Marcus splits from Coxon on timelines while agreeing on the absence of a plan
Marcus on AI (Gary Marcus)·medium signal
In a Sept 9 post pairing Tao's warning with Coxon's resignation, Marcus says Coxon exaggerates near-term capability and disputes his end-of-decade framing, while agreeing that neither Anthropic nor OpenAI has a credible alignment plan. His position is that catastrophic harm is plausible even if extinction is not, and that moving past LLM architectures may be required. Useful as the contrarian read on a story where most commentary this week amplified rather than examined the risk claim.