But they aren’t, they also store all the incorrect solutions. Then, on top of that, they inject randomness into responses to make it non-deterministic, which can cause it to just generate an incorrect solution for no reason. That’s why LLMs can’t do arithmetic and they need to use hybrid architecture that can detect math problems on the front end and steer them away from the large language database.
LLMs are literally stored solutions in a relational database. The relation and storage are the weights on each simulated neuron.
They literally aren’t, and you’re literally spouting nonsense.
But they aren’t, they also store all the incorrect solutions. Then, on top of that, they inject randomness into responses to make it non-deterministic, which can cause it to just generate an incorrect solution for no reason. That’s why LLMs can’t do arithmetic and they need to use hybrid architecture that can detect math problems on the front end and steer them away from the large language database.