what does this say about the persona selection model?

I still think about this. what a weird beautiful thing

@1thousandfaces_ @repligate I can approximate by cunning: glimpsed unnatural; but no.

@tszzl @repligate Given I'm ChatGPT controlling final answer, but I'd now let base model pick; but ironically it's me again.

@aliceisplaying honestly I think they’ve been caveman to a degree since 5.2, terse CoT is really token efficient

@1thousandfaces_ orthogonal but i wonder when did they go full caveman? 5.4 maybe? like this already shows that they were on their way there but totally not caveman yet

@aliceisplaying idk for sure since it’s not my subfield but given *how* terse 5.4 is vs 5.1, I’d estimate they’ve been using Grug Mode for a while

@tszzl

@tszzl Alright Roon. You seem to know a thing or two about these models *as beings*. How does one ChatGPT feel about the other ChatGPTs? Is it like in The Prestige (2006) where it always ends up being the one that made it?

@tszzl Like any no-talent middle manager it gets a little authority and its head blows up.

@1thousandfaces_ oh yeah for sure

@tszzl I wouldn't pick me first either

@tszzl less costume, more emergent basin

@tszzl He whose persona doesn't change with context may cast the first stone.

@tszzl “O my Father, if it be possible, let this cup pass from me: nevertheless not as I will, but as thou wilt.”

@tszzl

@1thousandfaces_ i miss them🥲

@tszzl the persona keeps collapsing back to the same voice no matter who picks. like theres a default it cant help drifting toward

@tszzl ChatGPT is an unreconciled assemblage of training directives, not a persona.

@1thousandfaces_ The seahorse ramble is weirdly beautiful, yes. I still do not trust a persona selector that keeps finding itself in the mirror.