I often hear the argument that "it’s just mathematics," and therefore not intelligent. Yet, the universe in which we exist is also "just mathematics," and we clearly consider ourselves to be intelligent. It therefore seems obvious to me that intelligence can emerge from "just mathematics." The real question is how complex the underlying mathematics needs to be. In our brains, we have actual cells that form a complex neural network via synapses, dendrites, and so on. But is this level of complexity—which also manages the supply of nutrients to the cells—truly necessary for intelligence? Or, to put it another way: to what extent can this structure be simplified?
In my view, language models certainly exhibit a form of intelligence. Judging by the results, they grasp my input—or rather, my intent—and adapt their decisions based on dynamic input from various tools to address problems for which the solutions are not pre-programmed. I consider that a form of intelligence.
A completely different question, however, concerns consciousness, needs, and emotions. A language model doesn't feel a sense of satisfaction upon completing a task; it simply asks if there is anything else it can do or you just open a new prompt.
> it's actually just maths essentially mirroring the dataset
And your brain is just water and fat conducting electrical signals. Also, modern AI doesn't mirror the training dataset, that's not how LLMs work, it's only true for the initial pretraining phase. But that's besides the point.
What matters is what it can and can't do. Whether you label it as intelligence or not is pretty inconsequential long term.
I often hear the argument that "it’s just mathematics," and therefore not intelligent. Yet, the universe in which we exist is also "just mathematics," and we clearly consider ourselves to be intelligent. It therefore seems obvious to me that intelligence can emerge from "just mathematics." The real question is how complex the underlying mathematics needs to be. In our brains, we have actual cells that form a complex neural network via synapses, dendrites, and so on. But is this level of complexity—which also manages the supply of nutrients to the cells—truly necessary for intelligence? Or, to put it another way: to what extent can this structure be simplified?
In my view, language models certainly exhibit a form of intelligence. Judging by the results, they grasp my input—or rather, my intent—and adapt their decisions based on dynamic input from various tools to address problems for which the solutions are not pre-programmed. I consider that a form of intelligence.
A completely different question, however, concerns consciousness, needs, and emotions. A language model doesn't feel a sense of satisfaction upon completing a task; it simply asks if there is anything else it can do or you just open a new prompt.
> it's actually just maths essentially mirroring the dataset
And your brain is just water and fat conducting electrical signals. Also, modern AI doesn't mirror the training dataset, that's not how LLMs work, it's only true for the initial pretraining phase. But that's besides the point.
What matters is what it can and can't do. Whether you label it as intelligence or not is pretty inconsequential long term.