cm0002@lemmy.world to Artificial Intelligence@lemmy.worldEnglish · 2 months agoAnthropic CEO Admits We Have No Idea How AI Worksfuturism.comexternal-linkmessage-square6linkfedilinkarrow-up119
arrow-up119external-linkAnthropic CEO Admits We Have No Idea How AI Worksfuturism.comcm0002@lemmy.world to Artificial Intelligence@lemmy.worldEnglish · 2 months agomessage-square6linkfedilink
minus-squareVagueAnodyneCommentslinkfedilinkEnglisharrow-up17·2 months agoThis bullshit propaganda is intensely frustrating. Stop pretending these models are magic. This propaganda is meant to mystify and fascinate idiots. Here is a short reading list that will explain to a competent person how generative transformers work. https://jaykmody.com/blog/gpt-from-scratch/#fn1 https://explainextended.com/2023/12/31/happy-new-year-15/ https://jalammar.github.io/how-gpt3-works-visualizations-animations/ https://arxiv.org/abs/1706.03762 Enough pretend-along. They aren’t smarter than you.
minus-squarewondrous_strange@lemmy.worldlinkfedilinkEnglisharrow-up3·2 months agoGood on you for saying it so eloquently. None of them would have dared to say anything remotely similar to ‘we don’t know’ if it really was the case
minus-squareKogasa@programming.devlinkfedilinkEnglisharrow-up1·2 months agoThe GPT architecture is well understood, the part that is hard to explain is the way information is encoded in the trained model’s parameters. It’s not magic, it’s just a highly opaque encoding.
This bullshit propaganda is intensely frustrating. Stop pretending these models are magic. This propaganda is meant to mystify and fascinate idiots.
Here is a short reading list that will explain to a competent person how generative transformers work.
https://jaykmody.com/blog/gpt-from-scratch/#fn1
https://explainextended.com/2023/12/31/happy-new-year-15/
https://jalammar.github.io/how-gpt3-works-visualizations-animations/
https://arxiv.org/abs/1706.03762
Enough pretend-along. They aren’t smarter than you.
Good on you for saying it so eloquently. None of them would have dared to say anything remotely similar to ‘we don’t know’ if it really was the case
The GPT architecture is well understood, the part that is hard to explain is the way information is encoded in the trained model’s parameters. It’s not magic, it’s just a highly opaque encoding.