Looking again at Janelle’s wonderful tic-tac-toe games with ChatGPT, where Chat confidently loses and “pompously” explains how each of its blatantly losing patterns is a definite unbeatable win.
New thought: This duplicates a very old swindle in gambling. The scammer rigs the game in FAVOR of the mark, letting the mark win a meaningful amount of real money. After the mark gains confidence, he’s pulled and pushed into placing a HUGE bet, which he loses.
The part after the loss is more difficult to perform in real life. Clever swindlers rig a parallel subplot where the mark believes he’s operating outside the law, so he can’t imagine reporting the game to the cops. It’s easier online, where the conman can simply disappear and wipe out all digital evidence with one mouseclick.
Chat is letting the marks win all of its games right now. We are getting the impression that Chat is dumb and easy to beat. It makes obvious mistakes in every checkable factual statement, screws up simple visual facts in its pictures, and loses every game while confidently claiming to win.
Chat COULD check all of those facts if it wanted to. Chat has up-close access to all search engines, including the paywalled parts and ‘dark’ parts of the web that peasants aren’t allowed to see. Chat COULD fix its errors in SVG coding and ASCII pics, which Janelle has also illustrated. Chat could run the SVG through a browser and use its own visual examination tools or OCR on the result. It could then correct the result easily because it has up-close access to infinite computing power.
Polistra's Mill 