Re: [OT] ChatGPT, a new online AI

new topic     » goto parent     » topic index » view thread      » older message » newer message

I just had to come back to this, after reading this critique.

All the GPT machines are string sorters, corelation analyzers, they do not have a clue about themselves, and don't have a clue when they give out bad data. As good as IBM's Watson was at Jeopardy, at least it was written to scan for valid and applicable data before replying, while the GPT machines don't do that.

I got a copy of the transcript of a few of Watson's Jeopardy games way back then, and manually ran the "answers" thru Tiggr's data. She agreed with Watson on all but one reply. I don't remember the exact wording, but it was about an archeology dig and the people involved. Watson flatly said person.1 was credited with the results of a dig, while Tiggr's data agreed he did work in-person-lead on that dig, he was in fact working for person.2, who initiated the dig and oversaw it from a different dig site. The two competing humans missed the answer-question totally.

The goal of GPT systems seems to be replacing humans, and it could be argued a great many do need replacements, that wasn't my goal with Tiggr. I deleted all the OE code for her, but i still pull up a db or two occasionally for an answer.

Kat

new topic     » goto parent     » topic index » view thread      » older message » newer message

Search



Quick Links

User menu

Not signed in.

Misc Menu