In depth explanation + examples of why sometimes you don't need rag (especially with gemini 2.0)

video up on youtube too

It's simpler, but still costs a lot more. Maybe break down 1 query into several prompts like "is management optimistic about iphone sales next Q" and feed that to a "risks"prmpr and an "mda" prmpt, retrieve more chunks then feed those chunks to Gemini to get a more rounded answer

Would you rather pay more and have a accurate answer ? I think the cost equation starts to matter less and less

Thanks for sharing!

amazing explainer video, Sully. one thing that you've missed, which is HUGE: **context caching**[1]. basically, now you can send this whole 50k transcript, cache it for subsequent calls, and never ever bother breaking this down for RAG, even for long conversations, because this whole thing will get cached. I am switching most of my LLM use to Gemini after this. [1]:

Agreed I didn’t even mentioned which makes it way cheaper I will say googles caching needs work

KAG vs RAG

great video, i just don't understand how google was able to achieve this without crazy hallucinations

Thanks Sully! Great work

im extracting information from bank statments and need it REALLY accurate what would you recommend?

Gemini