Open
Description
One of the most valuable tools on Gnothi is dream interpretation. Journal a dream, submit, prompt "interpret this dream". GPT does a phenomenal job, and it's lead to some powerful personal insights.
One way to improve this is a two step pipeline (and we may want to consider multi-step prompt flows in general).
- User: Journal > "Interpret this dream"
- Assistant: "It seems like someone in your life could be causing some anxiety around work..."
- (the new idea) cosine-similarity fetch all sentences the assistance's reply matches, take 5. This is called RAG, and we currently have paragraph embeddings so this is doable. Inject a new user prompt like "Interesting! Well here are some thoughts that may be relevant in my life: . How would you rephrase the interpretation with that in mind?"
- Assistant: "With that context, it seems like your coworker Tom might be... I recommend..."
That step 3 (like 1) would be automated; an interpolated prompt (only made to sound like it's coming from the user directly).