Real World AI Application Issue // Resolution
H/T to David George
Cognitive Corp has many autonomous AI agents powering web-based applications - one of these Apps is the 'ask the expert' (here is an example: https://lnkd.in/grqnWtdG)
David was attempting to use the agent and was getting an un-expected response (see image).
We dove in and this is what we found...
A bit of context - our AI Agents use a minimum of three AI bots and with one always being a fact check before publishing. OpenAI LLM is used by at least one of the other bots via API.
OpenAI LLM is in beta release which means it is in constant update and error solution correction mode. The OpenAI LLM is very overloaded. When working the API calls to Open AI it is common to get two types of errors 1) the model is overloaded 2) the model is not available.
No bid deal, we handle errors in coding, in this case our error handling to "retry" the LLM request was performed the maximum number of times (its good practice to limit API calls) and still no sufficient answer from OpenAI.
The fact-checking agent, being the diligent fellow that he is, returned the only answer he knew to be true.
Issue is resolved!
At least for now - let's remember AI-enabled tools are still in beta, many work very well.
It is only when we attempt to hide, cover-up, or otherwise gloss over the fact that all AI applications are in beta that we will lose trust.
Have a great day - and thank you again David!!!!! 🙂