Hi - I’m looking at the reflection example provided in module 2 eg. fact checking an essay by providing external information. The guidance is to provide authoritative sources to fact check the output of step 1. If authoritative sources are known, wouldn’t it be better to build this constraint into the original prompt to either prioritize or only use the provided authoritative sources?
Hello @deeplycurious,
That’s an interesting thought - why wouldn’t we provide those authoritative sources before step 1 but step 2? I think you have already pointed out a very critical condition for this to be done, which is that the sources are known. If we are building a chatbot for specifically answering question about a product, I think we can provide those upfront, but for a general chatbot, we can’t and we rely on “something” to process the user query to find out what sources to look for. What is that “something”? LLM! We instruct it to work on a draft and to propose the sources for fact-checking.
Cheers,
Raymond
Also the idea behind separating the source check point in reflection by tool agent is probably to differentiate/check how llm response for a prompt versus how llm response with a source checkpoint.
What you mentioned can surely be also used, and ofcourse it is used, but module in question is dividing each step of
prompt==>reflect==>generate code==>refine the output. So the both approach are understandable, achievable and resourceful.
Regards
Dr. Deepti