tfarny
Poster Extraordinaire
ChatGpT isn't continually trolling the web, reading responses. It is a static Large Language Model that has about 1/3 the library of congress scanned into a database. That's what it works from when responding. If you ask it what happened in Poughkeepsie last week it has nothing to say. But future versions could certainly do that and a bunch of other stuff.Has anyone considered that ChatGPT is “reading” this thread and has already “learned” something about what it is getting wrong? I was surprised to see more output that mentioned safety—after several people mentioned it didn’t talk enough about safety. The feedback loop we are creating really is hard to comprehend.