a little philosophy on llm

does llm really think?

nowadays, it really gets interesting. llm appears to think like us. but it only seems...

in “AI: A Modern Approach” makes a distinction between “having intelligence” and “appearing to have intelligence”. on the philosophy side, there are experimental approaches like the Chinese Room Test, the Turing Test etc. we have a lot ideas to talk..

in my view, the issue goes beyond that. what we are facing right now is the battle between rationalism and empiricism. we, as human beings, are chasing theories that claim to govern the entire world with little knowledge. it’s quite easy to observe this in our experience. “people dressed in x, y, and z would have personalities like a, b, and c.. dude, these guys have been like acting like this for 500 years... Euclid discovered all 5 postulates of geometry... ”. we have many theories these lines that encompass both the past and future. of course, when we question these theories and go beyond them, we face intellectual revolutions. but, the number of people who do this is very small.

let’s go back to llm. it seems that the llm is taking very different approach to creating theories. after processig an enormous dataset, it can produce theses. it can even access very interesting knowledge connections that we may not be able to reach.

we, who act based on principles, prejudices, or theories, aim to act a lot in life with little information.and we get all these prejudices, theories, or principles with little knowledge. on the other hand, the llm tries to obtain principles through a vast amount of experience. also, it has a chance to reach more principles than we can.

in the current situation:

1. the battle between rationalism and empiricism gets deeper.
2. llm studies focus on processing information in a more optimized way. deepseek is a serious example of this.
3. we are changing the way we think in areas where mental abilities are showcased. since we cannot process large amounts of data, we work alongside llm.

in my opinion, llm is still at the level of primitiveness of a person holding a stone stick in their hand. it is doing important things by relying on the computing power behind them.

ok, what’s the power that allows a person to reach principles with little data? a writer can sense the end of the story from the opening sentence. a mathematician can feel discomfort about the object she is dealing with (for example, from Euclid’s fifth postulate). a musician can visualize notes in the her mind’s sky. what lies behind all these abilities? is it tied to the nature of the objects we engage with — numbers, notes, or words? is it related to properties of numbers, notes, or words? cannot these properties be grasped by artificial neural networks’ in hidden layers? or is it because we have already possess well-trained neurons? or is there something beyond the computation?

all we need to take a look inside of us.