What I find interesting here is how far a well working QA application with LLMs (such as this one) is away from anything that can be generalized to other topics.
Thats probably where we are right now: I have seen quite a few purpose built and tuned AI systems for one specific use case or topic which work really well.
By contrast, I have yet to see any general AI bot that does this with arbitrary data for any reasonable definition of good.
I mean, take any of these Chat-with-data bots, load up a huge document and ask it for information that is spread on many pages (like make a list of prices for every product in a catalogue). Then see it fail.
Definitely feel this way too. Sometimes I think to myself that it'd be really great to have an LLM give me a well researched report on say like, recent trends on undisclosed marketing online, wished that we supported that on Lumona, but realized that we'll have to do it eventually, but pretty tough with the current infrastructure
Thats probably where we are right now: I have seen quite a few purpose built and tuned AI systems for one specific use case or topic which work really well. By contrast, I have yet to see any general AI bot that does this with arbitrary data for any reasonable definition of good.
I mean, take any of these Chat-with-data bots, load up a huge document and ask it for information that is spread on many pages (like make a list of prices for every product in a catalogue). Then see it fail.
Exciting times.