Get the latest tech news
Here’s a new way to lose an argument online: the appeal to AI
Not even authority, just the signifiers of authority
Systems like ChatGPT, as anyone familiar with large language models knows, predict likely responses to prompts by generating sequences of words based on patterns in a library of training data. There is a huge amount of human-created information online, and so these responses are frequently correct: ask it “what is the capital of California,” for instance, and it will answer with Sacramento, plus another unnecessary sentence. I also find it depressing that no matter how many of these there are — whether it’s fake presidential pardons, bogus citations, made up case law, or fabricated movie quotes — they seem to make no impact.
Or read this on The Verge
