Early content generation before the RLHF safety tuning was a lot more interesting and nuanced. It's significantly worse now, both dumber and less diverse output

Early content generation before the RLHF safety tuning was a lot more interesting and nuanced. It's significantly worse now, both dumber and less diverse output. [1] We're seeing the enshittification of early AI products in real time, presumably so they can be sold to the lowest common denominator who can't distinguish AI output from human-embodied thought and opinion, and so that companies like micro$oft can safely sell advertising space in/alongside it, or whatever their plan is. [1] https://www.youtube.com/watch?v=qbIk7-JPB2c https://www.youtube.com/watch?v=qbIk7-JPB2c - mentioned fairly early in the talk that the "sparks of AGI" are not reproducible with the release version of GPT4.