Personal update: I'm excited to be joining Meta! I'm deeply grateful for the opportunities I've had at Ai2 over the past 6 years (including three paper awards in the last two years). Onward to the next chapter! 🥳

TPUs might be seen as the single best investment google has ever made

Vibe coding is here to stay. I'd been worried it might be a fad, but I talked to the founder of an infrastructure company who's in a position to see how well vibe-coded apps are doing, and he said a lot of them are making money.

Top Component Libraries I can't live without - ui.aceternity.com - magicui.design - cult-ui.com - animata.design - headlessui.com - ui.lukacho.com - eldoraui.site - nyxbui.design

have 2 shopify apps to sell $11.7k MRR $20k MRR stable revenue, very sticky, not too needy support + dev included have bigger fish to fry and no time to dedicate to the apps lookin for 3.5x ARR, no nego can sell together or separate and looking for quick transactions dm!

end-to-end encrypted collaborative docs and office suite

A tiny plug-in memory lets frozen LLMs adapt to domains quickly, with low cost and no retriever. Memory Decoder is a small transformer that learns the behavior of a retriever, then plugs into any base model that shares the tokenizer. It does not touch the base weights. At

We lived this at Cars.com in 2019 There’s only 1 right answer Elixir

Cloudflare’s DX just got a massive upgrade: Alchemy, a TypeScript-native IaC library that actually makes sense. alchemy

Introducing Codespace A virtual computer that runs custom Claude Code on server. It is using 2 models: Kimi K2 + Claude Sonnet 4 in a sync. I achieved same code quality with 2x better reasoning and 53% less model cost. Access it today CodeGuide (500 credits for everyone)

This is how offline calibration + PPO compares against GRPO on helpfulness BT rewards. Would be curious to see how this might help your usecases.

Brutal, but incredibly fair feedback from my editor, Mr. Claude 🥺 I’ll make you proud one day..

Mario Zechner Cool, I have been working on something very similar github.com/nibzard/agentp…

Appears to be a breakthrough result but no mention of Nagarajan (Vaishnavh Nagarajan ) and his insight into pitfalls of next token prediction. arxiv.org/abs/2508.08222

If you're wondering how to do this, v0 has an API to white label its tech inside your app. Bring your own files & biz logic too. DM me if you're evaluating building this at your company, Vercel will help you scope it out.

This is a real video

I have an unpopular opinion. Claude code in my terminal is a far superior experience to the equivalent model (even in max mode) in Cursor agent mode. Here’s what I do in my IDE that I don’t do with Claude code - i write underspecified prompts and instructions because I have the

Considering getting an iPad air for reading articles, ebooks, watching videos. Just feels more convenient solution than Mac. Thoughts and feedback from people who have done it?

it's wild to me that i can put up PRs like this in like a solid 5-hour coding session whole new world / doesn't get old

Hi. I have thoughts. The short answer is that the App Router is easy, but not simple. It's easy to get started, but complexity ramps fast. I've said previously that if you don't need streaming, then the App Router might not be a good fit. Here's a random list of places where