#llmerrors

Why Models Make Things Up

Last time, we saw how a model's working memory, its context window, can get overloaded, causing long conversations to drift. This time we tackle the most infamous side effect of that same predictive e