I implemented a ‘token intensive’ API application using the GPT-4 model and find it quite promising, although it still requires substantial testing. However, each time I review my API usage after a testing session, I feel discouraged and consider resigning from using the API altogether. Can anyone else relate to this experience?
In my testing experience, I first try with the GPT-3 model. If this succeed, I stay with the model or use GPT-4.
Most of my task can be done with GPT-3 even their are complicated. It will costs more time on prompt engineering, because GPT-3 needs better self-reflection mechanism to succeed reasoning some tasks, but it is worth the effort.
1 Like