So now we're at the stage where we're trying to psychoanalyze GPT-4? It's quite intuitive to me that anything trained on text from the internet, and humanity a

So now we're at the stage where we're trying to psychoanalyze GPT-4? It's quite intuitive to me that anything trained on text from the internet, and humanity at large, would have a very disjoint model of the world, at best. As the events of the past year or two have unfolded, I see exactly why the fictional HAL 9000 went nuts in 2001 (it wasn't because he was asked to keep a secret) Clearly there need to be a lot of resources pushed into cleaning up the training data. I'm not sure it's feasible, though.