

10·
28 days agoI’d like to see similar testing done comparing models where the “misaligned” data is present during training, as opposed to fine-tuning. That would be a much harder thing to pull off, though.
I’d like to see similar testing done comparing models where the “misaligned” data is present during training, as opposed to fine-tuning. That would be a much harder thing to pull off, though.
I mean, Cyberpunk 2077 does have construction workers being contractually obligated to receive strength-augmenting implants that are low quality and frequently malfunction and/or drive the wearer to homicidal insanity.