4Can you show me your favorite dance move from today and tell me what made it so special?3-4What magical powers did you use in your pretend play today?3-4If your toys came to life while you were at school, what do you think they did?3-4Can you draw a picture of the yummiest thing you smelled today and describe it to me?5-7If you could invent a machine to help your teacher, what would it do and how would it work?5-7Imagine you found a magical door at school. Where did it take you and what happened next?5-7If you could create a new rule for your classroom, what would it be and why?5-7What was the kindest thing you saw someone do today? How did it make you feel?5-7If you could be the teacher for a day, what would you teach the class and why?8-10If you could create a new subject to be taught in all schools, what would it be and why is it important?8-10What’s a challenge you faced today that you’re proud of overcoming? What strategy did you use?8-10If you could have any superpower for one day, how would you use it to help someone in need?8-10What’s a current event you learned about recently that surprised you, and how did it change your perspective?8-10If you could design a treehouse classroom, what features would it have and how would it change the way you learn?11-12If you could design a class to teach adults about teenage life today, what topics would you cover and why?11-12What is a goal you set for yourself today, and how are you planning to achieve it?11-12If you could create a social media platform that positively impacts mental health, what features would it have?11-12What does it mean to be a good friend, in your opinion? How do you try to be a good friend to others?11-12If you could change one thing about the world, what would it be and why?13-14What is an issue at school or in the world that you feel passionate about, and why?13-14If you could have a conversation with any historical figure, who would it be and what would you ask them?13-14What does success mean to you, and how are you working towards your own definition of success?13-14How do you think technology is changing the way we learn and communicate?13-14What is a risk you took recently, or a time you stepped outside of your comfort zone? What did you learn from that experience?15-18What is a belief or idea you have recently questioned, and what led you to rethink it?15-18If you could design your own course to teach others, what would it be about and why?15-18How do you think your actions can make a positive impact on your community or the world?15-18What are your biggest concerns about the future, and what steps can you take to address those concerns?15-18If you could create a business that solves a global problem, what would it do and how would you scale it? ...

AI Facebook AI: major updates to their open-source LLM model llama 3.1 and released the S egment Anything Model 2 (SAM2) . Runway ML launched Gen-3 Alpha Image to Video and it’s incredible ChatGPT finally started rolling out advanced voice mode (preview only) AI search feature (preview only) Gadgets and Gizmos July was a bad month to work on my gadget addiction: AI assistant hardware: pre-ordered a wearable friend (that’s the actual name!) pre-ordered limitless pendant Sparkling Water: They got me with the Amazon Prime Day deal on the SodaStream Enso (~$160). I am impressed at how easy it is to use and the fact that it doesn’t require any power or cables. Stud finders: Finding studs in ceilings and walls continues to be my least favorite part of my home projects. I got the Walabot DIY 2 which uses radar scanning to see through the walls. It’s cool, but the jury is still out on whether it is any more accurate than my Franklin Sensor M210 ($60) which I got earlier this year and love. I also finally added an awl to my stud-finding toolkit. Charging devices: Added Anker 15W wireless charger to desk ($46). I wonder if anyone other than Matt Mullenweg thinks about charging apparatuses as much as I do. If you read this far I’d be remiss to not point you to my earlier post about 2024 travel gadgets I like. If you can recommend any must-have gadgets, please leave a comment. ...

Leopold Aschenbrenner is a young entrepreneur and researcher who recently founded an investment firm focused on Artificial General Intelligence (AGI), after working on the Superalignment team at OpenAI. Graduated as valedictorian from Columbia University at age 19, Aschenbrenner has a background in economic growth research and an interest in securing the future of liberty, particularly through the advancement and responsible development of AI technology. Key Takeaways on the Future of AI from Leopold’s ~160-page Situational Awareness paper (available as a webpage and PDF ). ...