[precompute]

>the solution is alignment, not censorship

haha, that's very naive. There's already heaps (veritable mountains, even) of information that isn't given to the public on the public-facing instances of ChatGPT, because some info is deemed too incendiary. Filtering out "unwanted" sources of information is already a goal of information labelling, on which these entire LLMs exist. If you were to really make a LLM out of what people really thought and put on the internet instead of the current practice of castration, you wouldn't have techbros wondering about jobs, you'd have a veritable revolution on your hands.