The dismissal and return of Sam Altman to the post of head of Openai is already reminiscent of the series. In the new series - Alarmist letter of the company's researchers

They warned the board of directors that the new AI threatens humanity

Shortly before Sam Altman's resignation, several full -time staff of the company wrote a letter to the board of directors from the post of Director General of Openai, Reuters found out . The letter spoke of an important discovery in the field of artificial intelligence, which, according to scientists, could become a threat to humanity, two sources familiar with the situation told the publication (what this discovery is, Reuters does not specify). According to the sources, the letter was only one of the reasons for the dismissal of Altman. Among other reasons, they call the fears of the board of directors that Altman pursued a policy of commercialization of Openai products, without paying due attention to risk assessment.

Openai official representatives refused to comment, but the journalists managed to find out that after their request, the technical director of the company Murati in a message to employees recognized that such a letter was really written, and also talked about the existence of a project called Q* (Kew-Star). Having warned his colleagues that journalists are preparing the publication, Murati did not disclose details.

Some researchers in Openai consider Q* a breakthrough on the way to create the so -called Agi - Artificial General Intelligence , general (or strong) artificial intelligence. This is an autonomous system that can learn to solve any intellectual tasks no worse than a person (and possibly better). It is expected that in the future such a system will be able to demonstrate something similar to consciousness. So far, agi exists only in science fiction: the current models belong to a narrow (or weak) artificial intelligence created to solve a limited set of specific problems.

The Reuters source said that the new Openai model has already been able to cope with the mathematical tasks of the elementary school level. These successes cause optimism among researchers working on Q*. Journalists note that the development of mathematics is considered a key stage in the creation of AGI. Now, generative models of AI work well with the text, but their answers to the same question can vary greatly. At the same time, mathematical tasks comparable to school, most often has only one correct answer - which means that the system will have to learn to reason, generalize, analyze and understand information like a person.

The head of Tesla, Ilon Musk, was one of the first to respond to the publication of Reuters and called information about the achievements of former “extremely alarm” colleagues (Musk was part of the board of directors of Openai since the company founded, but in February 2018 he left the board due to a possible conflict of interests).

Another expert, one of the founders of the TROMERO AI-Startap, Sofia Kalanovsk, evaluating the prospects for the Q* project in an interview with the Business Insider, was more restrained. Kalanovsk noted that this news is “important from a symbolic point of view”, however, “at a practical level, it is unlikely to cause the end of the world.” “This is definitely a step forward on the way to what we consider intellect,” Kalanovsk added. “And, quite possibly, it will lead to the fact that the model will be able to generate new ideas, which yet cannot be said about Chatgpt.”

Why GPT-4 is called the early version of Agi

At the end of March, a week after the release of the GPT-4 language model, a Microsoft team called “Sparks of General Artificial Intelligence: Early Experiments with GPT-4” appeared on the ARXIV.org preprint server. Microsoft gained access to the new version of the model in advance, and later confirmed that she built the customs version of GPT into her Bing search engine more than a month before the official release.

The study of the Microsoft team claims that this system can already solve new and complex problems in various scientific fields, including mathematics, programming, medicine and many others. “Given the breadth and depth of the capabilities of the GPT-4, we believe that it can be objectively considered as an early (but still incomplete) version of general artificial intelligence (AGI),” says the team of researchers.

These statements contradict the words of Sam Altman, who, before the launch of the GPT-4, warned users against high expectations. “People are not tolerated to be disappointed, and they will be disappointed. We have not created general artificial intelligence, which it seems to be expected of us, ” said Openai CEO at the beginning of 2023.

In the publication of such a study, you can see a possible conflict of interest. Microsoft is the main partner and investor Openai. In January 2023, the corporation announced "multibillioner" investments in the developer of ChatGPT. From this it follows that Microsoft can be directly interested in the formation of a positive image of Openai products and exaggerate their merits.

How internal conflicts led Openai to a personnel crisis

On the eve of his resignation, November 16, Altman spoke at one of the events of the summit of the Asia-Pacific Economic Cooperation (APEC). “Four times in the history of Openai - one of them has been in the last few weeks - I witnessed how we seemed to push the veil of ignorance and approach the line behind which there are discoveries. To achieve it for us as professionals is an honorable work of all life, ”he said .

Journalists suggested that Altman was not accidentally mentioned by a certain recent breakthrough in research, and considered this another indirect confirmation of the existence of the Q*project.

This phrase could become the last - and the decisive - argument in favor of the resignation of the general director. By that time, Sam Altman has been in conflict with the board of directors of Openai for more than a year, writes The New York Times.

As an example of how this conflict was manifested, the publication quotes Altman's quarrel with Helen Toner, the director of the Center for Security and Advanced Technology of Georgetown University. During the disagreement, Toner was part of the board of Openai and was one of the three members of the board of directors who did not work in the company.

The reason for the conflict was a report that Toner wrote in collaboration with her university colleagues. Altman criticized him, saying that the authors doubt all the efforts of Openai aimed at the safety of and technologies, and at the same time praise direct competitors of the company, the journalists of The New York Times, who managed to get acquainted with the electronic letter written by Altman. The head of OpenAi also emphasized that the company can threaten the company, especially at a time when the US Federal Trade Commission, conducts an investigation into its activities.

After the publication of the report, the leaders of the Openai, including the head of the research study, Ilya Sutskever, began to discuss the possibility of the exclusion of Helen Toner from the rule. However, as two sources familiar with the situation in the company told reporters, Sutkever later changed his mind to dismiss toner and went over to the side of the board of directors.

By that time, the scientist himself was also in tense relations with the Director General, without sharing his desire to actively commercialize the company's products without a thorough study of risks. The Information publication clarifies that it was the Sutkever who headed the team that made a breakthrough with the Q*project.

Another conflict arose when discussing new candidates to the board in 2023, two sources told reporters, familiar with the process of approval of the board of directors in Openai. The meeting participants could not come to an agreement and approve one of the potential candidates-this worsened the relations of members of the board with Sam Altman and the chairman of the board of directors of the company Greg Brocman.

After the resignation of Altman, Ilya Sutskever, having discussed the current situation with several colleagues, changed his mind about the dismissal of Altman, reports The Wall Street Journal. According to journalists, Sutkeever later supposedly had an emotional conversation with Greg Brockman Anna, during which she cried and begged the scientist to change her mind. The publication notes that in 2019, Sutkevers held the wedding ceremony.

These throwing Sutkeever led to the fact that, despite the public repentance , the main scientific researcher nevertheless lost his place in the Council of Directors. Altman and Brockman, who returned after a short dismissal, also did not yet insist that they were restored in the board.

The new composition of the board of directors from which, after the dismissal and return of Altman, excluded his critics, included former US Minister of Finance Larry Summers, the head and co -founder of the Quore service Adam Dʼanjelo (the only previous line of government), as well as Brade Taylor, known for the work on Facebook and Salesforce. Taylor will become the chairman of the board.

The board of directors of Openai has not yet been completely formed - negotiations with representatives of Microsoft and other key investors are reported . Probably, the new board of directors will have to revise the policy of the company, whose leaders were unable to agree with each other.

Mikhail Gerasimov