OpenAI nears IPO, courts states

OpenAI spent May 20 and May 21 advancing three separate claims about its future: that one of its unreleased reasoning models solved a longstanding math problem, that it is helping shape AI rules in Democratic-led states, and that it is preparing for a public listing. The moves were reported across OpenAI’s own blog, POLITICO and Reuters, and together put the ChatGPT maker under a wider set of academic, political and financial tests. (politico.com) OpenAI has not publicly released the model behind the math claim, and it did not immediately respond to Reuters on the IPO report. (money.usnews.com) Chris Lehane, OpenAI’s chief global affairs officer, told POLITICO the company is pursuing what he called “reverse federalism” as Congress remains deadlocked on AI legislation. Reuters reported on May 20 that OpenAI is preparing to confidentially file for a U.S. initial public offering in the coming weeks, aiming to go public as early as September. ### What did OpenAI say it solved in mathematics? OpenAI said on May 20 that an unreleased reasoning model produced an original proof on the planar unit distance problem, a geometry question first posed by Paul Erdős in 1946. (indianexpress.com) The Indian Express, citing the company’s blog post, reported that the model found a new family of constructions that outperformed traditional grid-based arrangements. The planar unit distance problem asks how many pairs of points on a two-dimensional plane can be exactly one unit apart. (politico.com) The Indian Express said mathematicians had long worked under assumptions tied to square-grid-like arrangements, and reported that outside mathematicians were treating OpenAI’s new claim seriously rather than dismissing it outright. ### Why are mathematicians not brushing this off immediately? OpenAI’s latest math claim arrives after an earlier dispute over whether one of its models had truly solved unsolved Erdős problems. (indianexpress.com) The Indian Express reported that former OpenAI executive Kevin Weil deleted an earlier post after researchers including Yann LeCun and Google DeepMind chief executive Demis Hassabis said the model had rediscovered results already present in academic literature. That history makes outside validation central to the new claim. OpenAI said the new result involved an original proof, but the article says the model remains unreleased, leaving mathematicians and peer reviewers to determine whether the proof holds up. ### What is OpenAI doing in state capitals? Chris Lehane told POLITICO that OpenAI is urging large Democratic-led states to pass AI laws that “mirror each other” and create what he called a “de facto” national standard. (indianexpress.com) He said California and New York had already adopted rules that largely reflect the company’s preferred approach, and identified Illinois as the next target. POLITICO reported that the favored state laws focus on transparency and reporting requirements for developers of advanced AI models, while imposing relatively little new liability for catastrophic harms. (indianexpress.com) The outlet said the effort marks a shift from the industry’s earlier push to stop states from creating a patchwork of AI rules. ### Why does the state strategy matter now? State legislatures have introduced hundreds of AI bills and signed dozens into law while Congress has stalled, according to POLITICO. Lehane told the outlet OpenAI is trying to assemble a “critical mass” of states around one framework rather than wait for a federal law that may not come soon. (politico.com) Illinois lawmakers are now preparing legislation endorsed by OpenAI that would emulate laws passed in California and New York, POLITICO reported. That gives the company a near-term venue to shape rules covering advanced model developers while Washington debates broader national standards. (politico.com) ### How close is OpenAI to an IPO? Reuters reported on May 20 that OpenAI is preparing to confidentially file for a U.S. IPO in the coming weeks and is targeting a debut as early as September. Reuters said the company is working with Goldman Sachs and Morgan Stanley on a draft prospectus and had been valued at $852 billion in its latest funding round. (politico.com) A May 18 Reuters report said a U.S. jury ruled against Elon Musk in his lawsuit against OpenAI, removing what Reuters described as a major obstacle to an IPO. (politico.com) Reuters also reported that OpenAI had previously explored a listing that could value the company at up to $1 trillion and raise at least $60 billion at the low end of preliminary discussions. ### What will investors and regulators look at next? Reuters reported that ChatGPT serves more than 900 million weekly active users and that OpenAI has more than 50 million consumer subscribers. (money.usnews.com) The same report said the company has revised its product roadmap twice in recent months as it faces competition from Google and Anthropic. September is the next concrete date in the story. Before then, investors will be looking for OpenAI’s confidential filing, mathematicians will be testing the Erdős-related proof, and Illinois lawmakers are expected to take up legislation backed by the company’s policy team. (usnews.com) (money.usnews.com)