“ETH is fake”: The OpenAI paper that panicked crypto

Coin Encyclopedia Improve your knowledge of cryptos

The second market currency has not moved a bit. The ETH of theorists is wavering, and this story is much better than the misunderstanding.

Let's start from the beginning. Among computer scientists, the Exponential Time Hypothesis is a conjecture formulated in 1999 by Russell Impagliazzo and Ramamohan Paturi. According to her, a logic problem called 3-SAT is not solved much faster than trying all the combinations. Hundreds of academic papers are based on it.

OpenAI now claims the opposite. His team describes a deterministic algorithm which counts all the solutions of 3-SAT type problems in so-called subexponential time. The finding dates from October 16, and its verification in the Lean proof assistant from the 17th. The paper revised on October 25 is online on the OpenAI GitHub repository.

And there, the short circuit. On X, a certain Takumi blurts out “Damn algo traders just slaughtered $ETH”. In other words, “algo traders have just massacred ETH”. Andrew Jiang adds: “Vitalik has been real quiet since this came out”. Understand: Vitalik has been very quiet since this came out. Read too quickly, the title sounds like an obituary.

Except that the course saw nothing. Ether was trading around $2,350 this Wednesday, October 28 at 2 p.m. according to CoinGecko, almost stable over 24 hours. Meanwhile, Ethereum developers were dealing with the Glamsterdam fix on staking deposits.

First clarification for ether holders. Ethereum cryptography does not rely on theorists' ETH. The ECDSA signatures are due to the difficulty of the discrete logarithm on elliptic curves. Zero-knowledge proofs are based on yet other hypotheses. A faster algorithm for counting 3-SAT solutions doesn't give anyone your private key. The real substantive subject for blockchain remains the quantum computer.

Second clarification, for everyone. Even if the result holds, it does not say that all NP problems are solved quickly. The P versus NP question, the most famous in computer science, remains open. OpenAI itself writes this in its post. She said the rebuttal “reopens the question of how much faster these tasks could become.” And finding such algorithms “will require significant additional research.”

The series has a sequel. On October 26, Anthropic posted a competing paper, “A 2^o(n) Time Algorithm for 3-SAT.” According to the filing, an internal search Claude model found the algorithm. The paper warns, however, that "the result is purely asymptotic and says nothing about solving formulas in practice." Terence Tao predicted it back in September: “the mere rumor that someone is working on a problem can trigger a massive amount of AI-assisted effort.”

Machine-translated from French