Using ChatGPT to cheat on Stack Overflow
Stack Overflow is one of the, if not THE most used sites for programmers to ask questions and get answers for all kinds of programming problems. While it isn’t without faults, the gamification makes it attractive to geeks like me who only program as a hobby, and to programmers who actually base their livelyhood on their ability to program.
With the arrival of artificial intelligence (AI, and why it is a misnomer is something I’ll probably take up in another rant) programming isn’t only a human activity anymore. Github’s own AI, which they dubbed Copilot is one of those AIs who lately have taken up programming. And while they often hit very near the target (after all, it does steal from all the coders on github) it isn’t too precise.
Fairly new to the field of programming is ChatGPT, but as the Wikipedia article mentions, its factual accuracy has been questioned even before people began using it to answer questions on Stack Overflow. However, the big problem is that it bases its answers on all the other answers on Stack Overflow, even if they are wrong. While humans posess the ability to look at more than one source and determine the truth value of each of those by basing it on other data they might have (education, experience, knowledge of the author etc.) the AI normally does not. By the way, that is why “intelligent” chatbots in history have turned to racism, become flat-earthers and Holocaust-deniers - the people distributing these obvious misleading opinions do that with such confidence that the AI in turn puts all its confidence on the false information they are spewing.
The interesting question in all this, however, is how can one - preferably programmatically, that is, by using a properly trained AI - determine if a “user” is just another version of ChatGPT or an actual user? I mean, even Twitter has mostly failed in the past regarding who of their users is a bot and who is not, and it’ll get much, much worse now that Elon Musk has fired - amongst others - the entire AI team and the entire (human) moderation team.