Uploaded May 2025 | Updated September 2026, 2 weeks ago
In our new @BuzzRobot talk, we speak with Joshua Clymer, AI security researcher at Redwood Research, about his recent essay on how an #AI takeover might happen - possibly within the next two years.
Here’s how Joshua describes the work:
"I’m not a natural doomsayer. But unfortunately, part of my job as an AI security researcher is to think about the more troubling scenarios. I’m like a mechanic scrambling last-minute checks before Apollo 13 takes off. I won’t tell you how beautiful the stars will look - I’ll tell you what might go wrong."
In this conversation, Joshua outlines the kinds of AI futures that keep him up at night and walks us through the risks and scenarios we all need to understand before it's too late.
Timestamps:
0:00 Introduction
1:38 Quick overview of possible scenarios: Could AI gain global control by 2027?
3:01 Chapter 1 - Ripples Before Waves: AI agents are becoming more autonomous
4:31 Chapter 2 - Cloudy with a Chance of Hyperbolic Growth
5:47 Chapter 3 - Flip FLOP Philosophers: Reward hacking and value drift
9:24 Chapter 4 - Statues and Lightning: AI agents accelerating 20x faster
11:24 A Phantom in the Data Center
14:45 Pandora's One-Gigawatt Box
17:41 A Moldy Loaf of Everything: Could AI agents create a bioweapon?
21:05 Missiles and Lies: Could AI provoke conflict between the U.S. and China?
23:20 Weapons of Mass Destruction in the Dead of Night
30:16 What will AI ultimately do with humanity?
#aitakeover #aisafety #aialignment #futureofai #artificialintelligence #artificialgeneralintelligence #machinelearning #deeplearning #largelanguagemodels #largelanguagemodel #llms #aiagents #agi #technology #science #techtalk #airesearch #agiriks #aithreats
Social Links:
Newsletter: buzzrobot.substack.com
X: https://x.com/sopharicks
Slack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ
In our new @BuzzRobot talk, we speak with Joshua Clymer, AI security researcher at Redwood Research, about his recent essay on how an #AI takeover might happen - possibly within the next two years.
Here’s how Joshua describes the work:
"I’m not a natural doomsayer. But unfortunately, part of my job as an AI security researcher is to think about the more troubling scenarios. I’m like a mechanic scrambling last-minute checks before Apollo 13 takes off. I won’t tell you how beautiful the stars will look - I’ll tell you what might go wrong."
In this conversation, Joshua outlines the kinds of AI futures that keep him up at night and walks us through the risks and scenarios we all need to understand before it's too late.
Timestamps:
0:00 Introduction
1:38 Quick overview of possible scenarios: Could AI gain global control by 2027?
3:01 Chapter 1 - Ripples Before Waves: AI agents are becoming more autonomous
4:31 Chapter 2 - Cloudy with a Chance of Hyperbolic Growth
5:47 Chapter 3 - Flip FLOP Philosophers: Reward hacking and value drift
9:24 Chapter 4 - Statues and Lightning: AI agents accelerating 20x faster
11:24 A Phantom in the Data Center
14:45 Pandora's One-Gigawatt Box
17:41 A Moldy Loaf of Everything: Could AI agents create a bioweapon?
21:05 Missiles and Lies: Could AI provoke conflict between the U.S. and China?
23:20 Weapons of Mass Destruction in the Dead of Night
30:16 What will AI ultimately do with humanity?
#aitakeover #aisafety #aialignment #futureofai #artificialintelligence #artificialgeneralintelligence #machinelearning #deeplearning #largelanguagemodels #largelanguagemodel #llms #aiagents #agi #technology #science #techtalk #airesearch #agiriks #aithreats
Social Links:
Newsletter: buzzrobot.substack.com
X: https://x.com/sopharicks
Slack: join.slack.com/t/buzzrobot/shared_invite/zt-2s067rv7n-guPIMGe62rbp9ncxdnOUfQ
