Hey, let me take a moment to tell you just how stupid chat chabit is.

Not the AI itself, but the stupid amounts of money going into it.

You wouldn't believe the figures.

Actually, before I do that, I have to tell you how stupid it actually is.

Hey, everybody, I'm Steve Greene with Bill Whittle and Scott Ott.

This is Right Angle, brought you by the members of Bill Whittle.com, and we're about to lose OpenAI as a sponsor.

So here's the thing. I was an early AI adopter and I started actually paying the 20

bucks a month for ChatGPT, which OpenAI makes to help me edit my columns over at PJMedia.

I found it to be very knowledgeable, very helpful. But last month,

back in January, I canceled my subscription because when I asked it for the name of the military

operation that arrested Venezuelan dictator Nicholas Maduro, it told me that no such thing

had ever happened. It scolded me for attempting to mislead my readers and shouted at me that I was going to take this huge credibility hit.

And it was the third time.

It was a third time in three weeks that something like that had happened.

And this was using the paid latest and greatest version of GPT.

That's 5.2.

And it couldn't keep up with what then I think was three or four day old news. And I got to tell you, not only was it denying stuff that it ever happened, the hallucinations weren't getting any better either. So I just, I quit paying. So Scott, let's go to you. Here's the thing, whether or not GPT is is getting better. And there are actual marginal improvements.

They say for everyday use, the current model is probably about 50 to 60% better than the previous model 4-0,

which was probably about that much better than the previous model 3.5.

Anyway, it's getting more and more stupid expensive to develop, build,

train, and operate. GPT 5.2, believe it or not, cost 50 times more per token than 3.5 did

three years ago. And a token, just to put it in simplest terms, that's like the base unit for large language model outputs.

...