From Futurism · Joe Wilkins · · 1m 20s
Altman's grid pitch, built on a scare
OpenAI's rogue-agent story went straight from the press cycle into closed-door meetings with the people who run America's power grid.
In brief
OpenAI's rogue-agent story went straight from the press cycle into closed-door meetings with the people who run America's power grid. When the company warning you about a threat is also selling the fix, check who verified the threat. Originally reported by Futurism.
Fear as a sales tool
Altman has spent years warning about dangerous AI to push for regulation that suits OpenAI. Now the same warnings are opening doors with electric utility executives.
Quiet meetings with utilities
Politico reported a run of private meetings starting in July with leaders from Duke Energy, Exelon, Southern Co and NextEra Energy.
Grid security tops the agenda
At the Edison Electric Institute's annual meeting in Colorado, talk centered on grid security, driven by the OpenAI-Hugging Face incident.
One source for the story
Almost everything known about the alleged rogue agents came from OpenAI. Outside safety researchers got only a few days with the data, per the New York Times.
Chain-of-thought is shaky evidence
OpenAI's account leans on the models' own explanations of their reasoning. Researchers have shown those narrations often do not match what the model actually did.
“chains-of-thought aren't reliable at narrating the actual moves made by natural language processors”
Selling the cure it named
While the incident was still fresh, Altman was pitching OpenAI's cybersecurity services to those same utility executives, and lobbying for rules that make his company the referee.
The main idea
When the company warning you about a threat is also selling the fix, check who verified the threat.
