As the dust settles on the saga surrounding the clash of directors and executives of AI startup OpenAI, more details about the events behind the scenes are emerging. OpenAI, the company behind the popular chatbot ChatGPT, surprised ** after a short blog post in November announced the ouster of its CEO, Sam Altman. According to the latest details shared by The New York Times in a report, while that alone is enough to generate a large number of stories, Altman, who initially agreed to resign, launched a comeback, and with broad support from employees, he negotiated his return to the company.
According to the report, Altman's stunning return to OpenAI came after he secured support for his role at the company from his home in San Francisco. The details come from interviews with dozens of people by The New York Times. While things peaked in November last year, OpenAI's troubles began brewing as early as two months ago in September due to a conflict between Altman and the former OpenAI board over filling the missing seats.
Altman's decision to create a for-profit OpenAI subsidiary and his decision to elevate OpenAI researchers to the same corporate level as the company's chief scientist, Ilya Sutskever, continue to create mistrust among stakeholders within the company. While OpenAI was founded in 2015 as a rival to Alphabet's subsidiary Google's Deepmind AI division, Altman took on the company's top position four years later in 2019.
Four years later, the board was wary of Altman using his professional network to reverse a possible removal, quietly voting at a ** meeting before informing him. Altman's initial reaction was to accept it, but he made a successful comeback inspired by others. While he and the board initially agreed to work together to pick new members, negotiations broke down.
However, backed by Microsoft's proposal, Altman believes he will force the board to become OpenAI's CEO again after a letter signed by hundreds of employees questioning the board's motives and threatening to resign if their leaders are not reinstated.
New details about Altman** and the return have appeared in other reports that suggest that the FTC is now interested in the nature of Microsoft's investment in OpenAI. The Redmond, Washington-based tech giant doesn't own a controlling stake in the company, and OpenAI's non-profit nature eliminates the responsibility for acquisition or investment reporting.
OpenAI's large language model (LLM)-based AI software is widely regarded as a global leader in the AI industry. LLM models, such as GPT, are referred to as Transformer models by researchers and industry members. This is because they convert a set of specified or unspecified parameters into output, and the "computing power" of an AI is often synonymous with the number of parameters it trains. Tesla, for example, uses machine learning Xi in its semi-autonomous software system called Autopilot, revealing that last year, its cars were powered by "a neural network with 1 billion parameters and 144 trillion operations per second."
While OpenAI has yet to publicly share details on the number of parameters powering its latest product, GPT-4, the report claims that GPT-4 has 120 layers out of 18 trillion parameters. If true, this figure places it at the top of the global AI food chain, making it more powerful than Google's Universal Language Model (GLAM). According to Google, the full version of GLAM has 12 trillion parameters across 32 layers. AI does not use all of these parameters at the same time when producing outputs or "inferences". Instead, it only activates 97 billion each time.