21–30 of 46 posts
Having internal users ask questions then have a ChatGPTesque system answer with injected data would be nice. Would be very tightly controlled (or it couldn't answer what it doesn't have).
Something between a rigid chatbot that's exists now and ChatGPT that just makes up plausible answers.
Neat, this is like a SaaS version of Langchain[1]. For those who haven't already seen it, Langchain is a framework that gives you the ability to do this as well as integrate out of the box solutions for other common problems like memory. [1] https://langchain.readthedocs.io/en/latest/
Langchain is awesome. Another great library is GPT Index. This field is apparently being called "prompt orchestration" (getting data from external sources, chaining going back and forth with an LLM, etc).
No wait, that doesn't have a great acronym.
Stage after that: GPT, or some other AI constantly adapts / curates the agents for me based on my notes, conversations, etc. (Although a privacy nightmare)
If you ask an existing LLM for a stock quote, it will hallucinate one, or give you a price it found in the snapshot of the web that it was trained on. But the intent here is to build a system that knows it should call an external service to get the right answer. It seems that there's a risk, though: the platform might use the correct service and get an accurate answer. Or it might make up a plausible answer. Who know…
What's your guidance concerning the risk of prompt injection with applications built on this kind of platform? If developers on your platform are building code that accepts prompt input from untrusted users do you have any measures in place to avoid them tricking the system into executing functions in an inappropriate or unintended way? One thing I'd find useful would be a limit on the number of functions that can be…