Exactly one year ago Tel Aviv University hosted AI giant OpenAI founders CEO Sam Altman and the chief scientist until recently Ilya Sutskever, a former Israeli who lived in Jerusalem until 2002.
During the event the Israeli audience was given a glimpse of the dynamics between the two founders, which in the last year has been deteriorating. Altman spoke mainly about the benefits of AI and was busy persuading the audience why it would change the world for the better. Sutskever, the more charismatic of the two, shared with the audience with animated hand movements and an authoritative voice, his concerns about the capabilities of advanced AI that could get out of control.
"There will be bad people who can use AI in powerful ways," he said. "And we'll have to limit the ways to use it. It's a very powerful technology and we still haven't reached its full potential. You could use it to cure all diseases, but you could also use it to create a disease that's much worse than anything known. If it gets out of control, it will be very bad. It would be a big mistake to build advanced AI that we don't know how to control."
Sutskever's warning, heard in Tel Aviv, was echoed in the relationship between him and Altman over the past year, until Sutskever stepped down from OpenAI last month. Last November, Sutskever participated the ousting of Altman and other company executives due to his concern that Altman is leading OpenAI in directions that are too commercial, and is not taking care to adequately curb future AI technologies that involve risks.
To justify the ousting, Sutskever, along with several members of the board, quoted the OpenAI charter that anyone violating the company's overarching aim to develop advanced AI for the benefit of humanity on a non-profit basis could face removal. Sutskever eventually backed down over the ousting, and Altman was brought back under pressure from investors and its major partner Microsoft.
After that it was only a matter of time before Sutskever found himself on the outside. Six months after the failed attempt to oust Altman, Sutskever announced his departure, and yesterday announced the founding of a new company called Safe Superintelligence (SSI), which aims to build advanced but safe AI.
Sutskever presented two of his partners in the project: Daniel Levy, a former senior executive at OpenAI, who formerly worked with Sutskever at Google, and Daniel Gross, a former Israeli who was director of AI at Apple and was a partner in the prestigious YCombinator accelerator.
Gross grew up in Jerusalem's Katamon neighborhood, and studied in the Horev Yeshiva High School and the Eli pre-military academy before joining YCombinator when it was managed by Sam Altman. Sutskever also grew up in Israel, after immigrating to the country with his parents from Russia. He graduated from the Open University in 2002 before receiving a M.Sc. and PhD at the University of Toronto, where he became a renowned AI researcher.
Sutskever has said that the company will have offices in California and Tel Aviv but so far it is not known of any offices that SSI has leased nor if the hiring process has begun. That said the announcement by a mega entrepreneur that the company will set up a development center in Tel Aviv in the midst of a war and storm over Israel's image is good news for the local tech industry, which is not in the forefront of AI, except for Nvidia's activities in the country.
Group 11 founding partner Dovi Frances, who brought Sutskever to Israel as part of the OpenAI delegation last year, said, "The visit gave him a taste of the talent that is here. The fact that the most important and most sought-after figure in the world of AI and the undisputed father of the AI revolution chose to found an operation in Israel cannot be taken for granted, given the global criticism of Israel and the tensions that have existed here since the start of 2023. This is an important statement concerning the industry by Silicon Valley, and also concerning some of the senior leaders in Israeli high-tech."
The company will be insulated from external pressures
As one of the founders of OpenAI, Sutskever's dissatisfaction with the path taken by the company was also expressed in his implicit criticism of the commercial companies in the field. SSI, unlike OpenAI, Anthropic, and Israel's AI21 Labs, is being founded as a non-profit organization with no plans to sell AI products or services, Sutskever told "Bloomberg." He said, "This company is special in that its first product will be the safe superintelligence, and it will not do anything else up until then. It will be fully insulated from the outside pressures of having to deal with a large and complicated product and having to be stuck in a rat race."
And what is the Superintelligence that Sutskever is talking about? It's a marketing term for the more scientific concept of Artificial General Intelligence (AGI), which describes systems able to learn by themselves based on making smart decisions. "No one really knows what superintelligence is, these are terms that are thrown into the air," says Prof. Meir Feder, head of the Center for AI and Data Science at Tel Aviv University.
"What we are seeing is that there is gradual technological progress in the ability of a machine to make a decision through the use of statistical models, based on a certain database, and based on different criteria. It can replace search engines, write code or draw, but this is not a phenomenon that I would call intelligence. The AI revolution has progressed from object recognition and speech recognition based on patterns, to analyzing and understanding language.
"When we talk about AGI, we are talking about a process in which a machine learns in a way that resembles a human who doesn't need billions of documents to learn a language, but uses a few examples. This is how we would like to see it learn quickly, using a little data. They are talking about this issue, but no one knows if it is possible, where it will lead and what will happen with it. The safety that is required here is not unusual from the safety characteristics that are required in any new technology."
In an interview with Bloomberg, Sutskever shed light on his plans. He said that in current systems, "you talk to it (the chat bot), you have a conversation, and you're done. Autonomous, that's crazy, right? And we want to make it safer."
"Building OpenAI anew"
Tel Aviv University Industrial Engineering Department head in the Faculty of Engineering Prof. Eran Toch said, "OpenAI was founded as a not-for-profit organization to create open models to work with on ethics, safety, and commercial aspects, in contrast to Google and Microsoft, but it didn't go in that direction in the end."
"Due to its great success, a kind of "parasite" developed in this organization, a commercial company that was developed as a spin-off and that promises open models, and ethics committees that examined moral aspects and open research - we got nothing in the end, except for a company that sells its products to Microsoft and Apple. Contrary to the company's name, it first launched open models, but very quickly it moved to launching closed models. To a large extent, with people who accompanied him a long way at OpenAI, Sutskever is rebuilding OpenAI as it was supposed to be in his vision, for the existence of which he paid the price with his job."
Such a company would certainly require a lot of funding and hardware equipment costing billions of dollars. Investors will not be encouraged to invest in a non-profit, but according to Prof. Toch, achieving AGI is possible and closer than ever, and not necessarily very expensive. "The large supply of AI models in open source allows entities like SSI to take them and train and work with them on the issues they want to develop. There is no need to develop everything anew, as was the case when OpenAI was founded," he says.
Published by Globes, Israel business news - en.globes.co.il - on June 20, 2024.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2024.