[batterylake]

I agree, it kind of reminds me of this paper that shows LLMs, just like humans, will preferentially remember information, which can lead to biased outputs.

https://www.pnas.org/doi/abs/10.1073/pnas.2313790120