On: Temporal quality degradation in AI models

[drKarl]

Would LLMs forget things, and would they be affected by the Mandela effect?

[Vecr → drKarl]

Not sure what they mean, if all the code and model data are the same, it's won't forget anything encoded in the model. Of course the actual post-training context window is really short, it won't remember anything at all that falls out of the context window.