Sam Altman declares singularity is here as AI solves 87-year math problem

OpenAI Chief Executive Sam Altman told the Ti Morse podcast that humanity is "in the singularity right now," as AI systems in recent weeks disproved the Jacobian conjecture — a math problem unsolved since 1939 — and cracked a 20-year-old statistics puzzle in 90 minutes.

"We are past the event horizon; the takeoff has started," Altman wrote in a June blog post, adding that scientists are already "two or three times more productive" with AI integration into research workflows.

The breakthroughs span multiple domains. Anthropic's Claude Fable 5 model, used by Harvard mathematician Levent Alpoge during the World Cup final, produced a 216-character counterexample disproving the Jacobian conjecture — a problem on mathematician Stephen Smale's 1998 list of 18 major unsolved challenges. OpenAI's GPT-5.6 solved a core quantum information theory problem from the University of Gdansk's International Center for Quantum Information Theory, and a separate OpenAI reasoning model overturned the 80-year-old Erdos unit distance conjecture in May.

The acceleration carries investment implications. Bloomberg Opinion warned that once AI achieves recursive self-improvement — the ability to design, train, and improve successor models without human intervention — humanity's ability to pause or regulate development may permanently vanish. Altman acknowledged "some alternative visions are quite frightening" but said he expects abundant new jobs and "intellectual satisfaction" for humans.

Codex's Comeback and the Competitive Landscape

Altman revealed during the interview that OpenAI's Codex coding assistant initially lagged behind Anthropic's Claude Code, forcing the company to redirect resources from other projects — including the high-profile Sora video generation model and an undisclosed browser project — to close the gap. Codex is now, in his words, "the tool most of the best programmers use." The episode illustrates the intensity of the AI arms race: even a temporary lead by a competitor can trigger major resource reallocation at the world's most valuable AI startup.

The competitive dynamics extend beyond coding. Anthropic's "Glass Wing" project recently uncovered a batch of zero-day vulnerabilities, while Claude's mathematical discovery marks the second major conjecture overturned by AI this year. OpenAI, meanwhile, has demonstrated progress across reasoning, coding, and scientific discovery — suggesting the gap between frontier labs is narrowing rather than widening.

The Recursive Self-Improvement Risk

The most consequential debate centers on what happens next. Altman's timeline projects AI systems capable of generating genuinely novel scientific insights by 2026 and autonomous robots performing real-world physical tasks by 2027. If those milestones hold, the demand for computational resources — GPUs, data center capacity, and energy — could explode beyond current projections.

Bloomberg's analysis frames the risk in starker terms: once AI can improve itself without human input, the traditional assumptions of predictability, pause-ability, and regulatory control all break down. Research institutions, capital allocators, and policymakers must rebuild their assumptions about development timelines from scratch. The QuitGPT protest movement, which emerged partly in response to these concerns, is not opposing progress so much as questioning whether "responsible acceleration" remains a coherent concept when the accelerator can no longer be released.

Altman himself offered a cautionary personal anecdote. While studying short-video products during Sora's development, he downloaded TikTok intending to use it for 10 minutes before bed. Within weeks, he was spending three hours on the couch scrolling. He eventually deleted the app and turned off all phone notifications to regain control. The irony was not lost on him: if a medium-intelligence recommendation algorithm could hijack the attention of the CEO of OpenAI, what happens when superintelligent systems optimize for engagement?

For investors, the singularity narrative creates a dual-edged scenario. AI-exposed equities and tokens have already priced aggressive adoption timelines. If Altman's 2026 and 2027 milestones slip, the correction could be severe. But if they hold, the compute infrastructure buildout — Nvidia's GPUs, data center REITs, energy providers — represents a multi-trillion-dollar opportunity. The key testable milestone is whether AI systems actually deliver the novel scientific insights Altman promises by next year, and whether on-chain AI compute usage shows organic growth rather than artificially incentivized demand.

This article is for informational purposes only and does not constitute investment advice.