Enhanced GPT Tool Creation

These suggestions I feel would go far into helping users create better GPTs.

1: problem; GPT creation is problematic for complex tasks that require strict adherence. The creation mode is not as effective to train an AI session as using train of thought in a standard chat session, by teaching chat GPT over a longer conversation you can train it to do some complex tasks.and with the new memory feature this will become powerful. However writing prompts in configure mode can be problematic due to the apparent conversation first approach. My proposal is to have the ability to export a GPT4 conversation as a custom GPT to enable the full training of the AI on a task and to then create a GPT with the training instead of the current models that don’t work as well.

2: Implement a toggle mode within GPT4 that enables features of GPT 3.5 Intruct. For building GPTs as tools, the GPT4 model is fantastic as it has better creative and interpretation capabilities but this also makes it prone to not following prompt instructions precisely such as if it is given a specific length of characters, words or tokens (tokens work better) for creating an initial sentence or paragraph it will ignore this initial limit request, but if told to reword to say 55 token length it will do so, if told not to use certain words in text creation it will flat out use those words, but if told to reword text and not use those words it will indeed take them out. It would be useful to have stricter instruction adherence while still using the GPT4 features.