Alex Graveley
@alexgraveley
Co-creator of GitHub Copilot, Dropbox Paper, AI Tinkerers, Hackpad, MobileCoin, Minion AI, etc. Working on @PerplexityComet. Survivor 🎗️
Many of you asked for custom shortcuts. Shipping next week!
shortcuts for repetitive tasks rolling out next week on comet. more invites will be sent next week too. the browser is going to be your personal console for getting work done.
Perplexity x Tatiana Moodboard #7! The code: --p kuwwd66 The link is in the thread. A sref combo we used to create these images: --sref 257047628 --profile l3h4vio --sw 500 --stylize 500
Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains 'We introduce Rubrics as Rewards (RaR), a framework that uses structured, checklist-style rubrics as interpretable reward signals for on-policy training with GRPO. Our best RaR method yields up to a relative…
OK, Perplexity’s Assistant in the new Comet browser is good. Really good.
BREAKING 🚨: Comet browser is the first AI assistant which can distribute itself and onboard more users to install it! More Comet invites below ☄️
Have you checked your inbox lately? Comet invites are going out daily to the waitlist.
It’s surprising there’s not more breakout apps per year due to AI coding tools. Outside those tools, have there been any?
Working at a well-funded AI startup where you're able to try stuff *before* worrying about cost is an amazing blessing.
How to train a State-of-the-art agent model. Let's talk about the Kimi K2 paper.
Next level: Restructuring existing codebase to minimize claude code errors, and a tool to assess.
I open sourced Sniffly, a tool that analyzes Claude Code logs to help me understand my usage patterns and errors. Key learnings. 1. The biggest type of errors Claude Code made is Content Not Found (20 - 30%). It tries to find files or functions that don't exist. So I…
My bar for AGI is doing my taxes. Including gathering the materials, requesting necessary info from IRS, e-filing, checking and notifying me when return deposited.
A striking thing about OpenAI's IMO gold math model is how terse it is, it really tries to express itself in single tokens. Often breaking the rules of grammar and spelling to do so. They say compression is intelligence. We may be seeing a totally novel way to do compression…
🔥 At ICML 2025, we’re delighted to introduce BFCL V4 Agentic. As function-calling (also called tool-calling) forms the bed-rock of Agentic systems, BFCL V4 Agentic benchmark focuses on tool-calling in real-world agentic settings — including: 🔍 Web search with multi-hop…
Awesome finding!
Albert's excellent blog post on "model alloys" – a clever technique for combining the strengths of different models without making extra queries – is live! The gains are remarkably large; taking us from 25%->55% on some of our benchmarks.
🤯
I just used the Comet browser by @perplexity_ai to pay my Virginia state employment tax bill 🤯
Comet actually just a good browser
I really love it btw. Not even sure why. Not using many AI features. It's just a blazing fast browser that works really well. Native ad block and all. Beautiful.
Check out this DM I got after someone tried out Perplexity Comet 👀
Well-deserved praise for Daylight!
Joe Rogan talking about Daylight was not on our Bingo card until at least 2027 much love to @KONCRETE ☀️🫶