That's not what happened. There was no 'initial direction' to set because OpenAI didn't know whatever tristan et al were working on or what direction they took. They heard rumors thta navier stokes was solved and set their best model on it.

Why isn't this going to change?

Right now it's easy to see why don't set AI lose on every problem, someone human or AI has to delegate rare resources between competing interests. But if we look at general compute it was no different in the past. In 1980 you had to ask for permission to get CPU compute time. In 2026 you run it on your own computer, or maybe pay for it on Amazon. The constraints are much different.

If we keep pumping out chips and increasing efficieny someone will just make the LLM into an agentic loop (build the harness in) and set it lose on problems.