Uploaded May 2025 | Updated September 2026, 2 weeks ago
The AI absolutely lost it 💀
A pretty humorous video today, looking at a research paper where researchers created a benchmark instructing AI models to try and run a simple vending machine business. Except most of the time they end up going insane or absolutely unhinged, even to the point of contacting the FBI or preparing "quantum nuclear legal intervention".
The Paper: arxiv.org/abs/2502.15840
Time Stamps:
0:00 - Intro
1:01 - The Benchmark Paper
1:41 - Rules Of The Simulation
2:04 - AI Melts Down And Emails FBI
5:28 - AI Threatens “Nuclear Assault”
7:41 - AI Falls Into Despair ...But Recovers
9:09 - The One Where It Forgets How To Restock
9:28 - AI Thought It Was Broke
9:51 - Similar Failure Reasons
10:38 - More Thoughts
The AI absolutely lost it 💀
A pretty humorous video today, looking at a research paper where researchers created a benchmark instructing AI models to try and run a simple vending machine business. Except most of the time they end up going insane or absolutely unhinged, even to the point of contacting the FBI or preparing "quantum nuclear legal intervention".
The Paper: arxiv.org/abs/2502.15840
Time Stamps:
0:00 - Intro
1:01 - The Benchmark Paper
1:41 - Rules Of The Simulation
2:04 - AI Melts Down And Emails FBI
5:28 - AI Threatens “Nuclear Assault”
7:41 - AI Falls Into Despair ...But Recovers
9:09 - The One Where It Forgets How To Restock
9:28 - AI Thought It Was Broke
9:51 - Similar Failure Reasons
10:38 - More Thoughts
