Thanks for a great course. I put the learning to work and create a AI Cat goddess Vtuber and have started doing some live stream tests with chat input.
Stream recording 20230503
I have a few questions about how to get better results
-
Would adding multiple system inputs for different information I,e Output format system, back story system…
-
Temperature seems to affect the output format when asking for JSON in a specific style. Is there a way to only allow the temperature to affect certain areas of the LLM. i.e not the system format
-
I added in context append for each user input and the answers from the chatbot but found after the 6 answers the chatbot would just repeat the previous question result and not answer the new question. Even when adding an answer to the most recent question to the system information it still leads to the same issue. Is there a way to add weighting to certain parts of the system information to maybe solve this issue?