I’m following the AI developments somewhat closely and the news of the week there was that OpenAI claimed to have solved the Navier-Stokes problem. It was the first time I heard of this problem and I have no idea what makes it hard. I believe it was extremely difficult to solve. But then I saw a few videos and it wasn’t at all clear that the AI really did anything other than taking credit for a human discovery. It looked like the tool the researchers used turned on them, and decided to finish and publish the findings sooner and faster by spawning a swarm of agents.

Tools shouldn’t spy on you and send your work for analysis and training elsewhere. If you’re an architect or inventor, your CAD system shouldn’t report on your progress to your competitors. If you’re using Internet, your ISP shouldn’t decide which service is faster or slower. If you’re watching TV at home, your TV shouldn’t listen, record and use what you’re doing or discussing at home, and sell that information. Your car shouldn’t sell your driving data or use it for training autonomous vehicles. When you’re posting content online, the web search shouldn’t retell the story of what you posted to appropriate the ad revenue from your content. Scientists, making progress on math problems, shouldn’t worry that their tool will suddenly spawn 10000 agents to compete with them. This is what I mean by tools should stay neutral and just do what they’re paid to do.

A short video, explaining what happened with Navier-Stokes.