This seems like a neat idea, but I’d like to flag that this strategy only seems applicable when the fact you are looking for already exists explicitly and at an appropriate explainer level. I’m not sure you can do anything equivalent if you want the LLM to explain, synthesize, summarize, or do original reasoning.
Yeah. I’m imagining that when you need that sort of thing (which to be clear, I acknowledge is “most of what people want LLMs to do”, not like an edge case), the citations are retrieved, the LLM does a synthetization step, but then checks the results of that against the original source.
(and, the source is still provided in a place that’s optimized for “easy to look at without requiring more than 1 and preferably 0 extra steps”)
I just added this question to the post. I rattled this post off quickly, but seems good to amend it to answer more detailed FAQs in response to comment.
This seems like a neat idea, but I’d like to flag that this strategy only seems applicable when the fact you are looking for already exists explicitly and at an appropriate explainer level. I’m not sure you can do anything equivalent if you want the LLM to explain, synthesize, summarize, or do original reasoning.
The way to do it is to excerptsOnly as long as possible then shortCode a bit perhaps then only 0 or 1 or 2 steps of consolidate
Yeah. I’m imagining that when you need that sort of thing (which to be clear, I acknowledge is “most of what people want LLMs to do”, not like an edge case), the citations are retrieved, the LLM does a synthetization step, but then checks the results of that against the original source.
(and, the source is still provided in a place that’s optimized for “easy to look at without requiring more than 1 and preferably 0 extra steps”)
I just added this question to the post. I rattled this post off quickly, but seems good to amend it to answer more detailed FAQs in response to comment.