The current approach to AI alignment involves using less advanced models to guide more sophisticated ones, mirroring human supervision of superintelligent AI. OpenAI acknowledges the limitations of this analogy and is actively seeking broader input through a $10 million grant program for research on superintelligent alignment, alongside plans for an academic conference in 2025. This suggests a cautious yet proactive stance on addressing the complexities of AI development and control.
