Comment by funnygiraffe

I was under the impression that with LLMs, in order to get high-quality answers, it's always best to keep context short. Is that not the case anymore? Does Claude under this usage paradigm not struggle with very long contexts in ways as for example described in the "lost in the middle" paper (https://arxiv.org/abs/2307.03172)?