Catching illicit distributed training operations during an AI pause
Last year, my colleagues on MIRI’s Technical Governance Team proposed an international agreement to halt risky development of superhuman artificial intelligence until it can be done safely. The agreement would require all clusters of AI chips with more…