ENGLISH - The startup OpenAI wants to save the world, but without saying how. Silicon Valley is worried

On closer inspection, the chic gray building with white window frames doesn’t fit into the neighborhood. San Francisco's Mission District is a neighborhood of Hispanic workers – garages and taquerias line up here, and homeless people live in tents and trailers. On warm days, the smell of urine hangs in the air.

But in front of the large house at 2181 Folsom Street, the sidewalk is spotless, video cameras monitor activity outside, and on the first floor, blinds block the view inside. Nowhere is there a sign indicating that one of the world's leading artificial intelligence companies is based here.

OpenAI has shown us all, more than any other company, how far AI has come – and how this technology is likely to change all our lives.

The first thunderclap came last summer with the DALL·E 2 image generation software. Nestlé now also uses images created by DALL·E to promote its yogurts. OpenAI triggered a veritable earthquake when it released its chatbot ChatGPT to the public on Nov. 30, and public interest is so strong that ChatGPTs servers are regularly unavailable. Recently, the chatbot answered questions about the licensing procedure for doctors in the United States so well that it almost passed all three theoretical parts of the exam. Some financial firms are now having the program write a first draft of their quarterly reports.

But that is far from the culmination of what OpenAI has set out to do. Who exactly is the startup from San Francisco?

Founded by the Silicon Valley elite

© Provided by NZZ in English Elon Musk is one of the founders of OpenAI.  AP

Elon Musk, Peter Thiel, Sam Altman and other high-profile Silicon Valley founders were driven by the same idea in 2015: AI was advancing rapidly, but research was taking place behind closed doors in the big tech companies. Worse, early examples showed that this key technology was being misused for racism and bullying. Fundamental ethical questions also arose about the application of such tools.

In total, these individuals pledged $1 billion to create a nonprofit organization called OpenAI. Their goals were ambitious: They wanted to be the first to achieve «general artificial intelligence,» that is, AI that would be indistinguishable from human intelligence. Second, OpenAI would ensure that «all of humanity would benefit.» And third, the company would act so openly that the concept became part of its name.

In the event that competitors should achieve the goal of general artificial intelligence before OpenAI, the charter states that they will drop out of the race and support the competitors. It's exactly the kind of noble promise that attracts many Silicon Valley employees. In fact, the founders managed to attract just under a dozen of the best AI researchers.

Sam Altman radically rebuilds the company in 2019

© Provided by NZZ in English OpenAI's CEO Sam Altman drastically rebuilt what was once a purely nonprofit organization. Brendan Mcdermid / Reuters

But OpenAI soon realized that top AI research and the nonprofit business structure were in conflict with each other. The computer models ate up a lot of computing power and thus start-up capital.

In 2019, the startup underwent a radical transformation, with Sam Altman, co-founder of OpenAI and legendary as the founder of startup incubator YCombinator, stepping down from his leadership role there and moving from the board of directors to the role of OpenAI CEO. Altman, now 37, is notorious for his talents in marketing and sales. What he touches often turns to gold.

He also had clear goals with OpenAI. As a freshly arrived chief executive, he proclaimed, «We need to spend billions of dollars on cloud computing over the next few years, as well as on the best researchers, and to build AI supercomputers.» Presumably, Open AI would need more capital «than any nonprofit organization has ever raised in history,» he said.

Altman therefore gave OpenAI a new corporate structure, and a «‹capped› for profit» organization was created to complement the nonprofit arm. This company division was primarily intended to make money – however, the opportunity for returns was capped at 100 times the investment. So anyone who invested $1 million in OpenAI can earn returns of a maximum of $100 million.

It was a clear departure from the 2015 founding charter, which still said the company wanted to be free «from the need to generate financial returns. Because our research is free of financial obligations, we can better focus on making a positive impact on humanity.»

Many in the AI community expressed disappointment. How did OpenAI now differ from the big tech companies from which it wanted to differentiate itself? As the Wall Street Journal reported, many employees left the startup at that time.

A query to ChatGPT is seven times more expensive than one to Google

Those that stayed had a clear direction to follow: Build AI products that consumers and companies would be willing to pay for. Altman gave employees shares in OpenAI and announced that it would now license the products commercially to companies. He flew to Seattle and convinced Microsoft's CEO Satya Nadella to invest $1 billion with a demonstration of what was then the current version of ChatGPT. In addition, Microsoft agreed to provide the cloud infrastructure the company needed to train the startup's language models.

Publicly, the new CEO asserted that the new corporate structure hadn’t changed anything. What's more, employees today are reportedly compensated in part based on how closely they follow the charter's guidelines. Three hundred and seventy-five employees now work for OpenAI. «Our self-image is still that we work for the whole world,» Altman said in a recent interview. AI research is very capital-intensive, he said.

What exactly the computing power of ChatGPT costs is not publicly known. Altman talks about «pennies» per request. Scott Galloway, an economics professor at New York University, estimates that a question to ChatGPT costs about 2 cents – seven times as much as a Google search query. Tom Goldstein, professor of computer science at the University of Maryland, estimates that ChatGPT eats up $100,000 a day.

Microsoft invests another $10 billion

According to Reuters news agency, OpenAI is expected to generate $200 million in revenue in 2023 and as much as $1 billion next year. Microsoft announced last week that it would incorporate OpenAI technology into all Microsoft products in the future. CEO Satya Nadella is now shelling out another $10 billion, as confirmed Monday, Jan. 23, as part of a years-long partnership with the startup. That would give the startup a value of $29 billion. Investors believe OpenAI could become the next Google – that is, the next trillion-dollar company.

But Altman's conversion has also earned him criticism, including from co-founder Elon Musk. Musk had already left the company's board in 2018; on the one hand, because he saw conflicts of interest with Tesla's AI research, and on the other hand, because he didn’t agree with OpenAI's course, which was already becoming apparent at that time. OpenAI now lacked transparency, Musk criticized. He was «not confident» that the company was prioritizing the promised safety in AI development, he tweeted in 2020.

A few days earlier, the magazine MIT Technology Review had published a deeply reported article on Open AI that did not cast the company in a good light. Based on dozens of interviews with employees and former employees, the author recorded how profit and competitive thinking had erased the former idealism. Like a cult, the employees spent time together; Greg Brockman, one of the co-founders, even got married in the office, according to reports.

Even after recent breakthroughs with DALL·E and ChatGPT, there is criticism of OpenAI. «You pretend you're still not profit-driven – but the first tens of billions of dollars go to Microsoft, the employees and investors,» Scott Galloway, an economics professor at New York University, said recently at the DLD conference in Munich, referring to the capped-for-profit structure. «Until it actually becomes a nonprofit organization, OpenAI needs to be one of the top five most profitable companies in history. What unbelievable stupidity.»

But if you ask around in Silicon Valley, many also hide behind noncommittal statements. The AI community is tightly knit, one interviewee tells us in confidence. OpenAI now has the status of a saint with whom people don’t want to mess, this person says. To date, Altman and OpenAI have granted virtually no interviews to the media, and requests for interviews from this newspaper have also been unsuccessful.

For an organization that is dedicated to the well-being of humanity, this is very worrying, says Ramesh Srinivasan, a professor of information studies at the University of Los Angeles and author of the book «Beyond the Valley.» «What data are they collecting? With which algorithms do they establish correlations? And what exactly do they mean by promises to work ‹ethically› and ‹for the good of all mankind›? There are no answers.» That's typical for large AI companies, he said.

In addition, Time magazine recently revealed how low-wage workers in Kenya support OpenAI's AI models. Workers there have to manually train algorithms to recognize speech that is racist, sexist and violent for an hourly wage of less than $2. In the same way, Facebook and Google, for example, also «immunize» their AI models against hate speech. For this, the low-wage workers have to spend hours studying these unwanted images and texts, a job that is often traumatizing.

The goal: AI as our personal assistant

The discussion about transparency is relevant above all because we are likely to hear a lot more from the company. First, the technology underlying the chatbot should now be getting better and better – some experts believe that future versions of GPT will have the capacity to improve themselves on an ongoing basis, and will no longer require new version updates from the company.

Second, as we all know, the company's ultimate goal, according to its charter, is «artificial general intelligence» – thus, the point at which computers are able to think, act and feel in such a way that they are indistinguishable from humans. At present, that goal seems a long way off; even Altman says ChatGPT is in an «embryonic state.» But Musk also believes that «we are not far away from dangerously powerful AI.»

AI experts argue passionately about how far we are from creating thinking and feeling computers. German-born AI expert Richard Socher has a bet going with one of OpenAI's founders on whether the startup will have achieved general artificial intelligence by 2025. «I’ve bet against it and should win – but in the end, it's the winner,» he says. «Because OpenAI has advanced AI research to the point where it now has an insanely large impact on all research and applications.»

On the day he delivered ChatGPT to the public, Altman offered a glimpse into his view of the future. «Soon we'll have assistants talking to you, answering your questions, giving you advice.» And at some point, the AI will go off for you and not only do things, but even come back with new knowledge, he said.

Latest NZZ in English articles

The Neue Zürcher Zeitung (or NZZ for short) is one of the preeminent news sources in the German-speaking world, with a tradition of independent, high-quality journalism reaching back over 240 years.

With its curated selection of English-language articles on Swiss and international news, politics, business, technology and society, NZZ in English provides readers with a fresh and independent perspective. Learn more about the project here.

To stay up to date with our award-winning reporting, sign up for our Weekly Edition newsletter or follow @NZZinEnglish on Twitter.