The Big Story

OpenAI released a big collection of new mathematical results created by one of its most advanced internal models. These results tackle open problems in math that researchers have been working on for years. Instead of keeping everything secret, the company put the findings in a public GitHub repository along with computer-checked versions of the proofs written in a language called Lean.

Think of it like this: imagine your phone’s autocomplete suddenly got so good it could finish entire research papers instead of just the next sentence. That’s closer to what happened here. The model didn’t just guess answers — it produced full proofs that can be verified by other mathematicians and by software.

This matters because math is the foundation for almost everything in science, engineering, and even the apps you use every day. When AI starts making real progress on unsolved problems, it could speed up discoveries in physics, medicine, and technology that affect your future career options. For students, it hints that the way we learn and do research might change faster than anyone expected.

You might not solve these problems yourself right now, but the results are public, so anyone curious can look at the GitHub repo and see what the AI actually produced. The company also shared details about how much computing power each result took, which helps people understand what these systems are really doing behind the scenes. Source: openai.com

Explain Like I'm 14

You know how when you’re playing a video game and you have to solve a puzzle, you try one step, see what happens, then try the next step based on what you learned? That’s basically how these AI math systems work, except the “puzzle” is a math problem that might take humans years to finish.

The model starts by breaking the problem into smaller pieces it already knows how to handle. Then it tries different paths, checks whether each path leads somewhere useful, and keeps the ones that make sense while dropping the ones that don’t. It does this millions of times very quickly.

What makes it powerful is that it can also write its reasoning in a special computer language called Lean, which acts like an automatic referee that instantly tells it whether each step is correct. If the referee says no, the model goes back and tries again. Over many rounds it can build an entire proof that both humans and computers can trust.

The surprising part is that the AI isn’t memorizing answers — it’s exploring possibilities the same way a curious student might, just at a scale no human could match in one lifetime. That’s why releasing the actual proofs and the computer-checked versions matters: it lets real mathematicians check the work instead of just taking the company’s word for it.

Cool Stuff & Try This

Game developer turns ChatGPT into a 1,200-piece art factory A longtime artist and game developer used ChatGPT to create more than 1,200 pieces of artwork for a strategy game about airships. The AI handled rooms, icons, enemies, backgrounds, and achievements — work that would have taken the artist years to finish alone. The game is inspired by games like FTL and is planned for release on Steam soon.

If you like making games or just want to see what AI art can do for creative projects, this story shows how one person turned a long list of art needs into finished images in days instead of years. You can try something similar right now: go to chatgpt.com, describe a character or scene you want for a story or game idea, and ask it to generate several variations. Then keep refining the prompt until you get something you like. Source: reddit.com

ChatGPT now adds transcripts to audio uploads for paid users ChatGPT can now automatically create transcripts when you upload audio files, but only if you have a paid plan. This makes it easier to turn voice notes, lectures, or recordings into searchable text without extra apps.

If you record interviews, school notes, or podcast ideas on your phone, this feature could save you time turning spoken words into written form. You can test the idea for free by recording a short voice memo and asking any free AI chatbot to summarize what you said, then compare how accurate it feels. Source: digitbin.com

Quick Bits

Brian Chesky on why AI makes skilled people even more skilled Airbnb’s CEO says AI doesn’t make everyone equally good at things — it actually widens the gap between people who ask great questions and people who don’t, because everyone now has the same tools. He keeps a daily notebook of ideas and uses AI to find patterns across thousands of thoughts each month. Source: reddit.com

Anthropic opens advanced models to more security researchers Anthropic is expanding a program that gives verified security professionals access to its strongest models for defensive work like finding weaknesses in systems before bad actors do. It’s also adding options for authorized testing that helps improve safety. Source: x.com

💬 Reply to this email — Patrick reads every one.

Share: X · LinkedIn · WhatsApp

Forwarded this email? Subscribe here — it's free.