disagree. These are ultimately still do what I say interfaces. You need to be specific and intentional and they respond in pretty exact alignment to what you ask it to do. They're basically unconventional programming languages.

You can ask absurd things to gpt and it will try to respond to your absurd request.

For instance, I asked it "What year in the 1900s had the most tuesdays" and it spit back a python program that almost emitted the correct answer. On Google, to compare things, I get "1900s" crossed out and the wikipedia entry for the Ruby Tuesday restaurant chain as the first result, maybe because it's around dinner time.

The difference here is between that and the "guess what I mean based on crude demographic information and popularity" interfaces that ignore the user's clear intentions in favor of gross statistical markers like how google changes the position of images, shopping, maps, etc in the results based on the query presuming your intention based on crude vague guesses or the search systems that seem to only return 50% of what you asked for and the other 50% is simply what it thinks you want to see instead.

Those are not tools, they are broken trash. It'd be like if you had a knife that randomly turned into a spoon or a fork based on what time of day it is and what room you're using it in.

ChatGPT tried to do exactly, precisely what I said regardless of the fact that there is no answer since it's a list of years and not a single one. It's the "you told me to do X and I did exactly X".