>>16485627
> Why? In my opinion it gives great high-level overview for concepts that you can then investigate further yourself with the classic literature.

The simple answer is that it has no internal logical system by which it can evaluate whether or not its presentation of the material is correct. The big problem with LLM's and these kinds of things is that "good enough" is the best they can do. They are fundamentally interpolation systems which seek to find the closest in-distribution answer to your question.

If you primarily learn from Chat-GPT or LLM derived summaries, you have no real understanding of what parts are entirely sourced from a human author in the field (i.e., are likely to be accurate to the discipline) and what parts are interpolated based on what is "good enough" but in-distribution. My advisor loves to dummy check these things and you'd be amazed the number of simple subtle mistakes LLM's (even 4o) make around things like dimensionality for matrix multiplication or ensuring that the two statements surrounding a set of equality signs are actually consistent.

Even if 80% of the content is a legitimate summary, you'll never know what parts are hallucination if you don't spend the same time actually checking it with a legitimate source in the field.

Back | Go to board

Board:

/sci/

Thread:

is 25 to late to get into physics/math?