Yes, people have likened pre-LLM Internet content to low-background steel.

If in the hypothetical future the continual learning problem gets solved, the AI could just learn from the real world instead of publications and retain that data.