I am thinking about an AI system which can: we can talk to it with regular language, teach it rules directions to think/analyze files. Then we can just give it files, it could create report based on the rules we taught it.
We could keep giving it new rules, it could keep learning rules, even analyze rules and tell some conflict rules.
Is this doable? How?
Thanks!
It is doable but takes a lot of work. To explain how in a single forum post is not possible.
Now see what state of the art LLM’s can do and how much development they needed and think about how many resources and development this might take!
Although your problem might be more simple it needs to be trained on a lot of similar data or scenarios.
Thanks @gent.spah !
I tried this with chatgpt, I uploaded a log file, then tell a rule like “performance log should happen once per each 5 minutes for each device”, then, ask chatgtp to check if something wrong in this log file. Chatgpt did it very well and pointed out two missing case in the log file. Then I tell another rule saying " there should be one ‘xxx started’ if one box get restarted", then ask "did you see box restarted? ", chatgpt still answer it right.
So, I think, this is doable in theory to build a local system to do similar thing. Could be very basic from the very beginning. I hope it can remember the rules/steps I told , help me to do analyze work.
Just don’t know where to start. Need some basic ideas.
I think some of the short courses offered here might give you knowledge on this process, for example LangChain courses can create efficient chatbots that remember quite a few rules!