I doubt that it would be useful for most newcomers to try to compete with GPT/Claude, etc for pure text LLM's now.
If someone is just starting AI research for something like a PhD or startup, now, I think it'll be more useful to get familiar with robot simulation framework, such as Nvidia Omniverse.
While there's a lot of competition around humanoid robots, I'm sure there are plenty of more specialized possibilities. Maybe some agricultural machine, maybe medical, mining, etc.
If someone is just starting AI research for something like a PhD or startup, now, I think it'll be more useful to get familiar with robot simulation framework, such as Nvidia Omniverse.
While there's a lot of competition around humanoid robots, I'm sure there are plenty of more specialized possibilities. Maybe some agricultural machine, maybe medical, mining, etc.