Where basically it’s: accepting user input -> LLM to figure out which tool to call (user’s intent) -> JSON -> call API with strict security boundaries just like with a web app.
What’s the worse that could happen that couldn’t happen with a web app if I have bad security around the underlying API?
I’m sure that if you did successfully break my systems they could get it to say inappropriate things back to them for the lulz, but who cares?
Most of my use cases for using LLMs in production is call centers
https://news.ycombinator.com/item?id=47241412
Where basically it’s: accepting user input -> LLM to figure out which tool to call (user’s intent) -> JSON -> call API with strict security boundaries just like with a web app.
What’s the worse that could happen that couldn’t happen with a web app if I have bad security around the underlying API?
I’m sure that if you did successfully break my systems they could get it to say inappropriate things back to them for the lulz, but who cares?