Here’s the thing though, when you keep having to “prompt it”, are you really doing the work yourself by coaxing a good enough bullshit answer that you’ll believe it? It could still be useful, not sure.
A minimum sort of knowledge of the field would be helpful.
But I feel that if one is inquisitive enough, they'd know when exploring a topic to ask broad questions of the sort:
- what are the common types of problems encountered in X field
- what are the main roles and responsibilities of personnel in X field
- list some open questions in the field of X
After these open-ended questions one can dive into each of them with:
- what are some common solutions to X, compare and contrast X vs Y
The risk of seeing hallucinated info is always there, but just for an example when I was exploring the types of space-time metrics and asking chatgpt which tensors are most suited for which scenario (and also which coordinates are best for which scenario, such as moving from Schwarzchild to Eddington Finkelstein coordinates to deal with the event horizons near black holes), their individual wiki pages said generally the same things.
I was able to ask ChatGPT in very naïve "idk please explain like im 5" ways and it was able to very patiently rephrase dense jargon into surprisingly understandable explanations.
I'm sure if an expert in GR were to drill down deeper into the details they will find inaccuracies in ChatGPT's responses, but for the inquisitive information sponge user, a dive through ChatGPT to get a general feel of a topic is insanely useful. After that, one could dive into the appropriate books/papers/primary sources to start the real learning once they find what they're interested in.
But I feel that if one is inquisitive enough, they'd know when exploring a topic to ask broad questions of the sort:
- what are the common types of problems encountered in X field
- what are the main roles and responsibilities of personnel in X field
- list some open questions in the field of X
After these open-ended questions one can dive into each of them with:
- what are some common solutions to X, compare and contrast X vs Y
The risk of seeing hallucinated info is always there, but just for an example when I was exploring the types of space-time metrics and asking chatgpt which tensors are most suited for which scenario (and also which coordinates are best for which scenario, such as moving from Schwarzchild to Eddington Finkelstein coordinates to deal with the event horizons near black holes), their individual wiki pages said generally the same things.
I was able to ask ChatGPT in very naïve "idk please explain like im 5" ways and it was able to very patiently rephrase dense jargon into surprisingly understandable explanations.
I'm sure if an expert in GR were to drill down deeper into the details they will find inaccuracies in ChatGPT's responses, but for the inquisitive information sponge user, a dive through ChatGPT to get a general feel of a topic is insanely useful. After that, one could dive into the appropriate books/papers/primary sources to start the real learning once they find what they're interested in.