nicole m

@mazasiel @OwainEvans_UK I actually thought about, isn’t this already the case to a certain extent? Maybe we haven’t measured it but if we look at the behavior of LLMs today, we can probably see congruencies between LLM outputs and human behavior/tendencies that aren’t obvious in the training data

@OwainEvans_UK @OwainEvans_UK your paper makes me wonder if human-> model transfer is possible through the same mechanism. i know- absurd. but maybe? maybe?

Our paper on Subliminal Learning was just published in Nature! Last July we released our preprint. It showed that LLMs can transmit traits (e.g. liking owls) through data that is unrelated to that trait (numbers that appear meaningless). What’s new?🧵