Scott analyzes OpenAI's 'neuralese recurrence' technology that lets AI think in internal representations between processing steps, explaining the safety implications and arguing for clear taboos on recurrent architectures. Longer summary

Scott examines OpenAI's development of 'neuralese recurrence' in their Astra model, a technology that allows AI to think using internal representations rather than human-readable text between processing steps. He explains how transformers work with layers and chain-of-thought reasoning, why looping layers can create unmonitored thinking, and debates whether OpenAI's implementation crosses a dangerous threshold. The post concludes by discussing the need for clear taboos around recurrent AI architectures, drawing on Linchuan Zhang's argument that categorical boundaries work better than fuzzy thresholds for maintaining safety norms. Shorter summary

Scott argues that even if AGI requires a new paradigm beyond LLMs, we shouldn't expect significant delays, since Lindy's Law suggests major paradigm shifts could occur within 3-5 years, and new paradigms typically emerge precisely when scaling hits limits. Longer summary

Scott addresses the objection that AGI is far off because LLMs need a 'new paradigm' to reach AGI. He traces the evolutionary tree of AI development from neural networks through transformers to modern LLMs, then applies Lindy's Law to show that even paradigm shifts as major as deep learning or transformers should be expected within 3-5 years at the 25th percentile. He argues this timeline is comparable to LLM-only predictions anyway. Scott also makes a subtler point: new paradigms historically emerge when old ones hit scaling limits, meaning they won't cause delays but rather continue progress from where scaling left off. He concludes that extrapolating from current LLM scaling remains the best forecasting method whether or not LLMs themselves reach AGI. Shorter summary