Comment Re:This is why we use "agents" instead of "LLM's" (Score 4, Insightful) 114
The ChatGPT (or any AI product) people actually use is more than just an LLM. People get wound up over the inherent limitations of LLM's as if that's some brick wall preventing these things from ever becoming useful.
It's ludicrously simple to program a chatbot script to return the time instead of diving into the LLM if the user asks for the time. Same with mathematical operations - yes, LLM's sometimes get basic math wrong because that's not what they're designed for. But again, super simple to channel math requests to an engine that's built for that.
If they haven't fixed a particular shortcoming yet it's because they're seeking more complete solutions than picking off one tiny complaint at a time. But these are trivial to fix and smarmy posts like this will not age well.
The whole problem with this approach is a) we don’t have all the systems we need individually at a level of capability that meets the current level of hype and b) we don’t have a capable enough system management AI architecture, just using some weighted tokens is horribly deficient.