Safe to assume you like the original? I know it's not a comparison review, but I'm curious as hell how they, well, compare

"LLMs are just tools!"

Oh really? When was the last time you tried to hide the fact that you used a spell checker? When was the last time you pretended to be able to hammer in a nail with your bare hand?

People don't try to hide the fact that they're using tools.

Innnnnteresting

Honestly this is making the bsky experience considerably more fun

Gung vapyhqrf jung gurl frr, urne, fzryy, rgp. Nf gb jul gung gevttref frvmherf/pbzn, V unir ab vqrn. V fcraq n ybg bs gvzr jngpuvat guvaxvat nobhg ubj gurl qrpvqr ubj gb orunir nccebcevngryl gb erprvir bcgvzny/zbfg cebqhpgvir erfcbafr sebz Pneby.

V guvax jr unir qvssrerag pbaprcgvbaf bs ubj gur pbzzhany cebprffvat jbexf - V'z ivrjvat vg nf fgevpgyl bar zrtnoenva gung erprvirf qngn sebz rnpu obql ng rknpgyl gur fnzr gvzr, fb lryyvat ng (be pbzzhavpngvat ng nyy jvgu) bar crefba vf rknpgyl gur fnzr nf qbvat fb jvgu rirel bar bs gurz.

Isn't it just that

Vs bar bs gurz pna urne Pneby be gur bguref, gura nyy bs gurz pna, vafgnagnarbhfyl? Fvapr gurl funer pbafpvbhfarff, jbhyqa'g vasbezngvba erprvirq ol bar or vafgnagnarbhfyl erprvirq naq cebprffrq ol nyy?

To unpack a lot of the AI BS, it would be useful to isolate "learning" (as it is used by industry) from deliberately incentivized behavior. Models don't self-optimize, they optimize toward goals set by designers and patterns designers steer them toward based on data designers give them.

I'll die on this hill - operating generative AI in any form should require a license for which you have to pass a test demonstrating you understand what a probability engine and a GAN are, at the very least. And more specifically what they aren't.

(feeling pretty good, thanks)

Billy!

They're hurriedly generating AI videos announcing cuomo's win just to buy some time