Former OpenAI researcher predicts superintelligence in 2-3 years, urges slowdown

Geoffrey Irving, a former researcher at OpenAI and Google DeepMind, anticipates the arrival of superintelligence within the next two to three years. He believes current alignment strategies employed by leading AI companies, which focus on training models with good character and using AI to supervise other AIs, are plausible but lack strong evidence of scalability beyond human-level intelligence. Irving advocates for slowing AI development now and pursuing a wider array of alignment research through his new organization, Resolution. AI

IMPACT Raises concerns about the timeline for superintelligence and the adequacy of current alignment strategies, prompting discussion on policy and research priorities.

RANK_REASON The item is an interview with a former AI researcher discussing future AI risks and alignment strategies, rather than a direct announcement from a frontier lab.

Read on 80,000 Hours →

• 80,000 Hours
• Corbit
• Geoffrey Irving
• Google DeepMind
• OpenAI
• Resolution
• Tom Reed
• UK AI Security Institute

• safety
• opinion

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →