OpenAI and Anthropic's RSI Strategy and the Uncertainty of AI Profitability

OpenAI and Anthropic have set RSI (Recursive Self-Improvement) as a core objective to maintain their advantage as frontier labs. The gap between current AI models has narrowed to 1–2 generations, and the rise of competitors such as Deepseek and GLM could lead to low-margin cloud service models or 'speciation' specialized in specific domains. RSI is a strategy that allows models to develop subsequent models themselves, potentially leaping 20 generations ahead in a short period, but it is classified as a 'true unknown unknown' due to its high uncertainty. The fact that white-collar employment has not been significantly negatively impacted despite a 1000-fold increase in spending raises questions about simple scalar expansion of intelligence, and it is highly likely that AI will not generalize broadly even if it learns benchmarks.