LLMs in itself don’t do much because they’re inconsistent, they lack clarity and data structure. But, on the other hand, classical (deterministic) software has always been rigid, stiff, inflexible. I think it’s like a human, the deterministic software is the skeleton (bones), the LLMs are the muscles.
Just like a muscle would not at all stand without a bone and would collapse immediately, so are LLMs extremely useless in themselves because they have no consistency. However, i think that if you combine classical software with LLMs, you can arrive at better results than in any other way. It’s like adding a muscle and a bone together to make a functioning system.
LLMs in itself don’t do much because they’re inconsistent, they lack clarity and data structure. But, on the other hand, classical (deterministic) software has always been rigid, stiff, inflexible. I think it’s like a human, the deterministic software is the skeleton (bones), the LLMs are the muscles.
Just like a muscle would not at all stand without a bone and would collapse immediately, so are LLMs extremely useless in themselves because they have no consistency. However, i think that if you combine classical software with LLMs, you can arrive at better results than in any other way. It’s like adding a muscle and a bone together to make a functioning system.
I’m not sure I understand. Inconsistent input to a rigid program gives inconsistent output. Consistent input to an LLM gives inconsistent result.
I’m a programmer. Can you tell me the specification of what you want the rigid classical software to do?
provide databases for structured knowledge, mostly.