I'm trying to implement a PPT content generator using the GPT-3.5-turbo-16k model from Azure OpenAI. The client wants to have a Regenerate option included in the UI, so if the client/user feels the PPT content generated is not good enough they will hit regenerate option. The expectation is the regenerate option should not repeat the same content but it should be more with technical points and in bulleted fashion.
PPT Content generation specifications: If its a short ppt - It can have upto max 4 slides and its intended for Sales/Business SME's team. If its a Long ppt - It can have upto 12 slides max and its intended for CXO's/Leadership team.
I know we can tamper with the temperature parameter, but apart from that is there any other option to regenerate the content ?
I would suggest prompt engineering.
Generate similar contextual queries from the one given. Rank the new queries by similarity to the first query and send the top three (or how many you want) to the ppt llm.
Sorry, I don't have a source, it is just an idea off the top of my head.
Ask a new query
I don't think it's wrong to ask a new query. LLM's can only go so far, if a users query is too general. They should just be told to reformulate in my opinion.
Edit:
I had to look up some source for option 1:
https://medium.com/@lokaregns/masked-language-modeling-with-hugging-face-transformers-a-beginners-guide-b36af74a7560#:~:text=Masked%20Language%20Modeling%20(MLM)%20is%20a%20pre%2Dtraining%20technique,to%20predict%20the%20masked%20tokens
Look up all the nouns in the query and replace each with a mask then generate contextual close relatives with an MLM and reformulate the query.