For one month beginning on October 5, I ran an experiment: Every day, I asked ChatGPT 5 (more precisely, its “Extended Thinking” version) to find an error in “Today’s featured article”. In 28 of these 31 featured articles (90%), ChatGPT identified what I considered a valid error, often several. I have so far corrected 35 such errors.
Wait, you mean using Large Language Model that created to parse walls of text, to parse walls of text, is a legit use?
Those kids at openai would’ve been very upset if they could read.
Even for that it’s mid at best. I try using co-pilot at work often and it makes shit up constantly.
Chatbots aren’t the worst use case, too, even though we are headed in a wrong direction.