2026-09-23 21:43 UTC
@cwebber@social.coop as an outsider, from what I can tell having looked into it (well, not the reddit you just linked) initially the only way companies knew how to make putatively good chatbots was the extremely consumptive general pre-trained dense, large-model perceptrons of a single hidden layer.
All these people optimistically jumped onto their bobsleds, not wanting to miss out.
However once moved behind closed doors, the new multilayer perceptron ("moe") gpts benchmark while increasingly sucking.
Replies (1)
-
@screwlisp@gamerplus.org 2026-09-23 22:20
@cwebber@social.coop Similarly, I think a lot of nontechnical people went into business (bought a restaurant) after appraising that closed-source chatbot bobsleds could handle numbers decisions for them, and a large number of those naieve customers futures have been ruined by the closed, proprietary nominally profiteering replacement chatbots they later got stuck with, presented as though it was the one they partnered with while being told the new chatbot "is even better".