What programming tasks you saw GPT4 fail badly at

Question

It seems on HN there is a widespread belief that LLMs will just be another productivity jump for programmers and not an industry restructuring revolution. My own experiments with GPT4 make me feel that with similar more improvements in GPT (like GPT3 to GPT4), it will for sure have a dramatic impact on the industry.Have you observed otherwise. I would love to see prompts that GPT4 fails badly at. I would also like to hear the views of others on the best case and worst case scenarios for people currently employed in the software industry.

obpe · Accepted Answer

I haven't used GPT4 but my experience with Copilot is that it's just inconsistent. I think a big problem for me is I work in a legacy web environment with a lot of technologies that were never popular. So generating generic code that saves a google lookup is pretty good but anything more complex seems to be a waste of time.
The biggest successes I have seen are where AI does all, or most, of the work. I am not sure an AI could do my job but I do think, in the near future, AI could replace my entire team/department.
However, technology moves fast but business moves slow. I work for a pretty big enterprise that deploys dozens, if not hundreds, of web sites, apps and services. It's not like when AI becomes feasible, all of a sudden all of these things will be AI driven and there will be no need for any further technical expertise. Further, most businesses are not going to invest into a fanciful new technology with a proven cost/benefit model.