Both QA Retrieval Chain and QA Generation Chain, are using the same llm model. If I use the query generated by QA Generation Chain in QA Retrieval Chain to get the answer, then the predicted answer will be same/highly related as answers generated by Generation Chain.
Then there is no meaning for evaluating real answers against the predicted answers.
Am I right? If not, please correct and make me to understand in better way.
if the retrieval and generation chain are created with similar prompt task then perhaps they are doing same work, but I haven’t looked at the codes yet but most probably generated chain comes with a bit different code prompts to predict an answer and also the ability of a programer to make changes on What kind generated lang chain output one is looking.
like if I am using a same llm, say Gemini and I used retrieval chain to find how can I make my biryani more delicious, chances are their answer might be not be same based on retrieval providing all the answer the Gemini has and the generated chain might or might not give the same output based on a summarisation of the retrieved chain and the prompt task used in creation of generated chain.
Can you point to which lesson you got this doubt from, so I can look at the codes.