Uploaded May 2025 | Updated September 2026, 3 weeks ago
How does Google DeepMind treat AGI safety and the potential threats it poses to humanity as a whole?
In this video, @BuzzRobot talks to Rohin Shah, who leads the AGI Safety & Alignment team at Google DeepMind. Rohin shares the insights from a very extensive paper on AI safety measures, called "An Approach to Technical AGI Safety and Security". We talk about what gets identified as areas of severe risks, mitigation approaches, safety training and capability suppression, oversight, scaling AI safety measures, and more.
Read the full paper "An Approach to Technical AGI Safety and Security" here: arxiv.org/abs/2504.01849
Taking a responsible path to AGI blog post: https://deepmind.google/discover/blog/taking-a-responsible-path-to-agi/
If you'd like to catch these talks live and participate in Q&As with the speakers, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ
Timestamps:
0:00 Introduction
0:52 Context and Scope: Severe harm to humanity
3:10 Areas of severe risks
07:04 Technical approaches to misuse
10:34 Technical approaches to misalignment
15:11 Alignment approaches: Enablers
16:37 Safer design patterns
Join BuzzRobot:
Newsletter: buzzrobot.substack.com
X: https://x.com/sopharicks
Slack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ
#ai #artificialintelligence #artificialgeneralintelligence #agi #machinelearning #deeplearning #neuralnetworks #llms #largelanguagemodels #deepmind #aisafety #googleai #technology #science #aitalk
How does Google DeepMind treat AGI safety and the potential threats it poses to humanity as a whole?
In this video, @BuzzRobot talks to Rohin Shah, who leads the AGI Safety & Alignment team at Google DeepMind. Rohin shares the insights from a very extensive paper on AI safety measures, called "An Approach to Technical AGI Safety and Security". We talk about what gets identified as areas of severe risks, mitigation approaches, safety training and capability suppression, oversight, scaling AI safety measures, and more.
Read the full paper "An Approach to Technical AGI Safety and Security" here: arxiv.org/abs/2504.01849
Taking a responsible path to AGI blog post: https://deepmind.google/discover/blog/taking-a-responsible-path-to-agi/
If you'd like to catch these talks live and participate in Q&As with the speakers, join the BuzzRobot community: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ
Timestamps:
0:00 Introduction
0:52 Context and Scope: Severe harm to humanity
3:10 Areas of severe risks
07:04 Technical approaches to misuse
10:34 Technical approaches to misalignment
15:11 Alignment approaches: Enablers
16:37 Safer design patterns
Join BuzzRobot:
Newsletter: buzzrobot.substack.com
X: https://x.com/sopharicks
Slack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ
#ai #artificialintelligence #artificialgeneralintelligence #agi #machinelearning #deeplearning #neuralnetworks #llms #largelanguagemodels #deepmind #aisafety #googleai #technology #science #aitalk
