OpenAI Claims Its AI Refuted Exponential Time Hypothesis

OpenAI says its AI has disproved the exponential time hypothesis, the long-held conjecture that the hardest instances of the 3-SAT problem require exponential time to solve. If confirmed, the result would challenge a major limit assumed in theoretical computer science. It would not, however, settle the famous P versus NP question.

The proof has been formalized in Lean and published on GitHub, but it still needs independent review, notably by the two researchers who originally put forward the hypothesis. Until that happens, the claim remains unverified.

Claude Code's token drain

In another AI story this week, Flamerix AI released a script that disables the automatic summaries in Claude Code, which it blames for 15 to 20 percent of the tokens consumed. According to the firm, Claude Code generates a label of just a few words every 30 seconds by sending the model the full history of every active agent.

Anthropic confirms the mechanism but puts its cost far lower: roughly 2 percent through the API and 6 percent in the terminal. The company has started rolling out a fix that relies on caching, which should cut the waste without users needing a workaround.

Musk's Odyssey

Elon Musk, meanwhile, has relaunched Grok Imagine to produce an AI-generated series that he describes as "historically accurate", drawn from Homer's poem and pitched against Christopher Nolan's adaptation of The Odyssey. Musk has publicly attacked Nolan's casting, claiming the choices reflected diversity requirements imposed by the Academy.

The seven-minute first episode, produced with Wonder Studios, showcases progress in AI video generation. It also exposes frozen animations, a lack of physical presence and sharply divided reactions.

Together, the three stories show how far generative AI now reaches: into pure mathematics, into the everyday costs of developer tools, and into cultural rivalries. The first claim will hinge on expert verification, the second on how quickly Anthropic's fix lands.