Math has a new heavy hitter. For decades, Erdős problems have stood as some of the most difficult challenges in the field, solvable by only a tiny fraction of mathematicians. I was blown away when I saw this update from a savvy industry pro regarding a major breakthrough in this arena.
🤖 The Breakthrough
A quantitative researcher named Neil Somani has successfully solved Erdős problem 397. The fascinating part is that he didn’t do it alone; he used GPT-5.2 to generate the proof. Even more impressive is that Terence Tao, widely considered the greatest living mathematician, accepted and verified the solution.
The creator of this video highlights that the AI took only about 15 minutes to solve a problem that has baffled humans for years. This isn’t an isolated incident, either. Over a recent two-week span, six different open math problems were solved using artificial intelligence.
💥 The Intelligence Explosion
The expert explains that this signals the start of a recursive loop. Once AI can perform scientific and mathematical discovery, it can find more efficient ways to run itself. We saw a glimpse of this when Google’s Alpha Evolve improved matrix multiplication, the very heart of how AI functions.
Here is the cycle the video describes:
- Discovery: AI solves complex frontier math.
- Optimization: Those math solutions improve AI architecture and efficiency.
- Compound Growth: The improved system runs faster and smarter.
- Repetition: The AI solves even harder problems.
💡 Why It Matters
We are reaching an inflection point. The systems being built now are capable of open-ended scientific discovery. The post’s author notes that this applies to everything from curing diseases to designing better servers.
Neil Somani put it best in the video: “Many open problems are sitting there waiting for someone to prompt ChatGPT to solve them.”
📌 Key Insights
- Speed: GPT-5.2 cracked a historic problem in roughly 15 minutes.
- Verification: Top human mathematicians are validating these AI-generated proofs.
- Scale: With enough GPUs and energy, we can parallelize this intelligence to work 24/7.
Check out the full video from the original creator to see the charts and the specific proof details.