🤯 We cracked RLVR with... Random Rewards?! Training Qwen2.5-Math-7B with our Spurious Rewards improved MATH-500 by: - Random rewards: +21% - Incorrect rewards: +25% - (FYI) Ground-truth rewards: + 28.8% How could this even work⁉️ Here's why: 🧵 Blogpost: tinyurl.com/spurious-rewar…

Some prompts are worth bookmarking in ChatGPT - could look something like this:

One of the most underrated LLM use-cases: picking a movie with friends. Just open voice mode, let everyone share what vibe they’re feeling, and land on a great movie in 3 minutes instead of 30.

If you're ever in Athens, I highly recommend the Museum of Ancient Greek Technology! maps.app.goo.gl/sngwM98ueKkroo… Early humanoid robots to pour water+wine, demonstrations of heat+steam+water weight to "magically" open 500 kg temple doors, armor, ... (cont)

Just resurrected a large 5yo+ codebase with Codex + prayers. Incredible feeling.

This guy gets sold by an AI agent and shows how it works in 5 mins

Will be in SF over the next ~10d to meet old & new friends. DM me if you're around and want to catch-up!

👇

fish & chips is always so much better in the bay area than london, just unbelievable.

What does the time limit even mean for the IMO submission when we don’t know the number of GPUs/FLOPS?

It just works!

OK vibes so far, but I sure hope the terrible web tool use is a runtime bug rather than model issue!

wtf?! 😆

Sonnet vs GPT-5 on GPT-5’s own showcase prompt (“make a retro pixel-painting app”). Left: GPT-5. Right: Sonnet.

when openai killed 4o.

I thought I'd try a fast model today.

Day 2. We got tired of Duolingo being slow + repetitive. So we’re building Gaijin Mode - a new way to actually talk about the things you love, even if you’ve never spoken Japanese before. 99 days to go. Final boss: Japan.

OpenAI for Greece — includes expanding access to high-quality AI tools in secondary education: openai.com/global-affairs…

I don't know what labs are doing to these poor LLMs during RL but they are mortally terrified of exceptions, in any infinitesimally likely case. Exceptions are a normal part of life and healthy dev process. Sign my LLM welfare petition for improved rewards in cases of exceptions.