Alex Heath
speaker
672 appearances
6 recordings
1 series
first heard Oct 2025
last heard 3 Sep
Alex Heath’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 6 in all, peaking in Sep 2026 with 1.
Appearances
And I also think, look, is convenient timing.
I think both can be true.
I think it does.
Um, if you I would encourage everyone listening or watching this to go watch OpenAI's Black Hat presentation where their researchers I cited in the interview with Sam, where the researchers walk through how the model actually escaped and hacked hugging face.
It's remarkable.
And it's the first incident of this happening.
Um and at the same time, I think you're right to point out.
Open AI didn't have the safeguards in place.
If this is a company, one of the two main companies we're trusting with the future of AI, upending everything, uh, how should we feel about the fact they didn't have the safeguards in place to keep this from happening?
They didn't foresee what would happen.
Um, and I talked to, you know, the chief scientists and head of research and a bunch of people, head of alignment, about this.
They had safeguards that they had developed that would have stopped this.
They didn't implement them for the training run that led to the hugging face breach because they didn't think the model was going to be that capable.
So there is a sense of what's happening inside these labs at the frontier, is the model capability is getting ahead of the human researcher's ability to foresee what's happening.
Which is you get into this concept of an AI recursive self-improvement, RSI, which Sam and I also talk about in the interview.
Particularly around the IPO, which was I thought interesting the way he framed it around the IPO.
Um, people in the Frontier labs really feel like they're on the edge of like a precipice that's like potentially amazing and also cataclysmic in the sense of AI that can build itself recursively, self-improve, not need a human in the loop, even the chief scientist of open AI.
Um
And that's already starting to happen.
People talk about RSI as a thing that hasn't happened yet.
Showing 41–60 of 672 · page 3 of 34
← Previous
Next →