Large Language Models are Wrong

Imagine you move into an apartment building and you get assigned the new person spot on the lowest level of a garage that had the building developer been a more compassionate human would have made the clearance height something greater than the inch or two you have when you pull up in your mini cooper. Large language models are like your annoying uncle going through a mid life crisis and attempting to park in your parking spot with his brand new electric hummer from GM. It’s bad. LLMs are not the answer. They don’t learn like intelligent beings do. They come with bias that can’t be unlearned and it has no way of truly knowing the end users interacting with it. LLMs believe context can be achieved with requests by storing bits of data about the person and sending it in as text for all requests made. That’s stupid not scalable and cheating. Oh yeah and the only way to scale LLMs are with disk and compute which means that only the big boys can compete. It’s time we have another look at incremental learning at our best hope to achieve “Artificial General Intelligence (AGI)”.

Apr 29, 2025 - 03:38

Imagine you move into an apartment building and you get assigned the new person spot on the lowest level of a garage that had the building developer been a more compassionate human would have made the clearance height something greater than the inch or two you have when you pull up in your mini cooper. Large language models are like your annoying uncle going through a mid life crisis and attempting to park in your parking spot with his brand new electric hummer from GM.

It’s bad.

LLMs are not the answer. They don’t learn like intelligent beings do. They come with bias that can’t be unlearned and it has no way of truly knowing the end users interacting with it. LLMs believe context can be achieved with requests by storing bits of data about the person and sending it in as text for all requests made. That’s stupid not scalable and cheating. Oh yeah and the only way to scale LLMs are with disk and compute which means that only the big boys can compete.

It’s time we have another look at incremental learning at our best hope to achieve “Artificial General Intelligence (AGI)”.