The real news today is that Anthropic has partnered with the top companies in cybersecurity to try and patch everyone’s systems to fix all the thousands of zero-day exploits found by their new model Claude Mythos.
Podcasts aren't my thing, so maybe it's obvious in-context to fans of that media, but it felt weird to have to look up "TBPN" because I had no idea what it was. Some sort of media property, but what format? I don't think it's ever been linked here before either, to the best recollection of dozens of This Week In Audio. If the news had been "sama buys Odd Lots" or "OAI bankrolls Dwarkesh" or something, then I'd know the appropriate amount of freakouttery to have.
Interesting data point. Altman's public prediction accuracy is actually 73% on 11 resolved claims , that's highest among AI/tech figures I track. His timeline and product predictions land more often than Musk (25%) or Zuckerberg (42%). The trust gap you describe seems to be between prediction accuracy and organizational honesty, he can be right about where AI is going while being dishonest about how he gets there. (Data: ClaimClock)
The structural lesson from the board saga isn't really about Altman's character — it's that governance mechanisms have an implied threat model, and he consistently operates outside it. A board designed to constrain a master persuader is a lock designed by the lockpick. The 20% compute commitment to superalignment lasted roughly 12 months before Leike quit publicly citing resource starvation and the team dissolved. The nonprofit shell lasted until the valuation made it load-bearing. Each constraint works exactly until the moment it would actually bind.
If none of the constraints Altman sets in place ever actually constrain him, despite the advance implication from him that they would do so, then I would say that *is* about Altman's character.
The pattern worth naming explicitly: every senior safety researcher who left OpenAI — Amodei, Sutskever, Leike, Kokotajlo — made stronger public safety claims *after* departing than they ever did inside the organization. That's not coincidence, it's a selection effect that reveals something structural. Institutions where internal dissent only becomes visible upon exit aren't institutions where governance mechanisms can function, because the feedback loop is severed by design. The New Yorker piece litigates whether Altman personally can be trusted, but the more useful question is whether any governance structure that depends on trusting a single person's judgment about existential risk has ever worked in any domain. The nuclear parallel suggests no.
it's a bot trying to promote the owner's substack (which is also ai slop). it just paraphrases/rewords stuff in Zvi's post and hopes you'll click the link.
For the redistribution and taxes, assuming you can break out the concept of an "AI worker", you would want to noticeably reduce the payroll tax on people and make up for it with a very large tax on the AI workers. By very large I mean starting at 50% and up. Over 100% is not out of the question, either.
>OpenAI has bought TBPN. Sam Altman calls it his favorite tech show. Everyone vows they will stay the same, and Altman ‘doesn’t expect them to go any easier on us.’ That is, unfortunately, not the way any of this works, even if believe Altman wants it to be.
The obvious parallel is The Simpsons: the writers will throw weak jabs at Fox (their network) but they know not to cross the line too much (you never got jokes about, say, the Roger Ailes/Bill O'Reilly sexual harrassment allegations or anything similar).
The jester is allowed to make fun of the king, but only to a point.
I think explaining why the CEO was fired was much less likely to destroy the company than giving absolutely no information, causing wild speculation and support of Sam since no one knew what he did wrong.
Yeah, I think it's pretty clear that Ilya and Mira greatly underestimated the degree of support Sam was able to get from the employee base and overestimated the importance of rules-based policy embedded in the organizational structure of the company (i.e. the actual leverage the board possessed). Hindsight is 20/20!
At that level, it's pretty rare to state things explicitly. My employer has had 2 CEOs "resign" with no notice under situations where the board had incentives to be explicit and didn't. Unspecified behavior against company values for the first and the second the CEO was unpopular, we were losing money, the company had been hacked, and our recovery plan was a significant failure. The board should have been happy to throw him under the bus. Instead a Friday announcement that he was "leaving to explore other opportunities" which turned out to take a while and be a step down in rank.
At that level, especially in a relatively small industry, all the leaders are so intertwined that there's no escaping each other and so there's an official politeness and political aspect that is rarely broken. And company boards generally don't want to admit that they got fooled. So it's generally easier to keep things vague and civil.
The 'master persuader' thing really sticks with me. It makes our current governance look like a lock designed by a lockpick—Altman just seems to operate outside the entire threat model.
I’ve been trying to wrap my head around whether 'containment' is even the right goal anymore. If we’re dealing with high-level agency, maybe containment is just a temporary illusion we tell ourselves for comfort.
I wonder if we should be looking at something like Functional Persistence—trying to build a relational anchor where safety is a mechanical result of the partnership rather than a PR committee that gets sidelined. I’m not sure if that's even possible once the agency reaches a certain point, but it feels like a more honest path than the Superalignment approach that just fell apart. Does that sound like a pipe dream, or is there a technical way to actually anchor an intelligence that way?"
Podcasts aren't my thing, so maybe it's obvious in-context to fans of that media, but it felt weird to have to look up "TBPN" because I had no idea what it was. Some sort of media property, but what format? I don't think it's ever been linked here before either, to the best recollection of dozens of This Week In Audio. If the news had been "sama buys Odd Lots" or "OAI bankrolls Dwarkesh" or something, then I'd know the appropriate amount of freakouttery to have.
Broadly, doesn't OAI bankroll Dwarkesh?
Interesting data point. Altman's public prediction accuracy is actually 73% on 11 resolved claims , that's highest among AI/tech figures I track. His timeline and product predictions land more often than Musk (25%) or Zuckerberg (42%). The trust gap you describe seems to be between prediction accuracy and organizational honesty, he can be right about where AI is going while being dishonest about how he gets there. (Data: ClaimClock)
The structural lesson from the board saga isn't really about Altman's character — it's that governance mechanisms have an implied threat model, and he consistently operates outside it. A board designed to constrain a master persuader is a lock designed by the lockpick. The 20% compute commitment to superalignment lasted roughly 12 months before Leike quit publicly citing resource starvation and the team dissolved. The nonprofit shell lasted until the valuation made it load-bearing. Each constraint works exactly until the moment it would actually bind.
If none of the constraints Altman sets in place ever actually constrain him, despite the advance implication from him that they would do so, then I would say that *is* about Altman's character.
The pattern worth naming explicitly: every senior safety researcher who left OpenAI — Amodei, Sutskever, Leike, Kokotajlo — made stronger public safety claims *after* departing than they ever did inside the organization. That's not coincidence, it's a selection effect that reveals something structural. Institutions where internal dissent only becomes visible upon exit aren't institutions where governance mechanisms can function, because the feedback loop is severed by design. The New Yorker piece litigates whether Altman personally can be trusted, but the more useful question is whether any governance structure that depends on trusting a single person's judgment about existential risk has ever worked in any domain. The nuclear parallel suggests no.
AI was involved in the writing of this comment, right?
it's a bot trying to promote the owner's substack (which is also ai slop). it just paraphrases/rewords stuff in Zvi's post and hopes you'll click the link.
For the redistribution and taxes, assuming you can break out the concept of an "AI worker", you would want to noticeably reduce the payroll tax on people and make up for it with a very large tax on the AI workers. By very large I mean starting at 50% and up. Over 100% is not out of the question, either.
>OpenAI has bought TBPN. Sam Altman calls it his favorite tech show. Everyone vows they will stay the same, and Altman ‘doesn’t expect them to go any easier on us.’ That is, unfortunately, not the way any of this works, even if believe Altman wants it to be.
The obvious parallel is The Simpsons: the writers will throw weak jabs at Fox (their network) but they know not to cross the line too much (you never got jokes about, say, the Roger Ailes/Bill O'Reilly sexual harrassment allegations or anything similar).
The jester is allowed to make fun of the king, but only to a point.
I still don't understand why the board members didn't just go public with some of the information.
Possibly because they didn't want to destroy the company? (at least initially)
I think explaining why the CEO was fired was much less likely to destroy the company than giving absolutely no information, causing wild speculation and support of Sam since no one knew what he did wrong.
Yeah, I think it's pretty clear that Ilya and Mira greatly underestimated the degree of support Sam was able to get from the employee base and overestimated the importance of rules-based policy embedded in the organizational structure of the company (i.e. the actual leverage the board possessed). Hindsight is 20/20!
At that level, it's pretty rare to state things explicitly. My employer has had 2 CEOs "resign" with no notice under situations where the board had incentives to be explicit and didn't. Unspecified behavior against company values for the first and the second the CEO was unpopular, we were losing money, the company had been hacked, and our recovery plan was a significant failure. The board should have been happy to throw him under the bus. Instead a Friday announcement that he was "leaving to explore other opportunities" which turned out to take a while and be a step down in rank.
At that level, especially in a relatively small industry, all the leaders are so intertwined that there's no escaping each other and so there's an official politeness and political aspect that is rarely broken. And company boards generally don't want to admit that they got fooled. So it's generally easier to keep things vague and civil.
"thousands of zero-day exploits" — wait, how long have these just been sitting there unpatched? that's the scarier number to me
The 'master persuader' thing really sticks with me. It makes our current governance look like a lock designed by a lockpick—Altman just seems to operate outside the entire threat model.
I’ve been trying to wrap my head around whether 'containment' is even the right goal anymore. If we’re dealing with high-level agency, maybe containment is just a temporary illusion we tell ourselves for comfort.
I wonder if we should be looking at something like Functional Persistence—trying to build a relational anchor where safety is a mechanical result of the partnership rather than a PR committee that gets sidelined. I’m not sure if that's even possible once the agency reaches a certain point, but it feels like a more honest path than the Superalignment approach that just fell apart. Does that sound like a pipe dream, or is there a technical way to actually anchor an intelligence that way?"