Do LLMs have Proustian moments?

LLMs are famously unpredictable in what they can and can't recall at will from their training data. Proustian moments refers to the involuntary recollection of memories due to some sensory experience. Do LLMs have Proustian moments, where 'sensory' here is related to prompting or finetuning? Of course there are many papers showing ways to prompt or finetune more training data out of a model, but these seem like the more obvious case. I'm talking about cases where some key similarity drives the ability to verbalize recall: a similar tone, a single phrase, an image with a certain object.

LLM memory, to me, is still far more alien than LLM embodiment of lack thereof.