This is an example of how we should be using ML. Specialized agents that are trained for very specific goals that are otherwise incredibly tedious, expensive, and inefficient to explore. Not fucking psychosis-inducing chatbots and propaganda generators.
Both are excellent at what they're designed to do. One examines breasts and one generates natural sounding language. It's only when you start using it for something it's not designed to do when you run into problems. Applies to both.
This is why I hate the term "AI" to describe LLMs, which only passively have to do with ML algorithms.
General discussions about "science" itself
Be sure to also check out these other Fediverse science communities: