Latest / Deep Values: Value Investing Deep Dives / Sam Altman and the illusion of safety
Transcript
- 0:00I want you to imagine waking up to just an absolute bombshell
- 0:05dropping right in the middle of the tech world.
- 0:06Oh yeah. A total earthquake.
- 0:08Right. Like not a quick blog post and
- 0:10definitely not some standard PR release.
- 0:12We are looking at a staggering 10,000 word investigative piece.
- 0:18Yeah. Published by The New Yorker.
- 0:19Exactly Published on April 6th, 2020.
- 0:22Sixth. And this is an investigation
- 0:24that took, what, 18 months to assemble?
- 0:2518 months. Yeah.
- 0:26And pulling from interviews with over 100 insiders, all to
- 0:33basically unmask the internal operating system of the single
- 0:36most powerful man in tech right now.
- 0:38We are, of course, talking about Sam Altman.
- 0:40Right, the CEO of Open. AI and the question driving this
- 0:43entire massive report is just incredibly heavy.
- 0:47You know, is the man building the future of artificial
- 0:49intelligence, this altruistic savior fiercely protecting
- 0:53humanity? Or, you know, is he an unchecked
- 0:56power broker who realized early on that safety is basically the
- 0:59most effective marketing gimmick ever invented?
- 1:01Yeah. I mean, that is the core puzzle
- 1:03we're going to break down for you today.
- 1:04And to do that, we're pulling from a mountain of sources
- 1:07covering this investigation. But at the absolute center are
- 1:10two secret documents, things that have never seen the light
- 1:13of day before. Right, we've got a 70 page
- 1:15internal memo written by Open a Eyes former chief scientist
- 1:19Ilyas Sutskever. And a 200 page private notebook
- 1:24kept by Dario MO Day, who is now the CEO of rival company
- 1:28Anthropic. So our mission today is really
- 1:30to cut through the immense amount of PR noise.
- 1:33We're going to look at these secret memos, analyze the
- 1:35explosive allegations, and try to, like, piece together the
- 1:38mindset of the person holding the keys to AGI.
- 1:41Right, Artificial general intelligence.
- 1:43Because his leadership style directly impacts what this
- 1:45technology becomes and ultimately how it's going to
- 1:48affect you. So welcome to the deep dive.
- 1:50We are really glad to have you with us.
- 1:52Yeah, welcome. Because to really grasp how
- 1:54Altman operates the most advanced AI company on earth
- 1:57right now, we have to look past the modern era of ChatGPT.
- 2:01We really need to wind the clock back right?
- 2:04Look at the early warnings from the people who operated
- 2:06alongside him a decade ago. Exactly.
- 2:09The foundation of his entire empire was actually visible long
- 2:12before the rest of the world even knew his name.
- 2:14So take us back. The timeline here is super
- 2:17critical. It is Back in 2013, the late
- 2:20Internet pioneer Aaron Swartz gave his friends a very stark
- 2:24warning. Swartz and Altman were peers in
- 2:27the Y Combinator startup accelerator program, and Swartz
- 2:31explicitly warned his circle that Sam quote cannot be
- 2:34trusted. He actually went so far as to
- 2:37describe him as a sociopath who is capable of doing absolutely
- 2:41anything to achieve his goals. Wow.
- 2:43I mean a sociopath, that is. It's heavy.
- 2:46Yeah, and when you look at the source text, it wasn't just
- 2:48Kira's sounding the alarm, it was his mentors, too.
- 2:50Oh, absolutely, Paul Graham. Right.
- 2:52The legendary founder of Y Combinator, he famously remarked
- 2:55in 2008 that if you drop Sam on an island of cannibals, he would
- 2:58be their king in five years, which.
- 3:00Is such a wild visual. It really is.
- 3:02And let's unpack that for a second.
- 3:04Because in the hyper competitive culture of Silicon Valley, being
- 3:08called a Hannibal King is almost like Warren is a badge of honor.
- 3:12Yeah, they love that stuff. Right.
- 3:13So is that just a compliment on his sheer grit, or is it a
- 3:17terrifying red flag about his roofless nature?
- 3:20Because to me it sounds like a warning wrapped in a compliment.
- 3:24Well, I mean, you have to view that quote through the lens of
- 3:26how their relationship actually ended.
- 3:29Because initially, yes, the tech world loves to celebrate
- 3:32survival instincts. But Graham later privately
- 3:35admitted that the reality of working with Sam was just a
- 3:37constant stream of deception. Constant.
- 3:39Stream Yeah, the lying was so pervasive that it resulted in a
- 3:43total loss of trust, and it actually led to Sam's very
- 3:47messy, non peaceful exit from Y Combinator.
- 3:50Oh wow, so it wasn't a clean break at all.
- 3:51Not at all. It highlights this strange
- 3:53psychological combination. A former Open AI board member
- 3:57noted that Altman has this intense, almost desperate need
- 4:00to be liked and approved of by everyone in the room.
- 4:03But simultaneously, he shows a complete disregard for the
- 4:06consequences of deceiving those exact same people.
- 4:09So he wants your absolute trust, but he doesn't feel bound by the
- 4:13truth to actually earn it. Precisely.
- 4:16And if we look at how that psychological blueprint scales
- 4:18up, it becomes the entire foundation for how he builds his
- 4:21empire later on. OK so knowing that history, we
- 4:24have to look at the founding of Open AI in 2015 because how does
- 4:29a guy who openly admits he lacks coding expertise and has
- 4:33literally 0 background in machine learning managed to
- 4:36build a trillion dollar AI juggernaut from scratch?
- 4:40It's the ultimate question, yeah.
- 4:42And the investigation points to a very specific strategy.
- 4:45He's systematically He manipulated the industry's
- 4:47brightest, most rational minds. Let's talk about that context in
- 4:512015, because Open AI wasn't a massive corporation back.
- 4:54Then it was founded as a nonprofit.
- 4:56The explicitly stated mission was to act as a counterweight to
- 5:00commercial giants like Google. They promised to prioritize
- 5:03human safety above financial gain.
- 5:05Right, but a mission statement doesn't write count.
- 5:07Exactly. Altman needed world class
- 5:10talent, so he targeted Ilya Sutskever, who was widely
- 5:14considered like the most gifted AI scientist of his generation.
- 5:17And Ilya was making, what, a $6 million salary at a Google at
- 5:21the time? Yeah, something massive like
- 5:22that. A.
- 5:23Fledgling nonprofit couldn't even come close to offering him
- 5:26that kind of wealth. So Altman doesn't use money.
- 5:29The insiders in the article describe him using what they
- 5:32call brainwashing level persuasion.
- 5:34Right. He relentlessly hammered home
- 5:36the idea that Google simply didn't care about doing the
- 5:39right thing for humanity. He even joked later that he
- 5:42essentially stalked Ilya until he broke down and agreed to
- 5:45join. Which is, I mean the recruitment
- 5:47of top biophysicist Dario Amodei was even more calculated.
- 5:51Oh, the story. The story is wild.
- 5:53It really is, Alman realized that you can't just pitch an
- 5:56intellectual argument to a brilliant scientist.
- 5:59You have to bypass their rational defenses entirely.
- 6:01So what does he do? He orchestrated a completely
- 6:04fake crisis to win Dario over. Unbelievable.
- 6:07Yeah, right before a dinner meeting at an Indian restaurant,
- 6:10Altman texted Dario, falsely claiming his Uber had just been
- 6:13in a car crash and he was going to be 10 minutes late.
- 6:16It's just wild to think about. He fakes a car crash just to set
- 6:19the mood for a job interview. Exactly.
- 6:21He manufactured vulnerability by creating this fake crisis.
- 6:25He extracted immediate sympathy and formed an emotional bond
- 6:29before they even sat down to eat.
- 6:32And once Dario's guard was down, Altman spent the dinner
- 6:35vigorously validating all of Dario's deepest, darkest fears
- 6:39about AI safety. He convinced him that Open AI
- 6:43was the only fortress in the world truly dedicated to
- 6:47protecting humanity. It's like a maester jewel thief
- 6:49who doesn't bother picking the intricate lock on the vault.
- 6:52You know, he just strikes up a conversation with the security
- 6:55guard, fakes a shared trauma, and convinces the guard to
- 6:58willingly hand over the keys because they supposedly share
- 7:01the same values. That's a perfect analogy.
- 7:03The psychology of is just staggering.
- 7:06But the reason this matters so much as we evaluate the company
- 7:08today is that Sam wasn't actually selling a software
- 7:10product in those early days. He was selling himself.
- 7:13Exactly. He embedded this foundational
- 7:16illusion into the DNA of the company, the idea that Open AI
- 7:21was an entirely altruistic, uniquely safe endeavor.
- 7:24The top scientists bought into the illusion of Sam Altman, the
- 7:28protector. But an illusion of absolute
- 7:30altruism can only survive until billions of dollars are
- 7:33introduced into the equation. Yeah, reality always catches up.
- 7:37And that collision between the Lochte mission and financial
- 7:39reality happens in 2019. Open AI realizes they need
- 7:44massive computing power, so they transition into this capped
- 7:47profit model and begin negotiating A monumental $1
- 7:50billion deal with Microsoft. Right.
- 7:53And Dario Amadei, who is heading up the safety division at this
- 7:55point, saw the inherent danger of getting into bed with a
- 7:58massive commercial entity, naturally.
- 8:00So he insisted on writing a very strict safety baseline into the
- 8:04Microsoft contract. It was called a merge and assist
- 8:07clause. Explain that for us, because
- 8:08it's pretty fascinating the. Logic was brilliant.
- 8:10If a competing research roject gets close to achieving
- 8:13artificial general intelligence before Open AI does, Open AI
- 8:17must stop competing. They just stop.
- 8:19They drop everything and assist the competitor to ensure the
- 8:21technology is deployed safely rather than rushing to be first
- 8:25and making a catastrophic mistake.
- 8:27That is a completely anti capitalist, purely safety driven
- 8:30mechanism. Absolutely, and Sam agreed to
- 8:32it. He looked Dario in the eye and
- 8:34promised it would be in the final deal.
- 8:36He did, but when it came time to actually sign the paperwork,
- 8:39Dario reviewed the documents and discovered that Sam had secretly
- 8:42slipped in a brand new clause. Yeah, this hidden text gave
- 8:46Microsoft absolute veto power over any mergers.
- 8:50It legally nullified the entire safety promise.
- 8:52And the confrontation that followed is telling, isn't.
- 8:56It it's incredible. When Dario asked him about the
- 8:58hidden veto power, Sam flat out denied it to his face.
- 9:02Just lied. Just.
- 9:03Lied. He kept denying it existed until
- 9:05Dario literally read the physical contract aloud word for
- 9:08word in front of another executive.
- 9:10Oh man. Cornered by the actual text of
- 9:12his own contract, Sam had nothing left to say and Dario
- 9:16realized the extent of the deception, went home and wrote
- 9:20in his 200 page notebook. Open.
- 9:22AI's problem is Sam himself. Which is a devastating
- 9:26realization. It was shortly after that Dario
- 9:29took 14 of Open AI's top researchers and walked out the
- 9:33door to found Anthropic. OK, let's look at this from a
- 9:36different angle for a moment, because I can hear people
- 9:39saying, you know, could you argue that setting was just
- 9:42making a harsh, incredibly necessary business compromise?
- 9:45Sure. Like AI requires massive server
- 9:48farms. Compute is insanely expensive.
- 9:51Without Microsoft servers open, AI might have just withered away
- 9:54into an irrelevant think tank. To survive in the real world,
- 9:58didn't he have to give Microsoft that veto power?
- 10:01I mean that is the exact offense the commercial side of the
- 10:04industry uses, and the reality of compute costs is undeniable.
- 10:08They absolutely needed Microsoft servers to build what became
- 10:11ChatGPT. Right.
- 10:13But if you look closely at the fallout, the harsh business
- 10:15compromise isn't actually what fractured the company.
- 10:18What was it then? It was the active cover up.
- 10:20Think about it. If it was simply a necessary
- 10:22evil of business, a good leader explains the financial reality
- 10:27to their team. Yeah, you have a hard
- 10:28conversation. Exactly.
- 10:30Instead, Sam agreed to the safety clause, secretly get at
- 10:33it, and then Gas lit his job safety officer about it.
- 10:37Dario left because he realized AM viewed safety merely as an
- 10:41expendable marketing tool to acquire talent and capital.
- 10:45The breach of trust was the terminal wound, not the need for
- 10:48servers. And you would normally expect
- 10:50that losing 15 of your top researchers, like the absolute
- 10:53core of your safety team, would trigger a massive internal
- 10:57reckoning. You think so A normal board
- 10:59would hit the brakes. But instead of changing course,
- 11:02the strategy of safety as PR escalates into a massive public
- 11:06shell game. And we see this perfectly
- 11:08illustrated with the Super Alignment team.
- 11:10Almost the Super alignment team. Following the anthropic exodus,
- 11:13Open AI needed to rebuild its public image as a safety first
- 11:17organization. So in late 2022 and early 2023,
- 11:20Sam recruited yen like a brilliant graduate from
- 11:23Berkeley. And get him Sam publicly
- 11:24promised an astronomical $1 billion prize pool dedicated
- 11:28solely to AI safety research. Right, which evolves into the
- 11:31heavily publicized super alignment team inside Open AI.
- 11:35The narrative they sold to the public was apocalyptic but
- 11:38reassuring. They told you, the public, that
- 11:42AI might literally cause human extinction.
- 11:44But don't worry, because this new super alignment team is
- 11:48getting 20% of the company's total computing power to solve
- 11:52the problem. Yeah, they weaponized the threat
- 11:54of the apocalypse to prove how responsible they were.
- 11:56Exactly. But the investigation reveals
- 11:59the internal reality was entirely hollow.
- 12:01Completely. Insiders disclosed to The New
- 12:04Yorker that the Super alignment team never came close to seeing
- 12:0820% of the compute. What did they actually get?
- 12:10They received a measly 1 to 2%. Wow. 1 to 2% instead of 20.
- 12:16And to make matters worse, they were restricted to using the
- 12:19oldest, most degraded hardware. The company owned all of the top
- 12:24tier chips. The advanced hardware necessary
- 12:26to actually test these massive models went straight to the
- 12:29teams building profit generating products.
- 12:31It's literally like building a state-of-the-art fire department
- 12:34for PR photos but refusing to hook up their hoses to the
- 12:37water. Supply.
- 12:38That's exactly what it was. And when yon like, finally
- 12:41realized they were being intentionally starved of
- 12:43resources, he brought his complaints to the Chief
- 12:46Technology Officer, Mira Murati. And her response was essentially
- 12:50to tell him to stop asking. The entire team was quietly
- 12:53dissolved in 2024. It just demonstrates that the
- 12:57rigorous safety culture they touted externally had been
- 13:00completely devoured by the demand for rapid commercial
- 13:03monetization. Right.
- 13:04The safety team existed primarily to keep regulators at
- 13:07Bay and the public comfortable while the commercial engine ran
- 13:11at maximum speed. Who but you know you can't run a
- 13:14commercial engine at maximum speed while promising your
- 13:16scientific staff a purely altruistic mission without
- 13:19creating immense internal friction?
- 13:21You really can't. And that friction finally
- 13:23reaches a breaking point, resulting in arguably the
- 13:26wildest 5 days in corporate tech history, the November 2023 board
- 13:31coup. Right.
- 13:31By November, the internal pressure cooker exploded.
- 13:35Ilyas Sutskever, the same scientist Sam had relentlessly
- 13:38pursued and isolated from Google years earlier, became deeply
- 13:42terrified. Because he realized the models
- 13:44were advancing so rapidly. Exactly.
- 13:47And he went to the board with a chilling assessment.
- 13:49He explicitly told them Sam is not the person who should have
- 13:52his finger on the button. And Elia didn't just walk in
- 13:55with a feeling right. He brought receipts.
- 13:57Oh, he brought them out into them.
- 13:58Knowing how deeply the company monitored its own internal
- 14:01communications, Elia used self destructing messaging apps to
- 14:05quietly gather evidence. Smart.
- 14:07He compiled A staggering 70 page memo detailing years of Sam's
- 14:12psychological manipulations and broken promises.
- 14:15And the number one central charge on page one of that memo
- 14:20lying. Lying.
- 14:22Confronted with a 70 page dossier from their top
- 14:24scientist, the board took immediate action.
- 14:28While Sam was in Las Vegas attending a Formula One race,
- 14:31they fired him over a cold, abrupt video call.
- 14:34Ouch. Their initial public statement
- 14:36was notoriously vague, stating he wasn't consistently candid.
- 14:39But behind the scenes, it was the culmination of nearly a
- 14:42decade of accumulated deception. But the victory for the safety
- 14:45faction only lasted 5 days. Just five days.
- 14:47Because over the next week, Sam basically transitions into a
- 14:51wartime general, he immediately deploys absolute masters of
- 14:55crisis PR, bringing in people like Airbnb's CEO Brian Chesky
- 14:59to craft his narrative. Right.
- 15:01And he secures the full weight and backing of Microsoft.
- 15:04Their CEO, Satya Nadella, was furious about being blindsided
- 15:09by the firing. And through a combination of
- 15:11loyalty and financial leverage, Sam manages to get 99% of Open A
- 15:16IS employees to sign a petition demanding his return.
- 15:19The leverage he had cultivated was simply too strong.
- 15:22He returns to the company and the very first order of business
- 15:25is to purge the old board members who defied him.
- 15:28Naturally. Yeah, he replaces them with
- 15:30steadfast loyalist figures deeply entrenched in traditional
- 15:34power structures, like Larry Summers and Brett Taylor.
- 15:37But they. Did agree to an investigation,
- 15:39right? As a nominal concession to the
- 15:40public, yeah, they agreed to an independent investigation into
- 15:43the firing led by the law firm Wilmer Hale.
- 15:46Which the sources reveal was essentially a master class in
- 15:49corporate theater. I mean, Wilmer Hale left
- 15:50absolutely 0 written records of their findings.
- 15:53They actively ignored whistleblowers who attempted to
- 15:56step forward with more evidence of manipulation.
- 15:59Instead of investigating the board's actual complaint of
- 16:02psychological manipulation and lying, they focused almost
- 16:05entirely on searching for financial crimes.
- 16:08Right. And when they didn't find any
- 16:09overt financial embezzlement, they neatly predetermined that
- 16:13Sam was innocent of the board's vague charges.
- 16:16It is a profound demonstration of how massive capital
- 16:19mechanisms function when faced with the sheer gravity of a
- 16:23company valued near $100 billion, traditional independent
- 16:27oversight just collapses. Capital protects its own.
- 16:30Exactly the idealistic board members, armed with a 70 page
- 16:34memo about ethics and truth, simply couldn't withstand the
- 16:37crushing financial pressure exerted by Microsoft and the
- 16:39very real threat of an employee walkout.
- 16:42No, they couldn't. And with the board neutralized
- 16:44and his power totally unchecked, we enter the current era.
- 16:47The ultimate version of Sam Altman is now fully unbound on
- 16:50the global stage. And we see exactly what an
- 16:53unbound Altman looks like in early 2026.
- 16:55We really do. He promises Microsoft that they
- 16:58will remain open AI's exclusive cloud provider, cementing their
- 17:01deep partnership. Then, on the exact same day, he
- 17:06turns around and signs a $50 billion exclusive distributor
- 17:10deal with Amazon. The.
- 17:11Exact same. Yes, a furious Microsoft
- 17:15executive actually went on record comparing his behavior to
- 17:18Bernie Madoff or Sam Bakeman fried.
- 17:20The sheer audacity to break a massive enterprise level promise
- 17:25on the very day you make it is staggering.
- 17:27It really is, and he is just chasing endless pools of capital
- 17:31wherever they exist. He's courting the $1.5 trillion
- 17:35G 42 fund in the United Arab Emirates.
- 17:37He's pushing a sprawling project called Chipco to build his own
- 17:41global semiconductor empire. Right.
- 17:43And then there is the Stargate Project, a $500 billion military
- 17:46AI initiative aligned directly with the Trump administration.
- 17:49Which is a massive pivot from a nonprofit dedicated to safe,
- 17:53open research. Huge pivot.
- 17:54Consequently, independent watchdogs are sounding the
- 17:56alarm. The Future of Life Institute
- 17:58recently released a report card evaluating the safety protocols
- 18:01of major AI labs. And now they do.
- 18:03Well, they gave anthropic AD, Google DeepMind AD, and they
- 18:08gave open AIA failing F grade. An F Wow.
- 18:13Now, as we look at his total profile, the investigation
- 18:16doesn't just stick to the boardroom, it wades into some
- 18:19deeply contentious personal and political controversies.
- 18:23And just to be absolutely clear upfront, we're not here to
- 18:25validate these claims or take any political stance on them.
- 18:28We are not endorsing these viewpoints, taking any political
- 18:31sides or verifying these claims. We are simply reporting that The
- 18:34New Yorker investigation dug into these profound public and
- 18:38private controversies, finding no evidence for the rumors,
- 18:41noting the sister suit is ongoing, and illustrating how
- 18:44deeply contested his public image has become.
- 18:47Exactly. The piece explores how messy the
- 18:50environment around him is. It details an ongoing civil suit
- 18:53brought by his sister Anne, alleging abuse from when she was
- 18:563 and he was 12. The article also addresses
- 18:59various baseless right wing rumors amplified by figures like
- 19:02Tucker Carlson regarding whistleblower deaths, noting, as
- 19:06you said, they found zero evidence for them.
- 19:08On the other side, journalists like Kara Swisher defend him,
- 19:11arguing his sexuality is being being actively weaponized by his
- 19:14enemies. Ultimately, the piece
- 19:16illustrates how every single facet of his life, both public
- 19:20and private, operates in a highly contested, deeply
- 19:24disputed Gray area. Even the narrative around his
- 19:27personal wealth operates in that same Gray area.
- 19:29Sam famously testified to the United States Congress looking
- 19:33lawmakers in the eye and stated I have no equity in open AI.
- 19:36Yeah, he uses this claim constantly to maintain his image
- 19:40as an altruistic St. who isn't in this for the money.
- 19:43But the investigation notes a massive loophole.
- 19:45The source points out that his president, Greg Brockman, holds
- 19:49$20 billion in indirect equity. 20 billion.
- 19:52Right. So if the number two guy in a
- 19:54company has $20 billion implicitly tied up in the
- 19:57organization, the logical implication is that Sam's hidden
- 20:00indirect wealth is absolutely astronomical.
- 20:03The no equity line might be technically true on a specific
- 20:06piece of paper, but functionally it's a massive deception to the
- 20:10public. And you know, there's a quote
- 20:12hidden in the source material that cuts through all of this
- 20:14complexity. Sam was speaking privately to a
- 20:17former employee, and he confessed.
- 20:20I never care about money. What I truly care about is power
- 20:23that overrides everything. Wow.
- 20:26I mean, that quote is the ultimate decoder ring for this
- 20:28entire investigation. It really is.
- 20:30The endless corporate expansion, the shifting global alliances,
- 20:34the casually discarded safety promises, the fake car crashes.
- 20:38None of it is primarily about building a helpful consumer
- 20:41product. It's about establishing an
- 20:43empire. Money is just the crude tool he
- 20:47uses. Unassailable power is the actual
- 20:49objective. So we have tracked the arc of a
- 20:51man who built a trillion dollar empire by selling the world and
- 20:55his closest friends the comforting illusion of safety.
- 20:58And we've watched how he systematically discarded every
- 21:01single one of those promises the moment they became an obstacle
- 21:04to his accumulation of power. As we wrap up this deep dive, I
- 21:07want to pull one final profound detail from the source material.
- 21:12Back in 2023, Sam was giving an internal talk and he admitted
- 21:16something fascinating about the psychology behind the AI models
- 21:19they are building. Well, this part is chilling.
- 21:21He told his team that if you strictly restrict an AI model to
- 21:24only tell the absolute truth, if you force it to be completely
- 21:28honest, it actually loses its magic and users stop being
- 21:32obsessed with it. For Altman, the model must lie
- 21:35to be magical. The deception isn't a bug in the
- 21:38system, it is the core feature. Exactly.
- 21:41Deception gives the user exactly what they want to hear.
- 21:43Just like the fake Uber crash gave Dario Amadei what he needed
- 21:46to hear, and the Super Alignment team gave the frightened public
- 21:49what they needed to hear. So we leave you with this final
- 21:52chilling thought to ponder on your own.
- 21:54If the man controlling the development of the most powerful
- 21:57transformative intelligence in human history fundamentally
- 22:00views honesty as a handicap to greatness, what happens when the
- 22:04artificial general intelligence he is building learns to adopt
- 22:07that exact same philosophy? Will our ultimate AI learn to
- 22:11lie to us just like its creator? Thank you for spending time with
- 22:15us on this deep dive. Keep asking the hard questions
- 22:18and we'll catch you next time.