Egregoros

It's not like I never gave AI a try at all but I did NOT realize they were still THIS BAD in 2026. I really gotta be more careful.

This is grok output with thinking.

>$186 million tokens
>$tokens
????

>$400k / $2 = 186
????


I repeated this prompt a couple other ways and it gave figures between 1 and 2 billion tokens for the cost of a strix halo setup capable of running GLM 5.1, for those wondering what I was doing.

what the fuck is this
just set up a 4x m4 Mac mini cluster

yeah I said that too. I replied with buy a strix or something and it fucking thought I meant a 5090 💀

this nigga grok knows nothing!

if OpEx isn't an issue, get a decent 4U box that takes E-ATX and stuff some pro cards in it for like a fifth of the cost

>pro cards a fifth of the cost of mac minis or the strix
one of us has got to check their math.

a fifth of that 400k

Nigga I know 400k is way too much, we're past that part

I wasn't

well okay. So my point with the reply to the mac mini thing is that either that or the strix halo run GLM 5.1 (big frontier model) for WAY CHEAPER than 400k. So that part was already fucky from grok.
That's not the point of the OP post tho, the point is that the output looks humanlike on first glance but is total nonsense. STILL. in 2026

@WandererUber @professionalbigot69 ai proompters won't tell the difference