OpenAI says it just refuted the Exponential Time Hypothesis. That’s a big deal in computer science, if true. The kind of thing that would normally spark immediate collaboration, verification, and follow-up work across the research community.

Instead, it sparked frustration.

According to The Verge, researchers are “growing increasingly tired of OpenAI’s antics.” And honestly? They should be.

Here’s what happened. At the start of October, OpenAI dumped hundreds of mathematical preprints on the research community with little warning or context. Now, less than a month later, they’re announcing breakthroughs in math, computer science, and physics while claiming they have “an evolving approach to communicating its findings.”

Evolving is generous. Chaotic is more accurate.

The Problem With Playing It Both Ways

OpenAI seems to want credit for doing serious research without actually doing serious research the way it’s supposed to work. Real scientific breakthroughs don’t get announced via press release while the technical details remain locked behind corporate walls. They get published, peer-reviewed, replicated, and built upon.

The Exponential Time Hypothesis matters because it’s a fundamental assumption in computational complexity theory. If ETH is wrong, it changes how we think about what’s efficiently computable. That’s not just academic navel-gazing. It has implications for cryptography, optimization, and a lot of practical computing problems.

But here’s the thing: extraordinary claims require extraordinary evidence. And the research community can’t properly evaluate extraordinary evidence when it arrives in dribs and drabs, filtered through PR departments, with researchers expected to just take OpenAI’s word for it.

This Isn’t How Science Works

The scientific method has a pretty straightforward requirement: show your work. Not because scientists are suspicious or difficult, but because that’s how knowledge actually advances. Someone makes a claim, others check it, try to break it, attempt to replicate it, and build on it if it holds up.

OpenAI’s approach breaks this process. They want the credibility that comes from scientific breakthroughs without the accountability that makes scientific breakthroughs meaningful. They want researchers to engage with their work, but on OpenAI’s timeline and terms.

That’s not science. That’s marketing.

The October preprint dump is a perfect example. Hundreds of papers, all at once, with mathematicians left to sort through the mess and figure out what’s actually novel versus what’s incremental versus what might be wrong. It’s the research equivalent of forum-sliding, overwhelming the community’s ability to properly evaluate any single claim.

Why This Matters Beyond Academia

You might think this is just inside baseball, researcher drama that doesn’t affect the rest of us. You’d be wrong.

OpenAI is one of the most influential AI companies in the world. When they claim breakthroughs, it affects investment decisions, policy discussions, and public understanding of what AI can and can’t do. If those claims can’t be properly verified because OpenAI treats peer review as optional, we’re all flying blind.

And look, I get it. OpenAI is a company. They have competitive pressures, they want to protect intellectual property, they need to balance openness with business strategy. Sam Altman has made it clear that OpenAI isn’t primarily an academic research lab anymore.

Fine. But then don’t expect the research community to treat your announcements like peer-reviewed science. Don’t get indignant when researchers push back on claims they can’t verify. And don’t dump hundreds of preprints on the community and then act surprised when they’re “tired of your antics.”

The Trust Problem

Here’s what really bothers me about this pattern. Trust in AI research is already strained. We’ve got companies overpromising capabilities, cherry-picking benchmarks, and hyping incremental progress as revolutionary breakthroughs. The field desperately needs more rigor and transparency, not less.

OpenAI’s communication strategy does the opposite. It erodes trust by making verification difficult, overwhelming researchers with volume over clarity, and treating serious scientific claims as marketing opportunities.

The irony is that OpenAI probably did do interesting work here. Models that can solve novel problems in mathematics and theoretical computer science are genuinely impressive. But we can’t properly evaluate how impressive because the company won’t let us look under the hood in the systematic way that scientific evaluation requires.

What Should Happen Instead

The fix is straightforward: if you want credit for scientific breakthroughs, do science properly. Publish detailed papers with sufficient information for replication. Give the research community time to actually review and respond before announcing world-changing claims. Engage with criticism and verification attempts in good faith.

Or, alternatively, be honest about what you are. If you’re a product company that does some interesting research on the side, own that. Don’t demand the prestige of academic science while dodging the responsibilities that come with it.

Because right now, OpenAI is trying to have it both ways. And the researchers who would normally be OpenAI’s biggest advocates, the people who could verify and extend this work, are instead getting tired of the games.

That should worry OpenAI more than any theoretical computer science problem, solved or unsolved.