On 2025-08-21 13:32, Jonas Smedegaard wrote: >> https://kagi.com/assistant/552738ce-41df-42df-93a7-43d2307e5166 > > Notice the reference to this very mail thread: The LLM might have used > our conversation as basis for its explanation - and if so, it was not > in this case "pretty good", but was simply mimicking. > > (I use Kagi myself, and like how it includes such references to aid in > double-checking some of its reasoning).
Oh, great catch. Kagi does use RAG so it's indeed possible, or even likely, that this discussion influenced its outcome. FWIW, I've had the same pretty good experience with private texts. Here's some lunacy I made up, also including an inconsistency, which it found: https://kagi.com/assistant/ce250e69-f842-4ab9-be66-e87a43473233 Best, Christian
