The universal summarizer appears now to be providing information and references to sources beyond the scope of base material when using the "Discuss Further" feature. Kagi appears to be using an LLM, rather than its previous closed-context model.
This is particularly unhelpful when discussing specific academic articles, for example. Erroneous references are cited as though they originate from the base text, without mention of their having been supplied from an external source.
I expect the universal summarizer to constrain its responses to the source material, and not provide (without any communication that it is doing so) reference to or information from extra-source material.
