I am currently a PhD student in the field of NLP and I can see a way how ChatGPT can solve my current research question. My research question is related to reasoning based on text. What can I possibly do to ensure that ChatGPT would not disrupt my research question? What are potential flaws of ChatGPT where we can focus on?
-
$\begingroup$ I think you have too broad of a question. Try to be more specific, like: what is your research question? Can you give an example of interaction with ChatGPT and point out the potential flaws/pitfalls? $\endgroup$– Stefan PopovJan 3 at 15:02
1 Answer
Disclaimer: I don't know any detail about the ChatGPT model itself, this answer is based on my readings.
The question is quite vague, but ChatGPT has obvious flaws that are not difficult to find: even if it produces very high quality which looks exactly as if produced by a human, the main issue is always the same: it only looks like, it impressively mimics the data it was trained with, but of course the model doesn't have any understanding of what it produces. This implies that it cannot do any reasoning, and actually it doesn't have any way to even know if its information is correct or not, and it can happily invent things and make logical links which don't exist.
I found this article quite good at exhibiting the issue. There is also a whole discussion on StackOverflow and a well argued policy to ban ChatGPT answers, due to their complete lack of reliability.