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 Alexander shares his experiences and insights from the Asilomar Conference on Beneficial AI, covering various aspects of AI development, risks, and ethical considerations. Longer summary

Scott Alexander recounts his experience at the Asilomar Conference on Beneficial AI. The conference brought together diverse experts to discuss AI risks, from technological unemployment to superintelligence. Key points include: the normalization of AI safety research, economists' views on technological unemployment, proposed solutions like retraining workers, advances in AI goal alignment research, improvements in AlphaGo and its implications, issues with AI transparency, political considerations in AI development, and debates on ethical AI principles. Scott notes the star-studded attendance and the surreal experience of discussing crucial AI topics with leading experts in the field. Shorter summary