Frustrating crippling of responses

Hi,

I have a GPT that does SQL queries and since the 5.1 update, it has been performing so much more poorly.

For example, it will execute a query that returns 100 results. It will show 5 on screen and ask if I want the data in a CSV or Excel file. I say yes and it creates a file with… 3 entries!

I know that the SQL bridge works because I see the query, I test it in a SQL app and the JSON contains all the information.

The instructions (RAG + instructions) are about 5K.

The resulting JSON response is about 16k tokens for my test. That doesn’t seem to be excessive with a 400K window.

Other than exiting the ChatGPT environment and doing it with an API and a frontend, what can I do to make this work?

L