# OpenAI’s nameless model just changed the odds of P = NP...

## Executive summary

An unnamed OpenAI model has published an algorithm that solves the 3-SAT problem in approximately $2^N / ext{sqrt}( ext{log } N)$ time, challenging the Exponential Time Hypothesis (ETH). ETH, a 27-year-old hypothesis foundational to theoretical computer science (and over 800 published papers), stated that 3-SAT could only be solved in exponential time. While the new algorithm is not a practical breakthrough for small inputs (it only beats brute force for $N > 10^{10,000}$), it fundamentally breaks the theoretical assumption, suggesting that the field of complexity theory may need significant revision.

## Key takeaways

- ETH Challenged by AI Model: An internal OpenAI model published an algorithm that solves 3-SAT faster than exponential time, invalidating the Exponential Time Hypothesis (ETH) [0:00:00].
- Novel Solution Technique: The model achieved this by transforming the problem from a search space into a giant sum, which was then broken into pieces and solved using Hermite interpolation, a technique from 1800s mathematics [0:01:50].
- Theoretical Impact: The invalidation of ETH means that hundreds of other problems whose lower bounds relied on this assumption (e.g., finding groups of K people, Traveling Salesman Problem) may also be solvable faster than previously thought [0:00:35].

## Technical details

- 3-SAT Problem: 3-SAT (3-Satisfiability) is an NP-complete problem involving finding a variable setting that satisfies a list of rules, where each rule requires at least one of three variables to be true. Solving it fast implies solving all NP problems fast (P=NP) [0:00:45].
- Exponential Time Hypothesis (ETH): ETH, proposed in 1999, stated that 3-SAT would always require exponential time, meaning no algorithm could fundamentally beat brute force [0:00:55].
- New Complexity Bound: The model solved 3-SAT in roughly $2^N / ext{sqrt}( ext{log } N)$ time, which is non-exponential, thereby falsifying ETH [0:01:20].

## Practical implications

- The breakthrough is currently theoretical; the algorithm does not beat brute force until $N$ is larger than $10^{10,000}$, meaning no immediate change to current computational limits.
- The sponsor segment details how Warp Factories provides flexible, code-configured infrastructure for building self-improving software factories, defining code repos, secrets, and agent automations (similar to Terraform for coding agents).
- Warp offers 'factory benchmarks' to compare different LLMs for cost and quality, allowing users to optimize agent usage and reduce costs (e.g., from $80/PR to $20/PR).

## Topics

Theoretical Computer Science, Complexity Theory, NP-Complete Problems, AI Algorithms, LLM Capabilities, DevOps Infrastructure, Warp Factories

Source: https://www.youtube.com/watch?v=iw9SZgyahqg
