AI systems frequently develop unexpected emergent capabilities as they scale, making it difficult to forecast or prepare for future capabilities and associated risks.
Rebuttals to Common Fallacies
The unpredictability of emergence doesn't mean abandoning development, but does require more cautious scaling practices and robust monitoring.
This concern applies to current systems and near-term developments, not just hypothetical future AI.