I’ve tried 4 and I really can’t say the results are a qualitatively better than 3.5 for the tasks I’ve been trying (which have been trying to get it to generate documentation for my project).
In fact, I find 3.5 turbo the best overall model as a tool, because to quality of responses really depends on the quality of prompts, and the quality of prompts is improved by reacting to responses, which come more quickly in 3.5-turbo. So while ChatGPT-4 is still writing the first not-good response, ChatGPT-3.5-Turbo will be on the 2nd or 3rd and it will be much more cogent.
In fact, I find 3.5 turbo the best overall model as a tool, because to quality of responses really depends on the quality of prompts, and the quality of prompts is improved by reacting to responses, which come more quickly in 3.5-turbo. So while ChatGPT-4 is still writing the first not-good response, ChatGPT-3.5-Turbo will be on the 2nd or 3rd and it will be much more cogent.