Quote:
Originally Posted by jyl
Here is what I am starting to suspect.
Generative AI is going to be an groovy/easy to use front end/UI to send queries to the databases and applications and, yes, machine learning tools that already exist and are in use.
For rather trivial and imprecise stuff, maybe you have the AI come up with the answer. But if you’re trying to crunch financial data, examine a lesion, adjust an insurance claim, evaluate an employee, design a chip, I can’t understand how a LLM can be trusted to do anything other than interpret a natural language question and turn it into a structured query.
Anyone know about this stuff?
|
Right. Currently we're playing with the language models. Its word correlation, no semantics, no rationality. Is funny that they're terrible at things we normally use computers for. I personally believe these LM are literally how most human 'thought' works. We've nailed the mechanics of the peopleness of people.
Folks are now attaching these LM to calculators, theorum provers, etc. Not a 'structured query' but giving the LM a way to identify and offload the sort of work that can't be done with correlation. I think this capability is just getting started and will ramp up fast.