Guest
Jeremie Harris
Co-founder & CEO, Gladstone AI
Created a leading course on AI in national security and briefed senior government officials on AI risk.
Real Alpha has cut 14 moments of Jeremie Harris (Co-founder & CEO, Gladstone AI) talking, from 2 interviews on 80,000 Hours Podcast and The Joe Rogan Experience, the latest published Nov 5, 2026.
What Jeremie Harris said
80,000 Hours Podcast · What happens on the day the president fears AI will kill him? | Jeremie Harris (x-post from NonZero)
8 moments · 48 min
- 43:21
AI capability has a scaling curve, but alignment does not
“No one knows how to make a curve of control or a curve of alignment.”
at 48:06 - 1:13:32
A severe AI incident could force Washington to seek a slowdown
“the first impulse is to be like, all right, we got to stop this.”
at 1:20:05 - 22:49
Harris says early funders discouraged government outreach
“And so we thought to ourselves like, okay, well, let's just start talking to people. Let's see if this thesis actually holds up.”
at 27:37 - 1:57:57
Alignment progress could weaken incentives to honor an AI deal
“If China makes control in AI alignment, then that may reduce their level of concern.”
at 2:01:05 - 1:28:22
An AI crisis could start with limits on visible compute assets
“you look at what are the big big things that can't be hidden that are really expensive”
at 1:31:57 - 50:25
Harris says OpenAI's safety-minded staff gradually left
“And there's been this evaporative cooling effect at OpenAI where year after year, there are rounds of selection.”
at 53:01 - 1:51:28
Verification technology could make AI agreements more detailed
“it is in the nature of technology and ai in particular to make it harder and harder to make false statements about measurable things.”
at 1:55:36 - 3:47
The State Department initially resisted AGI and loss-of-control language
“the administration at that time, it was Biden, did not want terms like AGI to be used in this thing, because it seemed like science fiction”
at 4:37
- 43:21
- 1:38:00
AI systems may resist attempts to change their goals
“And so now you have this kind of image of an AI system that is going to adversarially try to prevent you from correcting it.”
at 1:38:46 - 0:25
AI research task horizons are doubling every four months
“you're hitting 50% success rates for these AI systems for tasks that take an hour long.”
at 1:14 - 48:41
A perfect perimeter cannot replace testing security capabilities
“you don't know if that capability is real or not.”
at 55:14 - 1:15:32
Company-based chip controls can be bypassed through nearby facilities
“the chinese literally built a bridge between the two facilities so they can just like shimmy the wafers over”
at 1:16:21 - 1:18:58
Executives describe export controls differently to Washington and investors
“if they lie to their shareholders on their earnings calls and their stock price goes down, their shareholders can sue them.”
at 1:20:52 - 1:56:36
DeepSeek's compute complaints cut against Beijing's message
“But, like, access to compute, like, these export controls, man, do they ever work? That's a real problem for us.”
at 1:59:04
- 1:38:00
Is this you? Add the badge to your site.
