Reasoning models are just LLMs

The post argues that reasoning models are fundamentally just large language models (LLMs), challenging the notion that LLMs lack the capability for useful reasoning. It discusses how recent advancements in models like DeepSeek R1 demonstrate that effective reasoning can occur through autoregressive methods without explicit symbolic reasoning. The author criticizes those who previously dismissed LLMs and now try to redefine them to appear correct, asserting that the underlying technology remains the same and capable of complex task performance through unsupervised learning and reinforcement methods.