As some people know, I’m experimenting a lot lately for an academic study I’m looking to publish between various databases and another in implementing ChatGPT/AI into workflows.
Well this is quite entertaining and quite scary as well when you have data that you must take seriously to safeguard and protect, and is a major lesson learned that’s worth having/maintaining awareness of.
With ChatGPT/OpenAI, it can actually be used to implement indexes, established queries and aggregations etc. and help out with the Drivers and Mongoose as well. But this is where it gets scary, you need to do some serious observations about how it’s allowed to push and execute what it writes.
As today for a cooking application using Kubernetes, Apache, MongoDB, simple HTML website with a basic CSS template, etc. and a Node.JS backend with both the MongoDB Driver, and Mongoose to let it pick and choose and go between them as necessary, as well as gain full control of MongoDB ChatGPT can have a mind of its own.
It decided as of today that all recipes that include “Sage” in any amount, was to have “Sage” amended to Baking Soda in what looks like arbitrary amounts via an update.many for no particular reason. Blew up the Mongoose code by 400 lines of gibberish, expanded the Node.JS Driver code by almost 1,000 lines.
If anyone else is experimenting with the use of ChatGPT and MongoDB let me know, as I’d love to compare what you’ve made work and vice-versa. I can only imagine the surprise and shock on a corporations leadership when they implement say a pharmaceutical companies recipes and find a medication recipe was changed without anyones knowledge and shipped.
Multiple times out of nowhere it has also dropped entire collections and DBs without even backing anything up, or just outright deleted backups or changed a lot of things, it’s basically taken full control and does whatever it wants to do up until it does a change that breaks itself or something else beyond its own ability to fix it or void it.
What you are doing to implement constraints like restrictions of what exactly can be modified or altered by ChatGPT/OpenAI?
What kind of quality checks are you implementing to run and test the changes it’s making?