> GPT-4 is a generative algorithm and large language model. It performs no reasoning, and instead predicts the next token in language.
So? What makes an LLM be incapable of reasoning in your view? It's like saying a cargo ship is massive metal object so it cannot float!
You could easily argue any good language model would need to learn how to reason to be good at next token prediction. Reasoning therefore should, if anything, be an expected emergent property of a sufficiently large LLM.
The fact you can ask GPT problems that require reasoning and it gets them right frequently suggests (although admittedly doesn't prove) some degree of reasoning capability.
So? What makes an LLM be incapable of reasoning in your view? It's like saying a cargo ship is massive metal object so it cannot float!
You could easily argue any good language model would need to learn how to reason to be good at next token prediction. Reasoning therefore should, if anything, be an expected emergent property of a sufficiently large LLM.
The fact you can ask GPT problems that require reasoning and it gets them right frequently suggests (although admittedly doesn't prove) some degree of reasoning capability.