Ex-OpenAI engineer who raised legal concerns about the technology he helped build has died

Article content

Sushir Balaji, a former OpenAI engineer and whistleblower who helped train the AI ​​systems behind ChatGPT, has died and later said he believed the practices violated copyright law, according to his parents and officials in San Francisco. He was 26 years old.

Article content

Article content

Balaji worked at OpenAI for about four years before resigning in August. He’s well-respected by colleagues at the San Francisco company, with one founder this week describing him as one of OpenAI’s strongest contributors who was essential to the development of some of its products.

Article content

“We were devastated to learn of this extremely sad news and our hearts go out to Sucher’s loved ones during this difficult time,” a statement from OpenAI said.

Balaji was found dead in his San Francisco apartment on November 26, in what police said was an “apparent suicide.” “No evidence of a crime was found during the initial investigation.” The city’s Office of the Chief Medical Examiner confirmed that the manner of death was suicide.

His parents, Purnima Ramarao and Balaji Ramamurthy, said they were still searching for answers, describing their son as a “happy, intelligent and brave young man” who loved hiking and had recently returned from a trip with friends.

Balaji grew up in the San Francisco Bay Area, and first arrived at the startup AI research lab for a summer internship in 2018 while studying computer science at the University of California, Berkeley. He returned a few years later to work at OpenAI, where one of his first projects, called WebGPT, helped pave the way for ChatGPT.

“Susheer’s contributions to this project were essential, and it would not have been successful without him,” John Shulman, co-founder of OpenAI, said in a social media post commemorating Balaji. Schulman, who recruited Balaji to his team, said what made him an exceptional engineer and scientist was his attention to detail and his ability to spot subtle errors or logical errors.

Article content

Article content

“He had a knack for finding simple solutions and writing elegant code that worked,” Shulman wrote. “He thought through the details of things carefully and meticulously.”

Balaji later turned to curating massive data sets of online writings and other media used to train GPT-4, the fourth generation of OpenAI’s pioneering large language model and the basis for the company’s popular chatbot. It was this work that eventually led Balaji to question the technology he helped build, especially after newspapers, novelists and others began suing OpenAI and other AI companies for copyright infringement.

He first raised his concerns with The New York Times, which published a profile of Balaji in October.

He later told the Associated Press that he would “try to testify” in the strongest copyright infringement cases and considered the lawsuit filed by The New York Times last year to be “the most serious.” Lawyers for the Times described him in a Nov. 18 lawsuit as someone who may have “unique and relevant documents” supporting allegations of intentional infringement of OpenAI’s copyrights.

Article content

Lawyers also sought his records in a separate case brought by the book’s authors, including comedian Sarah Silverman, according to the court filing.

“It is not appropriate to train on people’s data and then compete with them in the market,” Balaji told the AP in late October. “I don’t think you should be able to do that. I don’t think you should be able to do that legally.”

He told the AP that he gradually became disillusioned with OpenAI, especially after internal turmoil that led its board to fire and then rehire CEO Sam Altman last year. Balaji said he was widely concerned about how its products were commercialized, including their tendency to spread false information known as hallucinations.

But among the “range of issues” he was concerned about, he said he was focusing on copyright as the one “about which it is possible to do something.”

He acknowledged that this view is unpopular within the AI ​​research community, which is accustomed to pulling data from the Internet, but said “they will have to change and it is a matter of time.”

Article content

He has not been impeached, and it is not clear to what extent his revelations will be accepted as evidence in any legal cases after his death. He also published a personal blog post containing his opinions on the topic.

Schulman, who resigned from OpenAI in August, said he and Balaji coincidentally left on the same day and celebrated with colleagues that night with dinner and drinks at a San Francisco bar. One of Balaji’s mentors, co-founder and chief scientist Ilya Sutskever, had left OpenAI several months earlier, which Balaji saw as another incentive to leave.

Schulman said that Balaji told him earlier this year about his plans to leave OpenAI and that Balaji did not believe that better-than-human artificial general intelligence, known as artificial general intelligence, “was around the corner, as the rest of the company seemed to think.” The young engineer expressed interest in earning a doctorate and exploring “some unusual ideas about how to build intelligence,” Shulman said.

Balaji’s family said a memorial is scheduled for later this month at the Indian Community Center in Milpitas, California, near his hometown of Cupertino.

_____-

Editor’s Note – This story includes discussion of suicide. If you or someone you know needs help, the US National Suicide and Crisis Lifeline is available by calling or texting 988.

_____–

The Associated Press and OpenAI have a licensing and technology agreement that allows OpenAI to access a portion of the AP’s text archives.

Article content