I use chatGPT plus. When you ask for academic references about a very specific topic, chatGPT can produce fake ones.
For solving this problem I tried to feed a list of the most relevant article titles I knew, and asked chatGPT to discuss those with me. The list was too long and didn’t fit into one prompt so I break it into two long prompts.
I realized that all the answers I’m getting are only discussing my second prompt and the first list was completely neglected by chatGPT (even didn’t respond to simple exact sentences referencing).
I’m aware the limitation of 4000 tokens in memory of GPT3 model but I assumed openAI could have solved this with embedding for their greatest chatbot in history.
I tried to post my long list on an open-access internet webpage and repeat the link as a search area for answering questions. (I used ResearchGate or Wix blogs) but it didn’t work.
Please share any solution that comes to your mind. (except coding an API and paying per token)?