It would be interesting to chat with an LLM that has been trained on, with increasing weight 1) all the things up to Darwin's death that he might have read, 2) his own library, 3) his own words. It wouldn't sum up to Darwin's views or voice, but it would let us query them in the context in which they were written.
Datomic has a feature that lets you query a database from any given point in time. If an LLM could draw its embeddings from such a database I suppose you could just ask it to respond as of 1882. I'd love to be able to repeat a query once per decade to see how our zeitgeist evolves, like "how do species change?"
Datomic has a feature that lets you query a database from any given point in time. If an LLM could draw its embeddings from such a database I suppose you could just ask it to respond as of 1882. I'd love to be able to repeat a query once per decade to see how our zeitgeist evolves, like "how do species change?"