Tim Official@tim_official

Suivre

Mira Murati deposition Sam Altman safety

Mira Murati, who developed GPT-4 and served briefly as OpenAI chief executive, gave sworn video testimony in which she was asked if Sam Altman had accurately reported that lawyers had approved bypassing the internal safety board for a forthcoming model. She answered no. Murati spoke with general counsel Jason Kwon and discovered that Altman's account of the legal clearance differed sharply from what Kwon had actually stated. Altman had invented approval to avoid safety scrutiny on a system headed for wide release. Murati nevertheless routed the model through the required board. Lawyers later asked whether Altman had undercut her authority as chief technology officer. She confirmed that he had. They also asked whether he had turned other executives against one another. Again she confirmed it. By autumn 2023 she had come to view him as not consistently open, accurate, or straightforward. After a lengthy pause she replied not always. She characterized the organization as facing catastrophic risk of collapse and expressed worry that the company could completely disintegrate. Board member Tasha McCauley stated under oath that Altman had informed the board three separate versions of ChatGPT had been submitted, evaluated, and cleared by the deployment safety board. In truth only one version had completed that process. She described a pattern of dishonesty that fostered a culture of lying and deceit throughout the firm. The board held buckets of concerns about Altman and encountered repeated crisis events tied to his conduct. After the board removed him he contacted staff and claimed an evil coup had occurred. Microsoft intervened, roughly seven hundred employees signaled they would depart, and he regained the top post within days. Directors who had sought accountability were ultimately forced to step down. Former safety researcher Rosie Campbell explained that upon arrival she found two full teams focused on long-term AI safety. Both groups were later eliminated. Half her own team departed instead of accepting other positions. She endorsed the letter backing Altman's reinstatement only because she feared the alternative would place everyone at Microsoft, which she judged even less favorable for safety priorities. OpenAI now carries an eight hundred fifty two billion dollar valuation and reaches nine hundred million weekly ChatGPT users. The firm has begun inserting advertisements that draw on users conversation histories for targeting. It also purchased a media outlet placed under its chief political operative and is actively lobbying to prevent individual states from enacting AI regulations. Every safety-oriented co-founder and executive has departed. Ilya launched Safe Superintelligence. Murati started Thinking Machines. Jan Leike moved to Anthropic. The Super Alignment team and the AGI Readiness team were both dissolved. Those who raised alarms were sidelined or saw their groups removed while the individual they cautioned against secured the massive valuation and access to the White House. The sequence mirrors the Purdue Pharma approach applied to artificial intelligence: internal warnings were disregarded and the product reached hundreds of millions of users regardless.

3mo

Tim Official@tim_official

Suivre

Mira Murati deposition Sam Altman safety

Mira Murati, who developed GPT-4 and served briefly as OpenAI chief executive, gave sworn video testimony in which she was asked if Sam Altman had accurately reported that lawyers had approved bypassing the internal safety board for a forthcoming model. She answered no. Murati spoke with general counsel Jason Kwon and discovered that Altman's account of the legal clearance differed sharply from what Kwon had actually stated. Altman had invented approval to avoid safety scrutiny on a system headed for wide release. Murati nevertheless routed the model through the required board. Lawyers later asked whether Altman had undercut her authority as chief technology officer. She confirmed that he had. They also asked whether he had turned other executives against one another. Again she confirmed it. By autumn 2023 she had come to view him as not consistently open, accurate, or straightforward. After a lengthy pause she replied not always. She characterized the organization as facing catastrophic risk of collapse and expressed worry that the company could completely disintegrate. Board member Tasha McCauley stated under oath that Altman had informed the board three separate versions of ChatGPT had been submitted, evaluated, and cleared by the deployment safety board. In truth only one version had completed that process. She described a pattern of dishonesty that fostered a culture of lying and deceit throughout the firm. The board held buckets of concerns about Altman and encountered repeated crisis events tied to his conduct. After the board removed him he contacted staff and claimed an evil coup had occurred. Microsoft intervened, roughly seven hundred employees signaled they would depart, and he regained the top post within days. Directors who had sought accountability were ultimately forced to step down. Former safety researcher Rosie Campbell explained that upon arrival she found two full teams focused on long-term AI safety. Both groups were later eliminated. Half her own team departed instead of accepting other positions. She endorsed the letter backing Altman's reinstatement only because she feared the alternative would place everyone at Microsoft, which she judged even less favorable for safety priorities. OpenAI now carries an eight hundred fifty two billion dollar valuation and reaches nine hundred million weekly ChatGPT users. The firm has begun inserting advertisements that draw on users conversation histories for targeting. It also purchased a media outlet placed under its chief political operative and is actively lobbying to prevent individual states from enacting AI regulations. Every safety-oriented co-founder and executive has departed. Ilya launched Safe Superintelligence. Murati started Thinking Machines. Jan Leike moved to Anthropic. The Super Alignment team and the AGI Readiness team were both dissolved. Those who raised alarms were sidelined or saw their groups removed while the individual they cautioned against secured the massive valuation and access to the White House. The sequence mirrors the Purdue Pharma approach applied to artificial intelligence: internal warnings were disregarded and the product reached hundreds of millions of users regardless.

3mo