Majestical-psyche@alien.topB to LocalLLaMA@poweruser.forumEnglish · 1 year agoLlama 3 will be released in the first quarter of 2024?alien.topimagemessage-square32fedilinkarrow-up11arrow-down10file-text
arrow-up11arrow-down1imageLlama 3 will be released in the first quarter of 2024?alien.topMajestical-psyche@alien.topB to LocalLLaMA@poweruser.forumEnglish · 1 year agomessage-square32fedilinkfile-text
minus-squarestddealer@alien.topBlinkfedilinkEnglisharrow-up1·1 year agoPutting “safety” mechanism in foundational models is dumb imo. They are not just text generators, they are statistical models about human languages, and it shouldn’t have made up arbitrary biases about what language should look like.
minus-squareapi@alien.topBlinkfedilinkEnglisharrow-up1·1 year agoIt’s not hard to fine tune base models for any bias you want. “Zero bias” isn’t possible. There’s always some bias in the training data.
minus-squarestddealer@alien.topBlinkfedilinkEnglisharrow-up1·1 year agoSure, but that’s not a reason to purposefully add more biases into it.
Putting “safety” mechanism in foundational models is dumb imo. They are not just text generators, they are statistical models about human languages, and it shouldn’t have made up arbitrary biases about what language should look like.
It’s not hard to fine tune base models for any bias you want. “Zero bias” isn’t possible. There’s always some bias in the training data.
Sure, but that’s not a reason to purposefully add more biases into it.