There’s a distance between getting helped by an AI and actually getting something done. One thing that ChatGPT can do nicely is things that have been done often, but that are hard to find and remember. It’s indexing well lots of nuances in drawing charts, first steps of GUI, launching queries to database, wide range of typical data science algorithms, basics of computational mathematics, and such. Quality drops when leaving the comfort zone. It was fun to exchange messages on Liouville numbers and get clean code to reproduce the results in the article “A Numerical Investigation on Cumulative Sum of the Liouville Function” from 1980, but it seemed to totally fail chatting about Clebsch graph counting repeatedly some essentials wrong. Obviously there are other software that does it easy, but maybe you could say that currently some horizon is about there.
