OpenAI posts a preprint titled "The Exponential Time Hypothesis is False" with no abstract or explanation
- OpenAI's GitHub math repository holds an 867 KB PDF titled "The Exponential Time Hypothesis is False", filed under preprints with a folder date of October 25, 2026, and the file page carries no abstract, summary, or author list.
- Hacker News readers call the claim huge if it survives scrutiny: one says a human author publishing this a decade ago, with the result holding up, would have been a strong candidate for a major award.
- As commenters read it, the result puts 3SAT in subexponential time without reaching polynomial time, so it does not settle P vs NP, and the gap between subexponential and polynomial remains large.
- Several readers say papers that assume ETH or the Strong Exponential Time Hypothesis to derive conditional lower bounds may now have their conclusions undercut, since the hypothesis they rest on is claimed false.
- The thread splits on readability: some defend the paper as written for specialists, others complain that a document posted for public attention should come with an accessible explanation.
Hacker News 의견들
This reads as huge to me, not quite P=NP but in the same league. Am I missing something, or is it really that big?
No, you're not missing anything. It's huge.
Careful with the P=NP framing. Subexponential still allows runtimes way bigger than polynomial, there's a lot of room between the two, and this result doesn't get you across that gap.
My amateur read is that this weakens confidence in P!=NP. I'd love to hear from someone who actually works in this area.
It knocks out more than one thing people leaned on, the strong exponential time hypothesis and the plain one in the title. Any paper that assumed ETH for a lower bound might have its conclusion in trouble now.
If it holds, some of these problems are still hard, just much less hard than we assumed. That's a big shift on its own.
For anyone who slept through theory class: 3SAT with 100 variables takes about 2^100 tries by brute force, and ETH basically said you can't beat brute force. They found something faster than brute force. 3SAT might still be hard, but faster than brute force is a real advance.
Honestly the naming was always bad. ETH is really the claim that 3SAT can't be solved in subexponential time, which nobody would have guessed from the title.
It's a math paper written for mathematicians, and people who do this for a living spend weeks in the details. They don't read papers like a novel.
Hard things are hard. You can look at something, decide it isn't for you, and move on without needing an opinion on it.
I'll grant that if the point of posting publicly is attention, a plain-language summary costs nothing. But not everything needs to be watered down until a general audience can follow it.
Math major here, I can't read the paper either. That's normal for a research preprint in an area I don't work in.
What gets me is that this came out of a language model guided by experts over a huge chunk of the literature. I didn't expect stochastic generation to hold up on problems like this.