AI General Thread

Started by Legend, Dec 05, 2022, 04:35 AM

kitler53 and 5 Guests are viewing this topic.

the-pi-guy

Actually another super annoying thing that AI does:

"That's not what I mean, I mean _____"


"You're right, let's lock in to the real heart of the issue." (somehow still not talking about ____ )

Legend

Quote from: the-Pi-guy on May 15, 2026, 08:42 PMActually another super annoying thing that AI does:

"That's not what I mean, I mean _____"


"You're right, let's lock in to the real heart of the issue." (somehow still not talking about ____ )
"Hostile reframing."

So freaking lame. But a great reminder that llms still are clueless about basic convos.

darkknightkryta

Neural nets are going to do what neural nets do.  The faster the business men in charge realize this is just another data structure the better.  Like, they were better off making industry specific neural nets instead of the behemoth general "I know everything" nets which gets its data points crossed.

Legend

Quote from: darkknightkryta on May 18, 2026, 01:35 PMNeural nets are going to do what neural nets do.  The faster the business men in charge realize this is just another data structure the better.  Like, they were better off making industry specific neural nets instead of the behemoth general "I know everything" nets which gets its data points crossed.
The behemoth nets can do cross domain learning however which is very very useful. You might not have enough data to train a net for your specific industry in isolation but a behemoth net can be fine tuned on it just fine.

darkknightkryta

Quote from: Legend on May 18, 2026, 05:32 PMThe behemoth nets can do cross domain learning however which is very very useful. You might not have enough data to train a net for your specific industry in isolation but a behemoth net can be fine tuned on it just fine.
Not worth the money or data centers.  Hell I stopped using Gemini since it cross contaminates conversations referencing irrelevant things.  I spend more time trying to get it to stay on task than anything.  That's the problem with large neural nets.

Legend

Lol. I tried having the new gemini 3.5 help me with a programming thing.

"I could use wave function collapse, but is there something better for this specific thing?"

Gemini: "yes use this algo, a million times better!"

"Ok it's fast but it doesn't make sense, these two values are the same."

Gemini: "yes it can't handle stuff like that. Why, was that important?"

:P