> It helps to know that LLMs don’t “reason”. They predict ..
Semantics. Prediction is the training objective. The ability to reason can be, and very arguably is, an emergent property of that.
Semantics. Prediction is the training objective. The ability to reason can be, and very arguably is, an emergent property of that.
That being said, current generation LLMs do have issue, it's more productive to talk about those and their impact on real tasks (long term memory, continual learning, tokenization, context rot, reversal curse, etc)
Exactly. If LLMs don’t reason, then neither do humans.
Functionally, we can analyze LLM output and point to examples of reasoning, and examples that imply understanding.
The problem is there’s a lot of superstition around words like reasoning and understanding. People imagine some ineffable quality that only humans possess that distinguish what they do from what models can do, but they can’t define it - most likely, because it doesn’t exist.
I’d say we’ve figured out the fundamentals behind reasoning.
Sure. A photocopier fed with a maths paper.
What does this say about your reasoning?
But secondary to that, symbols != semantics. If I had a magical machine which could appear from the outside to be reasoning, but upon closer introspection was simply simulating reasoning, by definition it would not be reasoning. And yet that is the criterion you seem to think indicates reasoning ability. So it's not a sufficient criterion.
Claiming we've figured out the fundamentals behind reasoning is hopelessly naive.
1. Mimicking a process may not prove that the system is truly preforming the same process. In this case, a "reasoning" process.
2. A system that simulates a process cannot, by definition, be doing the process.
The first claim can be defended, but the second doesn't follow from the first.
Calling it a "simulation" is assuming the conclusion. What makes it a simulation rather than some implementation of reasoning on a different substrate? I.e. transistors rather than neurons.
Reasoning is an emergent property of complex systems. Prediction is not. Thus you cannot claim reasoning drives prediction.
> Where is the reasoning in linear regression?
The comment you are replying to was meant to expose you to your own reductionism.