
If you have 6.2-Astra work backward from a verifiable goal it's the best model in the world rn

Just a person, mostly. Model design & behavior @ OpenAI, co-founder of RescueTime, YC ‘08 alum, also likes 🎺 and 📸.

If you have 6.2-Astra work backward from a verifiable goal it's the best model in the world rn

It’s the best of times, it’s the worst of times.

I love 6.2 Astra but it’s too powerful for just vibes. Give it your most ambitious work or a verifiable and gnarly week long task and let it rip. Use 6.2 Sol Ultrafast for vibes.

Nihilism never really was much fun anyway.

My community jazz band welcomed me back and it’s super good for my work. Shaping and orchestrating AI feels like improv soloing. You know where you’re going and you quote things that work but the journey is a musical ride. (Not sure my artist friends will appreciate this analogy though rn)

Recursive self-improvement is a critical issue, but the rhetoric on the subject is becoming more poorly calibrated over time. As we approach new stages of AI development—and RSI is one of them—we need more precise frameworks and terminology. I fear the opposite is happening.

Whatever your politics, it's hard to be anything other than happy to see people can and will still fight back against insanely bad ideas and choices.

Bad news: somehow I messed up my Duo shipping address. Good news: My dot was nice to Barry 🥹

For AGI to benefit all of humanity, it should help people with the everyday tasks that matter most to them. So we built LifeVal, a new benchmark shaped by feedback from thousands of ChatGPT users on where they most want help from a personal assistant—from finding childcare to accessing government services to managing personal finances. Over the past year, we’ve used LifeVal to measure and improve how well our models handle these everyday tasks. openai.com/index/lifeval/

By popular request, you can now text your dot. Give it a task to take off your plate or something to keep an eye on, using text, photos, or even voice notes. It will text you proactively with relevant updates. Safe, secure texting via iMessage or RCS. You’re the only one who can text your dot, and your dot can only text you. Rolling out now as a preview for Pro users in the US.

I told one of my agents to visualize my pipeline in a creative and interesting way as task boxes and now it always lets me know when there's an update to "our little box factory!" I love the future.

Just added a "Orchestrator Assistant" that I forked off of the Orchestrator so I can ask questions without interrupting its work.

Current multi-agent recipe: Start a task in a Codex session to initialize a project. Give it all the docs and context it needs. Fork off "Specialist" sessions for areas you want to track progress in separately along with a "Coordinator" session that can manage/report on the specialists. Astra can do this automatically but manually splitting into focus areas keeps things legible for reporting purposes. I also usually have the Coordinator make a kanban board to keep me updated on tasks.

One of my pet peeves is when people dismiss what ai does as “pattern recognition.” Not because I think ai is doing something grander, but for the opposite reason: intelligence IS pattern recognition (ie compression).

there’s a genre of tech criticism i’d call black mirror brain: watch a product demo, invent a person, give them a tragic backstory, ruin their life with the product, then get furious at the company. meanwhile the product is helping someone manage their cat’s medication.

A weird amount of my work now is running a disk space cleanup crew on my laptop to make sure all of my local agents have room to do things.