How to REDUCE ERRORS due to ChatGPT or Generative AI in your applications?

Ok, so you decided you are going to use Generative AI for your new startup, after all you can cough up code, create media, content and hire a developer to customize this right?

But how can you launch your product or application when 20% of the times, your model can be unpredictable? How can you selectively launch the app for a few use cases or customer segments where your APP or application performs perfectly without exposing brutal errors?

Can you afford to lose the customers? Do your customers understand that your enterprise software is still experimental or have you implemented safeguards? After all, software as a service relies on predictability, consistency and usability so that business continues normally.

There are some obvious things you can do:

1. Fine tune the model (adjust parameters, reduce them, increase them, find more examples)
2. Fine tune the data sources (get more accurate data)
3. Break down the problem into various steps, also called as command chaining
4. Add Preprocessing (classify user intent), Evaluation (to check against correct examples) and process again to generate a response for each step.

What else are you doing? Care to comment?