The Human Brain

Our brain is all about predictions. We don't have the computational power to take everything into consideration, the only thing that we can do is to make the best guess. We condense our memories and experiences into patterns, and then use those patterns to make the best prediction of what comes next.

This is how we make millions of decisions every day.

We can read very fast in a familiar language because our brain guesses most of the words instead of actually perceiving every word.

LLM

Large Language Model (LLM) is also all about predictions. It was first designed as to predict the next word, giving the current context. Of course, nowadays LLMs can do much more than just predicting the next word, but the core is still the same: given the context (the prompt) and output the best prediction of what should come next.

This explains why LLMs are excellent at writing. They are trained on billions of billions of text, and they should have an excellent model of predicting the next sentence, the next paragraph, and even the entire article. But they might not be good at doing tasks that require accuracy. For example, when solving math problems, their "mind" seems to be blurred by that giant pre-trained clutter, and forget to use just simple equations.

Maybe LLMs also need to meditate, to clean up their "mind", to prevent over-complicating simple tasks.

Need More Context

Going back to humans ourselves, what should we do to make better decisions, or predictions, on our lives?

The answer is simple.

To get accurate LLM models, we need bigger datasets. To get better results from LLM models, we need to craft our prompts to include more details and context.

Therefore, our brain also need bigger datasets to work on, which means we need to try new things, talk to more people, and make more mistakes. Also, when we make decisions, we need more context, gather more information, and Do Your Own Research.

Good luck with your next guesses!