3 ms·
LLMs are trained to anthropomorphise themselves — their behaviour and output is what they (in the Bayesian sense of the word) "believe" users would up-vote. Wh
by ben_w 2y ago
LLMs are trained to anthropomorphise themselves — their behaviour and output is what they (in the Bayesian sense of the word) "believe" users would up-vote.
While they might be semi functional cargo-cult golems mimicking us at only a superficial level*, the very fact they're trained to mimic us, means that we can model them as models of us — a photocopy of a photocopy isn't great, but it's not ridiculous or unusable.
* nobody even knows how deep a copy would even need to be to *really* be like us