Hank Green examines AI proofs that outrun human understanding
Grant Sanderson of 3Blue1Brown joins him to explain why getting an answer and creating useful mathematics are not necessarily the same achievement.
Why this story? One of the most watched videos since yesterday, with 713,225 views.
AI is Proving Things No One Understands*
Things we can learn from the video
Press to jump
- 1:03Green separates truth-seeking, understanding and practical utility as distinct reasons to pursue scientific research.
- 6:44Solving a problem can matter less than developing mathematical tools that make other discoveries possible, Green argues.
- 20:45Sanderson explains why checkable mathematics and coding answers provide useful feedback for training AI models.
- 30:46Sanderson argues that OpenAI should give more prominent credit to the human mathematics underlying its results.
- 37:05Sanderson considers AI's current mathematical capabilities overrated, but its speed of improvement underrated.
- 45:39Turing's work illustrates how tackling a mathematical question can produce ideas with significance far beyond the answer.
A solved problem sounds like an uncomplicated victory. Hank Green wants to know what happens when the answer arrives without the understanding that makes it useful.
His video, "AI is Proving Things No One Understands*", drew roughly 713,000 views since yesterday. The asterisk matters: the description supplies the word "yet." This is not an argument that humans can never understand machine-generated mathematics, but a conversation about what counts as progress while they are catching up.
Green opens with OpenAI's release of hundreds of AI-generated mathematics preprints. He separates three reasons for doing science: discovering truth, building understanding and producing something useful. Those goals overlap, but they are not interchangeable. As he puts it, "Because you can have knowledge without understanding, this happens all the time."
For mathematics, he argues, the tools developed while tackling a problem can matter more than the final answer. A result about the breakdown of the Navier-Stokes equations, for example, does not automatically transform the practical business of engineering with fluids.
Then Grant Sanderson, the mathematician behind 3Blue1Brown, complicates Green's skepticism. He explains why mathematics attracts AI developers: answers that can be checked provide feedback for training models. He also challenges OpenAI's attribution of results, arguing that the human mathematicians whose work enabled them deserve more visible credit.
Sanderson's assessment is deliberately two-sided: "I feel like the position is overrated, but the velocity is not." He thinks present capabilities attract too much acclaim while the speed of improvement receives too little attention.
The discussion eventually reaches Alan Turing, whose work on a mathematical decision problem produced the idea of Turing machines. For Green and Sanderson, that is the distinction worth protecting: an answer can close a question, while the route to it opens whole new fields. Viewers are taking that idea into their own lives, discussing school grades, the value of struggling to learn and whether an automated destination can replace the journey.
Published on Trari
Where next?
- Channel
- Hank Green
- Subscribers
- 3,200,000
- Channel views, last 7 days
- 4,622,931
- Channel views, pace per day
- 950,507 (on a normal day: 452,393)
- Video published
- October 26, 2026
- Video views so far
- 1,111,801 (2 days online)
- Video views, pace per day
- 713,201 (on a normal day: 783,611)
The numbers
The video added roughly 713,000 views since yesterday. Its first-day comment section was also more than twice as busy as usual for Green's regular videos.
💡 Test your knowledge
Which channel is Grant Sanderson known for creating?
Sanderson is the creator of 3Blue1Brown, the visual mathematics channel named in Green's conversation.


