Quote:
Originally Posted by sovereignstar v2
How the ai? It have blackjack like TCY 1 ??
|
Slot machine, IIRC. A horribly rigged one designed so you'd always lose.
Going into the text/search thing. Context is everything. ChatGPT is as good as the database used to fill its repertoire. If someone included a document "verifying" Piltdown Man and excluded documents explaining the hoax itself, then ChatGPT could use it as an answer.
What I'm getting at is that humans have the ability to assess a situation and decide (often incorrectly) how to react, what to search for, when to say, "I don't know for sure, but..." That's consciousness. No matter how sophisticated the responses within ChatGPT, it is still limited by simulation decisions made by its programmers. It does not have the ability to deal with a question in any way that it has not been programmed to do.
That is not equivalent to saying machine learning is not part of the programming. Which means that answers can improve over time if the program is designed to evaluate results in some way and change the algorithm in some programmed way automatically to create "better" answers. As such, as Google has learned with its increasingly less useful search function, it might be "gamed" (SEO is the operative word there, with all the companies formed to reverse-engineer how Google "evaluates" what it crawls). Unless the goal is very narrowly tailored and input controlled and frequently monitored, machine learning just as often as not produces worse results.