I certainly can't claim any expertise on the topic Anon, but as I understand it, the transformer simply uses a statistical probability model constructed during it's training phase to make a weighted prediction of suitable following text regarding input strings. This is why it works from an initial 'seed' &tc. And no, it most definitely isn't 'thinking' in any human sense. I'm personally skeptical that will ever
happen, quite frankly.
>anons dont want to fall in love with a database...
Heh, very well-put Anon. I think we'll get there eventually with suitably satisfying simulacrums, just be patient.
>Lol, that's not very optimistic
Fair enough Anon. :^) My estimate is on what we would roughly call an AGI
today, and on second thoughts, I suspect my cost estimates are at least an order of magnitude low
>Within the next 5-7 years we are going to have something decent,
I think you are correct Anon. Certainly something well-enough suited for a 'starter robowaifu' if you will.
>I think as conversational ability degrades we will converge to AGI because people will be simpler than AI's in the current form.
LOL. I think just the opposite Anon, though I'm well aware of both the dynamics you're implying. Humans are profoundly capable and multifaceted. By far the most spectacular of all the natural creatures!