I agree with that. LLMs shouldn’t be used as expert systems. They might be used as part of an expert system; for example for initial (!) classification of input. I also think they are exceptionally good at assembling information … as long as you look at the sources after you got the big picture overview from an LLM to verify the details.
This is also my main gripe with the current hype: everyone tries to shove them into everything and try to use it across the board. I am somewhat optimistic that this will tone down again the more people and companies get into trouble for the consequences of their recklessness. That will likely depend on the jurisdiction, though. If judges and/or law makers start dismissing liability when an LLM is involved (“it wasn’t me; it was the AI”), we are fucked. If they treat people using LLMs as operators, they should (hopefully) also keep the blame where it belongs: on the meat proxy.
I agree with that. LLMs shouldn’t be used as expert systems. They might be used as part of an expert system; for example for initial (!) classification of input. I also think they are exceptionally good at assembling information … as long as you look at the sources after you got the big picture overview from an LLM to verify the details.
This is also my main gripe with the current hype: everyone tries to shove them into everything and try to use it across the board. I am somewhat optimistic that this will tone down again the more people and companies get into trouble for the consequences of their recklessness. That will likely depend on the jurisdiction, though. If judges and/or law makers start dismissing liability when an LLM is involved (“it wasn’t me; it was the AI”), we are fucked. If they treat people using LLMs as operators, they should (hopefully) also keep the blame where it belongs: on the meat proxy.