Sam Altman is building an empire with Openai. One with some lights and with many shadows

Sam Altman is a master of empathy. He listens to you as if you were the most interesting person in the world, learn what he needs about you and his speech fits what you want. And so convince you.

It is one of the first conclusions that Karen Hao arrives in her new book ‘Empire of ai’. In it we are narrated OpenAI origins and its evolution Thanks to hundreds of interviews with employees and former employees of the company, in addition to those made to professionals from other companies in the artificial intelligence industry.

Altman is loved or hated, there is no middle ground

We actually know the story – in Xataka We have been Speaking of Openai – but what Hao proposes to us is a visit to what happens behind the scenes, contributing many details that help us understand the past, present and perhaps the future of the company.

Many of those details focus on the figure of Sam Altman, which does not go especially well stopped. Brilliant as a seller of apparently impossible projects, Altman is sparse in words in his communications with other colleagues. Write emails with a single word, “Meet”to arrange appointments, and sometimes use a simple “?” Because who writes less seems to win the game. Of that Jeff Bezos knows a lot.

That, of course, when I wrote something, because according to Hao Altman leaves almost nothing written. Everything is verbal, something that allowed him to argue after people did not remember well what he had talked to him.

The opinions of those who talk about him in the book are significant. One of them commented that “it is so attentive. But partly uses it to find out how to influence you in different ways.” Others commented how Altman avoids expressing negative emotions and also confrontation. He dodged the word “no” in conversations with other people. “Others began to see him as someone diabolically capable of beating situations in his favor.”

Ilya Sutskever, one of the co -founders who came after her differences with him, left A disturbing statement: “I don’t think Sam is the right person to be the one who has your finger on the AGI button.”

OpenAi lives his own ‘Game of Thrones’

Paul Graham, his mentor in Yc Combinator, left two citations that leave a clear idea of ​​what Sam Altman is like. In the first commented that “you could throw in parachute to an island full of cannibals, return in five years and he would be the king.” In the second reinforced That vision of his protégé: “Sam is extremely good when it comes to becoming someone with power.”

It is something that Hao often mentions in the book and that makes it clear that Altman does very well one thing: win the battles for power. There are two clear examples, also known. The first, when managed to snatch Musk The direction of Openai at the beginning of that unique adventure. The second, when After his scandalous dismissal He returned more force than ever as the almighty CEO of OpenAi.

Those two moments in the history of this company are actually reflecting what happens in any empire: the view seen is usually impeccable, great, powerful. The hidden face is full of internal conflicts and wars, battles for power, and rivalries and differences of criteria that end badly.

In all these battles an Altman was imposed again and again that according to Hao used a singular tactic: he changed his speech according to the interlocutor. What he had told A was often what he had told B. The problem arose when A and B were talking about what Altman had told both of them.

That also happened with Openai’s original vision. Created as a laboratory for the development of a beneficial for the world, The approach would change soon. To share knowledge and details about its models, the company became a secretism bunker.

Seeking to be the AI ​​monopoly

Like Oppenheimer, Altman Believe That “technology occurs because it is possible”, and like others before him – including one of his mentors and friends, Peter Thiel – his goal (such as his competitors, of course) is clear according to Hao: What he wants is to create an AI monopoly.

We have seen that with the evolution of their models, increasingly powerful, and that They were there to earn money. That was the vision that has ended up winning. The other, to try to develop a safe and “aligned with the objectives of the human race” has been in the background.

In fact Hao reflects it well in the book. If Openai is leading the AI ​​career today it is not only for having been the first to launch a chatbot like Chatgpt, but for its apparently disproportionate climbing.

He has invested more than anyone from the beginning. To start, to capture talent. When the project began to create OpenAi Ilya Sutskever, I worked in Google Brain and was already considered a superlla of this segment.

The rest of the founding members were offered a salary of $ 175,000 and shares of YC Combinator or Spacex. But Sutskever was offered almost two million dollars annuallybut Google counteroffierted on a bid whose final figure is not known.

What is known is that Sutskever ended up abandoning Google to sign for Openai –And then leave it-. In 2016 of the 11 million that Openai spent, seven were for salaries. Initially the company “did not really know what I was doing,” explains Hao: few of the things they worked worked, and those who did it “seemed little original or something someone had already done.”

There were more ambitious bets. It is demonstrated by the famous demo of that kind of “GPT 2.5” that made Bill Gates in April 2019. Until then those who investigated the development of foundational models of AI did so training those models with a few dozen gpus. Darío Amodei – who ended up leaving Openai to co -confound Anthropic – launched that new version of GPT in 10,000 GPUS NVIDIA V100.

Probably at that time many that a very expensive extravagance, a waste – like so many others later. That decision was a testament of Openai’s total bet for its approach. One that allowed him to go ahead for years. That “gates demo” It was the one He ended up leaving astonished to the co -founder of Microsoft. And it was undoubtedly one of the reasons for Microsoft to end up investing Those first 1 billion dollars In the company. As emails revealed between Kevin Scott (CTO from Microsoft) to Satya Nadella and Bill Gates, the concern about the advantage this company carried made made them preferred have it as an ally Before as an enemy. As we know, the thing has ended twisting.

The dark face of the AI

But in ‘Empire of Ai’ there is also a strong criticism with Openai’s darker side. A tragic and terrible one that tells us how the company has taken advantage of cheap labor from countries like Kenia, where workers gained misery labeling toxic content to polish the operation of Chatgpt.

Users are surprised by how well and politely respond, but if the chatbot behaves like this, it is much thanks to people who now have post -traumatic stress disorder as the moderators had before of YouTube either Facebook.

Like others, Openai has also exploited workers in countries in South America such as Colombia, Chile – terrible what the author details there— or Uruguay. Not only that: Hao talks about how technological giants such as Google have tried to reach controversial agreements with some of those countries to build data centers that would further threaten the meager reserves of drinking water in the country.

It happened in Uruguay: Google bought 29 hectares there to build a data center that would consume 7.6 million liters of water a day for cooling tasks. That is equivalent to the daily consumption of 55,000 people, many of whom could not allow to buy bottled water.

There is much more than discovering this striking research work by Hao, no doubt. In his prologue he already points to one of his conclusions not only about Openai, but the rest of AI companies: “over the years, I have found only a metaphor that encapsulates the nature of what these powers of AI are: empires.”

In another reflection of Hao, speaking of Altman’s origins, he was even more raw:

“That would reveal to what extent the domain of that technology – which is already restructuring society and transforming our planet – is based on polarized values, the egos faced and the disorderly humanity of A little handful of fallible people“

And in the end it is always that.

Availability of ‘Empire of AI’

The book ‘Empire of ai’ Karen Hao is available in English through the Penguin Random House publishing house in different specialized online bookstores and stores. At the moment there are no details or departure date of a possible edition in Spanish.

More information | Penguin Random Books