OpenAI co-founder discusses technical bottlenecks in AI development

In a recent interview, OpenAI co-founder John Schulman and other AI researchers discussed the limitations of the current Transformer and reinforcement learning paradigm. They noted that recursive self-improvement may not lead to an immediate intelligence explosion and highlighted data efficiency and continuous learning as critical areas. The experts projected a 10-fold productivity increase in specific knowledge work within two years, with AGI potentially emerging in three to five years.