Google DeepMind: The Podcast The Arrival of AGI with Shane Legg (co-founder of DeepMind)
307 snips
Dec 11, 2025 
Join Shane Legg, co-founder and Chief AGI Scientist at Google DeepMind, as he tackles the intricate world of artificial general intelligence. He discusses the current capabilities of AI and their limitations compared to human cognition. Shane explores various levels of AGI, from minimal to superintelligence, and emphasizes the need for ethical considerations and robust testing methods. Furthermore, he predicts significant societal shifts and job automation trends, while suggesting a 50% chance of minimal AGI by 2028. This conversation is a deep dive into the future of intelligence!
AI Snips
Chapters
Transcript
Episode notes
Minimal AGI As A Practical Bar
- Shane Legg defines minimal AGI as an agent that can do the kinds of cognitive things people typically do.
- He argues this is a practical bar because higher bars exclude ordinary human abilities.
AI Strengths Are Uneven
- Current models are uneven: superb at language and knowledge but weak at continual learning and visual reasoning.
- Shane expects these weaknesses to be addressed over several years with targeted techniques.
Don't Rely Only On Scale
- Improving reliability requires more than scale: targeted data, new algorithms, and architectural changes.
- Build episodic memory and retraining processes for continual learning rather than just more data.