Using Full.CX’s unsupervised algorithms with GPT-3.5 to create complete complex product plans from the name to the acceptance criteria.<p>Choose 3.5 purely based on cost.<p>Significant use of input tokens. This is absolutely crucial to get reliable results, prompts can be up to 7k+ tokens.<p>GPT-4 produces superior results and will eventually switch over.<p>3.5 has issues with valid JSON using function calls but with the help of JSON repair libraries can overcome the issues.<p>Would love to see what ideas it produces and even better if someone actually makes it into a real product.