Peter Welinder (@npew) / X

Peter Welinder

2,255 posts

Peter Welinder

@npew

Devices @OpenAI

San Francisco, CA

Joined April 2010

Peter Welinder
@npew
8h
GPT-5.6 Sol is on a roll
Nisarg Shah
@nsrg_shah
19h
1/5 Oh wow. GPT-5.6-Sol seems to be on a whole other level. In just three weeks, it seems to have made nontrivial progress on 10+ problems in social choice. Exciting! (But I'm still processing what the existence of such a model means for research.) A snapshot:
Peter Welinder
@npew
8h
Look at that nice GPT-5.6 Luna/Sol pareto frontier
ARC Prize
@arcprize
16h
We re-tested GPT-5.6 Luna from @OpenAI on ARC-AGI (Verified) following its recent 80% price reduction: - ARC-AGI-2: 59.6%, $0.18/task - ARC-AGI-1: 90.7%, $0.07/task The new results match Luna's original performance at a much lower cost.
Peter Welinder
@npew
8h
Highly recommend watching this talk to understand the alignment challenges ahead.
Eric Wallace
@Eric_Wallace_
15h
Yesterday, my OpenAI collaborator and I gave a detailed talk on the Huggingface incident, our models creating "the message board", model misalignment, and more. youtube.com/watch?v=87DyyM… I hope it can answer a lot of the questions folks have, and we will release a full detailed
Peter Welinder
@npew
Aug 2
Our 9-year old is obsessed with Rubik’s Cube and is building a website for anyone who wants to learn to solve it. Check it out and give him feedback in the replies:
how2cube.com
How2Cube
Solve. Learn. Get faster. Digital or real—with profiles, Championships, eight bots, and a movable digital cube.
Peter Welinder
@npew
Aug 1
10 mathematical breakthroughs for 2 weeks of grad student salary. Hope this will lead to many more breakthroughs.
Noam Brown
@polynoamial
Aug 1
The cost of generating the proofs for all 10 of these breakthroughs combined was under $2,000 at Sol API prices. We’re excited to see what scientists and researchers are able to create with our upcoming Astra models!