An ex-OpenAI employee has explained the reasons behind his resignation in February 2024; the company’s focus on developing ‘newer, shinier products’ has forced safety to take a backseat.
William Saunders, who joined OpenAI in 2021 to work on their AI alignment team, spoke with the Big Technology Podcast alongside his lawyer to discuss his concerns about his former company, whose stated goal is to build smarter-than-human artificial intelligence.
Saunders had hopes that the OpenAI trajectory was more in line with the Apollo program — rigorous risk assessment, careful thought, and stringent safety protocols, rather than with the Titanic, where a profit-driven race between companies to build the biggest and bestest ship meant that corners were cut and, as a result, thousands died.
But, as the months rolled on, OpenAI released ground-breaking large language models GPT-3 and GPT-4, and Saunders was dismayed by the recklessness of his bosses:
‘OpenAI claimed that their mission was to build safe and beneficial AGI. And I thought that this would mean that they would prioritize, you know, putting safety first. But over time, it started to really feel like the decisions being made by leadership were more like the White Star Line building the Titanic, prioritizing getting out newer, shinier products, than, you know, really feeling like NASA during the days of the Apollo program.’
On the 15th of February 2024, OpenAI previewed Sora, a cutting-edge text-to-video model. Saunders quit the same day.
Not long after his departure, fellow members of the ‘superalignment’ team Daniel Kokotajlo and Jan Leike followed suit, along with co-founder and chief scientist Ilya Sutskever.
Leike made similar comments to those of Saunders, saying in a Twitter thread that development of AI that’s smarter than humans is an ‘inherently dangerous endeavor’, and suggested that OpenAI is not cognizant of the immense risk they are forcing the rest of humanity to live with. He argued that the company must prioritise safety, and called on the remaining employees to ‘act with the gravitas appropriate for what you’re building’.
Kokotajlo revealed he had forfeited around 85% of his net worth in order to not sign an NDA that would forbid him from criticising his former employer.
The controversy around AI companies’ race towards AGI with little regard for the existential risk such technology would pose has led to increasing calls for stricter regulation. Max Tegmark, the author of the 2023 letter calling for a six-month pause on training models more powerful than GPT-4, argued in a recent interview with the Guardian that the incentives of tech companies preclude true dedication to safety:
‘If a CEO of a tobacco company wakes up one morning and feels what they’re doing is not right, what’s going to happen? They’re going to replace the CEO. So the only way you can get safety first is if the government puts in place safety standards for everybody.’
