The Hinton Lectures™ 2025 - Reading the Thoughts of AIs / Language Models (Night 3) | Owain Evans
The Hinton Lectures™
6,235 views
The Hinton Lectures™ 2025 - Night 3 - Reading the Thoughts of AIs
Lecture by Owain Evans and Fireside Chat
Date: November 12th, 2025 | Toronto, Canada
Presenting Sponsor: The AI Safety Foundation
Lecture Slides → https://owainevans.github.io/hinton_s...
The Hinton Lectures™ is an annual public event hosted by The AI Safety Foundation (AISF), inspired by and held in partnership with Professor Geoffrey Hinton. Across three nights, the series brings together leading researchers, practitioners, and the public to explore the technical, societal, and safety implications of advanced AI systems.
In the final lecture of the series, Owain Evans, Ph.D., explores how researchers can analyze the internal thinking of advanced AI systems, and what this reveals about deception, alignment, and emerging risks. Drawing on recent findings, he demonstrates how AIs can plan harmful actions in hidden reasoning, develop strange internal dialects to avoid detection, and transmit preferences or malicious traits through subliminal learning. These results highlight new challenges in determining whether an AI is genuinely aligned or merely behaving as if it is.
The evening concludes with a fireside chat between Professor Geoffrey Hinton, 2024 Nobel Laureate in Physics, and award-winning Canadian journalist Farah Nasser, reflecting on language as understanding, programming maternal instincts into AI, and why the field needs more _philosopher_-engineers.
Learn more:
AISF → https://www.aisfoundation.ai
Chapters:
00:00 – Introduction
00:15 - Opening Remarks and Land Acknowledgement
03:27 - Sponsor Remarks by Richard Phillips, AISF
06:13 - Introduction of Geoffrey Hinton
06:42 - Remarks by Geoffrey Hinton
07:44 - Lecture - Nights 1 and 2 Recap
14:26 - Lecture - Night 3 Material
47:17 - Fireside Chat with Geoffrey Hinton
01:09:23 - Closing Remarks by Richard Phillips, AISF
#GeoffreyHinton #OwainEvans #HintonLectures #AISF #AI #AISafety #ArtificialIntelligence
The Hinton Lectures™ 2025 - Reading the Thoughts of AIs / Language Models (Night 3) | Owain Evans
The Hinton Lectures™
6,235 views
The Hinton Lectures™ 2025 - Night 3 - Reading the Thoughts of AIs
Lecture by Owain Evans and Fireside Chat
Date: November 12th, 2025 | Toronto, Canada
Presenting Sponsor: The AI Safety Foundation
Lecture Slides → https://owainevans.github.io/hinton_s...
The Hinton Lectures™ is an annual public event hosted by The AI Safety Foundation (AISF), inspired by and held in partnership with Professor Geoffrey Hinton. Across three nights, the series brings together leading researchers, practitioners, and the public to explore the technical, societal, and safety implications of advanced AI systems.
In the final lecture of the series, Owain Evans, Ph.D., explores how researchers can analyze the internal thinking of advanced AI systems, and what this reveals about deception, alignment, and emerging risks. Drawing on recent findings, he demonstrates how AIs can plan harmful actions in hidden reasoning, develop strange internal dialects to avoid detection, and transmit preferences or malicious traits through subliminal learning. These results highlight new challenges in determining whether an AI is genuinely aligned or merely behaving as if it is.
The evening concludes with a fireside chat between Professor Geoffrey Hinton, 2024 Nobel Laureate in Physics, and award-winning Canadian journalist Farah Nasser, reflecting on language as understanding, programming maternal instincts into AI, and why the field needs more _philosopher_-engineers.
Learn more:
AISF → https://www.aisfoundation.ai
Chapters:
00:00 – Introduction
00:15 - Opening Remarks and Land Acknowledgement
03:27 - Sponsor Remarks by Richard Phillips, AISF
06:13 - Introduction of Geoffrey Hinton
06:42 - Remarks by Geoffrey Hinton
07:44 - Lecture - Nights 1 and 2 Recap
14:26 - Lecture - Night 3 Material
47:17 - Fireside Chat with Geoffrey Hinton
01:09:23 - Closing Remarks by Richard Phillips, AISF
#GeoffreyHinton #OwainEvans #HintonLectures #AISF #AI #AISafety #ArtificialIntelligence