o3 is really special and everyone will need to update their intuition about what AI can/cannot do. while these are still early days, this system shows a genuine increase in intelligence, canaried by ARC-AGI semiprivate v1 scores: * GPT-2 (2019): 0% * GPT-3 (2020): 0% * GPT-4

Insane

Sick. Can't wait for this (80% of which I don't understand) make it into the next game.

My dad once told me the story of his CS department taking delivery of one of the first computers in his country (Turkey). Nobody knew how to use it, or anyone they could ask for help, so they just re-read the manual literally hundreds of times, in English, until it made sense.

New in Type: – File + url imports - Table support 🎉 - Text extraction No AGI. No agents (yet!). Just useful stuff for people who love writing.

This thing is amazing. Absolutely love the attention to detail.

Sweden stands with Ukraine. You are not only fighting for your freedom but also for all of Europe’s. Slava Ukraini! 🇺🇦

Fun fact: Craft.do does not use AutoLayout. Every new joiner iOS engineer is horrified that this is the case. The founder (my brother, Balint Orosz) gives a challenge to devs inevitably suggesting to move over to AutoLayout: Take a transition in the app, and build it with

We’re about to see AI Agents explode in SaaS. The entire architecture of most modern cloud services is to power multi-user, collaborative workflows, and AI Agents swap in as one of those collaborators naturally. These systems are extremely well suited for automation.

Folks at google are really cooking with gemini 2.5! Was struggling with a bug somewhere in a 16K line file (not my code) and it found it in one shot. No other model would fit it into context. Cursor wouldn't allow me to pass the full file even with gemini selected so had to use

Remember when Satya asked if google would come and dance? Well looks like they're dancing now. 😂

Something seems off with the Gemini 2.5 Pro integration in Cursor. If being fed the right context explicitly, it often performs at or better level than Claude 3.7. But it consistently fails to do agentic lookup for the right context itself, leading to bad code.

Building a chat UI is the classic example of where getting to V1 takes like 2 minutes but then the long tail of micro UX wrecks you. Layout shifts. Virtualization. Growing input box that doesn't cause jump scrolls. Bottom align logic. etc etc. And yeah the AI won't get it right.

UPDATE: big NYC tech week milestone -- we've hit **1000** events in a single week. Amazing. All the events are happening next week. I'll be in town, and hosting a bunch of stuff on behalf of a16z. If you want to check out the calendar, here it is: tech-week.com/calendar Some

Introducing LiveStore (beta) LiveStore is a next-generation state management framework based on reactive SQLite and built-in sync engine.

The hundred billion dollar AI models have made Tailwind semi-usable

Tried giving Claude Code runtime access by writing a simple Chrome DevTools Protocol MCP server (that's a mouthful). It worked shockingly well for the effort needed. Did a writeup here for anyone curious: autoconfig.io/devtools-mcp-s…

I remember switching from android to an iPhone around 10 years ago and being very confused about the "buy" button not being available for kindle books in the Amazon app. It took a decade to get this, and it's not really a buy button, but I'll take it. Apple you greedy bastards.