LLM training isn't even remotely brute force[1], even with tons of clever tricks to reduce the amount of compute necessary it's still fundamentally just a really hard problem.

[1]: https://youtu.be/MD2fYip6QsQ