AI Analyst Urges Stronger Evidence On R&D Acceleration Trends
what is the rate of algorithmic progress in frontier AI systems for fixed data, and how is this changing over time? what is the rate of “data progress” (1 GB new data = X GB of 2025 data)? capabilities explosion might imply these trends go ~vertical.
Joel Becker@joel_bkrhow far are various AI R&D automation loops from closing? is the fraction of spend on human vs AI data (relevant to model ‘intelligence’) going to 0? what fraction of the important-weighted research ideas for most recent AI systems are AI-originated?
i could go on. a common theme is that *stronger evidence on AI R&D acceleration is possible but requires much more information.*
Joel Becker@joel_bkrwhat is the rate of algorithmic progress in frontier AI systems for fixed data, and how is this changing over time? what is the rate of “data progress” (1 GB new data = X GB of 2025 data)? capabilities explosion might imply these trends go ~vertical.