A former OpenAI researcher, Andrew Ho, has publicly advised his former colleagues that frontier AI labs may be overvalued. Ho expressed concern that the rapid pace of spending on model training, driven by competition from cheaper alternatives like Moonshot's Kimi, could outstrip revenue growth. He believes that the pursuit of recursive self-improvement (RSI) as a path to exponential gains is flawed, as current models lack the 'research taste' to propose or identify significant experiments. Ho has left OpenAI to start a company focused on creating datasets for complex, judgment-heavy tasks that AI models currently struggle with, such as long-horizon scientific reasoning and statistical analysis. AI
IMPACT Suggests potential overvaluation in frontier AI labs and highlights challenges in AI research progress beyond verifiable tasks.
RANK_REASON Former employee expresses critical opinion on industry valuations and future prospects.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
