swyx
@swyx • 162,804 subscribers
achieve ambition with intentionality, intensity, integrity & insanity. affiliations: - @dxtipshq - @cognition - @temporalio - @aidotengineer - @latentspacepod
Shorts
Videos
Wow. GitHub CEO Thomas Dohmke just announced GitHub Copilot X: - Copilot Chat - "ChatGPT-like experience in your editor" powered by GPT-4 - Copilot for Pull Requests - AI-generated descriptions for pull requests on GitHub - GitHub Copilot for Docs - chat for *any* company's repos and docs, starting with React/Azure/MDN - GitHub Copilot CLI - compose commands and loops, and throw around obscure find flags to satisfy your query. as Benedict Evans might say, the incumbents are embracing and integrating AI into existing form factors faster than startups can build them. To win, startups have to create -new- form factors from first principles.
swyx2,052,112 просмотров • 3 лет назад
Am judging TreeHacks Stanford sama’s takes on: - the literal first day of OpenAI - “if you’re a sophomore, you’ll graduate to a world with AGI” - why Matt Schlicht’s Moltbook matters - what you should work on inspiring in how down to earth it sounds but still extremely impactful when done well.
swyx214,061 просмотров • 3 месяцев назад
ChatGPT casually dropped an APP STORE 🤯 It can now: - browse the web (RIP Bing waitlist, cutoff) - write and run Python (RIP replit?) - access org info (RIP docsearch startups) - add third party plugins from OpenTable, Wolfram, Instacart, Zapier, etc) - developer SDK in preview
swyx1,859,059 просмотров • 3 лет назад
🐣 Introducing `smol-developer`! ▸ Human-centric, coherent whole program synthesis ▸ your own junior developer ▸ develop, debug, decompile ▸ open source: ▸ 200 LOC, half english Insights: 💡 100k context can summarize both content and codebases 💡 Markdown is the best prompting DSL 💡 Copy and paste your errors as prompts 💡 Copy and paste curl output as prompts 💡 write CSS animation by describing what u want 💡 GPT4 >>> GPT3.5/Anthropic Claude for codegen see README for more thoughts!
swyx1,590,825 просмотров • 3 лет назад
insane line at the OpenClaw🦞 meetup tonight to see Peter Steinberger 🦞 do his thing
swyx182,249 просмотров • 3 месяцев назад
🤯 with OpenAI Whisper + TTS + GPT4V, you can just permanently share your screen with an AI and talk to it to get help on anything on screen. a generally intelligent assistant for your digital life. with a bit more engineering you could probably feed it extra docs and context based on the currently active tab/app.
swyx🛬 SFO1,051,628 просмотров • 2 лет назад
after 15 years of waiting, the developers of singapore gave up on waiting for the government to get the tech sector going and finally brought SF to SG. great showings from Daytona Tusk Arize AI and Zo Computer tonight (this is the SECONDARY venue for those not at ClawCon SG) ahead of AI Engineer SG
swyx32,147 просмотров • 20 дней назад
ok so let me explain why subagents kill long context Like you can spend $500m building 100 million context models, and they would be 1) slow, 2) expensive to use, 3) have huge context rot. O(n) is the lower bound. Cog's approach is something you learn in day 1 of CS50 - divide and parallelize. Embeddings are too dumb, Agentic Search is too slow. So train limited-agency (max 4 turns), natively parallel tool calling (avg parallelism of 7-8, custom toolset) fast (2800tok/s) subagents to give the performance of Agentic Search under an acceptable "Flow Window" that feels immaterially slower than Embeddings. The benefit of this is threefold: - 8 ^ 4 toolcalls cover a very large code search space. can compound subagent calls if more needed. - predictable cost & end to end latency - subagent outputs "clean" contexts, free of context failure modes like context poisoning and context rot (h/t Drew Breunig ) we originally called this Rapid Agentic Search, to contrast with RAG. but Fast Context rolls of the tongue better. there's 2 other perspectives that are worthwhile, i'll go into below, but just go try it out. here it is on Andrej Karpathy's fastchat
swyx 🇸🇬276,060 просмотров • 7 месяцев назад
OAI’s new ChatKit just rekt like 100 agent builder startups
swyx 🇸🇬265,336 просмотров • 8 месяцев назад
you guys have not felt the agi until you have vibe designed your 6000 person conference website at the climbing gym in between projects without reading a single line of code including 99% video asset performance optimization because why the heck not its 2026
swyx🛬 SFO123,929 просмотров • 3 месяцев назад
finally paid for a Gemini Ultra sub, and tried it out for an unsponsored unsolicited review. it has obvious flaws but... it's here! realtime playable video world model!! here's "arid desert with little tiny human towns here and there and big cliffs and lots of terrain to walk around in" where you're "a giant robot spider like in Wild Wild West". flaws: - it clips thru a lot of terrain. worse than in a game engine. - only too late do i discover you can spacebar and jump on things. - it also sometimes errors out - you can only live in the world for 60 seconds. feels like some kind of black mirror storyline - nothing else moves. ruins immersion. why does every world model lack physics? - prompt to prompt editing can have bad side effects like adding a thing removes a previous thing you liked but: - REALTIME VIDEO WORLD MODEL - instruction following was pretty good!!! - movement acommodates weird legs! - THEY NEVER EVER SHIP GENIE MODELS TO PROD. NEVER. EVER. WHAT TPU SORCERY IS THIS. congrats to the Genie 3 team, i can't believe this actually shipped, what absolute mad lads.
swyx121,819 просмотров • 4 месяцев назад
Inspired by Andrej Karpathy’s words on why you - yes YOU - should work on AI Agents
swyx839,489 просмотров • 2 лет назад
"more than twice the speed" is underselling Haiku tbh built a way to directly compare Sonnet v Haiku 4.5 and it's roughly 3.5x faster, but the UX feels SO much better because Haiku stays inside the "flow window". obviously end to end latency varies a lot so Ant can't report a real number without production usage but you should try heads up comparisons
swyx 🌉171,603 просмотров • 7 месяцев назад
late night low TAM tweet: did a Broadway solo song for the first time in ~18 years! only fumbled lyrics once!
swyx44,516 просмотров • 1 месяц назад
🎉Congrats to OpenAI for releasing o1: - Economics: tylercowen asked o1 basically to write a college essay - Genetics: Catherine Brownstein asked o1 to help her reason through "n of 1" cases - medical cases that nobody has ever seen - Physics: Mario Krenn used o1 to draft and reason through complex quantum physics equations - Code: Hongyu Ren prompted a full snake game and it was generated zero shot, working perfectly, and obeyed instructions to add obstacles
swyx384,689 просмотров • 1 год назад