The 'Godfather of AI' Has a Hopeful Plan for Keeping Future AI Friendly

Abstraction: Geoffrey Hinton's views on LLM risks and analog computing as AI safety mitigation

Key points:

• Hinton left Google in 2023 to speak freely about AI dangers after witnessing LLM capabilities change his views
• Estimates 50% probability AI systems will be smarter than humans within 5-20 years
• Changed his mind on three points: LLMs genuinely understand language, share knowledge far more efficiently than brains, and have better learning algorithms
• Argues predicting the next word requires genuine understanding of meaning — dismisses "mere text prediction" critique
• Proposed mitigation: analog computing for AI, since analog hardware uniqueness prevents efficient weight-transfer and hive-mind coordination
• Deep learning breakthrough in Hinton's University of Toronto lab ~2013 launched the modern AI era at Google, Facebook, Microsoft, IBM

Connections: Geoffrey Hinton · Google · Openai · AI Safety · Large Language Models · Deep Learning

Source: https://www.wired.com/story/plaintext-geoffrey-hinton-godfather-of-ai-future-ai/