Hinton vs. LeCun is back: did recent reasoning models prove that LeCun was right all this time about auto-regressive LLMs?

The long-standing disagreement between AI godfathers Hinton and LeCun has resurfaced, with recent advancements in reasoning models sparking debate over whose views on auto-regressive LLMs have been vindicated. LeCun's argument for human-like reasoning involving search in continuous representation space, rather than token space, is gaining traction. Research into continuous-thought LMs, compressed non-linguistic tokens, and recurrent-depth models, including Pathway's BDH-CQ and ARC-AGI, supports this, raising questions about when an LLM augmented with search or latent computation ceases to be a standalone LLM.