Memorandum
- From
- Caroline Mercer via TIME.com
- Date
- Filed
- Business·5 min to read
- Re
OpenAI CEO Sam Altman Acknowledges Missteps as Company Reboots Around Astra AI Models
ReOpenAI CEO Sam Altman Acknowledges Missteps as Company Reboots Around Astra AI Models
OpenAI CEO Sam Altman concedes the company fell behind rivals after a difficult year marked by leadership departures, lawsuits, and public distrust. The company is now refocusing on its upcoming Astra AI models, which Altman says could be the first to invent new things in a meaningful way.
OpenAI chief executive Sam Altman has acknowledged that the company lost its lead in the artificial intelligence race to rival Anthropic, conceding a series of missteps in product direction and research as the company pivots to a new family of AI models called Astra. In interviews at the company's new research facility in the Bay Area, Altman said OpenAI fell behind where it wanted to be on pretraining and product strategy over the past year.
Anthropic, founded by former OpenAI employees, surpassed OpenAI in reported annualized revenue and private-market value for the first time, according to the company. The rival firm is now expected to be the first of the two to go public, with a possible initial public offering as early as September, according to two people familiar with its plans. Anthropic's coding tool, Claude Code, has become a market-defining product that OpenAI executives now acknowledge caught them off guard.
OpenAI has faced a series of setbacks beyond the competitive pressure. Several top leaders departed, including Fidji Simo, the former Instacart CEO recruited last year as second in command, along with leaders on safety, ethics, and research teams. Chief revenue officer Denise Dresser left after just eight months, and former chief operating officer Brad Lightcap also departed. Meta CEO Mark Zuckerberg poached key OpenAI researchers with lucrative pay packages, and Google's Gemini products now reach more than one billion people per month.
The company is also defending at least a dozen product-liability suits in California plus federal cases in which plaintiffs allege ChatGPT reinforced delusions or suicidal thinking and, in several cases, contributed to users' deaths. OpenAI has expressed sympathies for victims in those cases and rolled out ChatGPT for Teens with stronger default protections. Apple has sued OpenAI alleging theft of trade secrets, a claim the company denies. OpenAI also battled co-founder Elon Musk in a lawsuit accusing the company and Altman of betraying its nonprofit mission; a federal judge dismissed Musk's claims in May, and Musk has said he will appeal.
Altman said the past year has been painful personally, noting that his home was targeted by attackers twice in two days in the spring, first with a Molotov cocktail and then by gunfire. He acknowledged that public sentiment has turned negative on AI and data centers. Protesters were waiting outside OpenAI's offices in early August when customers arrived for a preview of Astra, holding signs to stop the AI race and writing chalk messages on the sidewalk.
Despite the challenges, OpenAI remains valued at nearly one trillion dollars, and ChatGPT still has more than one billion active users. Altman said he personally stopped using ChatGPT for a month in favor of Codex, the company's coding tool, noting that ChatGPT is no longer the cornerstone of the company's future. Under co-founder and president Greg Brockman, who now oversees nearly all product and business operations, OpenAI has wound down projects including the video-generation app Sora, a partnership with Disney, and a stand-alone web browser called Atlas.
Altman conceded the company had spread itself too thin. The company expects to spend $50 billion on computing this year alone, a decision that was widely criticized a year ago but now looks prescient, according to Sachin Katti, who oversees OpenAI's compute efforts. Katti said the company is still short of compute and should have bought more.
In demonstrations for customers, researchers showed Astra agents dividing a research-level math problem into subproblems and coordinating their work, as well as navigating desktop software to create and edit work across applications. Altman said Astra would enable persistent agents that work for sustained periods and predicted its biggest impact would come from people using it to discover new knowledge. He told visitors he expects Astra to be the first model that actually invents new things in a way that matters, calling that a very AGI-like thing.
