Re:Stack Overflow Moderation Strike Update
by comex on Sunday June 11, 2023 @11:36PM
(#36288167)
I have never asked a question on Stack Overflow; I am too impatient. However, I frequently find answers on Stack Overflow through Google.
The problem is that, when I have "how do I do X using Y API"-type questions, I usually consult the API's official documentation (or sometimes the source code), since it's generally more complete and up-to-date. The questions that take me to Google are usually either obscure errors or other hard questions. Even Stack Overflow answers are often irrelevant or outright wrong. ChatGPT is considerably less trustworthy.
As a random example, today I Googled 'topological sort cyclic graph'. By definition a topological sort can only be done on an acyclic graph, because otherwise it's impossible to avoid violating some ordering constraints, but I wanted to find algorithms for minimizing the number of ordering violations. Among the Google results, there were a few true-but-unhelpful answers ('you can't'), a few wrong answers ('here's some code to do it', author doesn't seem to really understand the problem)… and finally a few answers, all on Stack Overflow, referencing the "feedback arc set problem" and "minimum feedback vertex set problem". Those were the magic keywords that allowed me to learn more, including the fact that minimizing the number of violations is NP-hard.
I just tried asking GPT-4, and it gave an answer that had some correct parts but also had significant errors, and did not include the aforementioned magic keywords.
The problem is that, when I have "how do I do X using Y API"-type questions, I usually consult the API's official documentation (or sometimes the source code), since it's generally more complete and up-to-date. The questions that take me to Google are usually either obscure errors or other hard questions. Even Stack Overflow answers are often irrelevant or outright wrong. ChatGPT is considerably less trustworthy.
As a random example, today I Googled 'topological sort cyclic graph'. By definition a topological sort can only be done on an acyclic graph, because otherwise it's impossible to avoid violating some ordering constraints, but I wanted to find algorithms for minimizing the number of ordering violations. Among the Google results, there were a few true-but-unhelpful answers ('you can't'), a few wrong answers ('here's some code to do it', author doesn't seem to really understand the problem)… and finally a few answers, all on Stack Overflow, referencing the "feedback arc set problem" and "minimum feedback vertex set problem". Those were the magic keywords that allowed me to learn more, including the fact that minimizing the number of violations is NP-hard.
I just tried asking GPT-4, and it gave an answer that had some correct parts but also had significant errors, and did not include the aforementioned magic keywords.