Singularity: An Anthropic Co-Founder Presents Evidence

During a speech at the IMF in Bangkok, a co-founder of Anthropic claims that the singularity has already begun. He bases this assertion on performance leaps in scientific assessments, concrete cases in mathematics, biology, and robotics, and calls for organizing the sharing of AI benefits while preparing for rapid changes.

From Score Leaps to Scientific Assessments and TheoPhysEval

AI systems have been tested on unprecedented theoretical physics problems. The 50 questions were crafted by physicists from around the world, and the AI systems achieved high scores. Indian physicist Suvrat Raju believes that no individual can solve a significant fraction of this series of problems without the help of AI. Beyond this case, several quantitative assessments are recording rapid increases. Humanity’s Last Exam, launched in January 2025 to measure skills on various challenges ranging from advanced chemistry to ecology, saw its best systems scoring 8% at launch. This week, Claude Opus 5.5 reached 67.7%. FrontierMath, deployed in November 2024 for hundreds of problems selected and graded by professional mathematicians, initially showed 2% for the best, compared to 94% today. Such sudden and steep jumps are reported across many tests in different scientific fields.

Concrete Examples in Mathematics, Biology, and Robotics

AI systems have been mobilized to propose solutions to long-standing mathematical problems such as the Navier-Stokes Millennium Prize, the Jacobian Conjecture, and open Erdős problems. In several cases, decisive ideas have been provided by AI. In biology, AI has been used to design viruses from scratch and to select new drug candidates, with AI-designed molecules already in trials. In robotics, systems autonomously operate laboratory equipment remotely, enhancing the automation of experiments in biology and materials science and accelerating sectors like batteries and hydrogen fuel cells. Frontier models like Claude, ChatGPT, or Gemini are presented as capable of working at the intersection of multiple disciplines.

Software Productivity and Ambivalent Effects in Cybersecurity

AI systems now exist to augment and automate parts of human work. In software development, they are described as more capable and efficient than the vast majority of engineers, to the point that a single person can now build applications that once required teams. At Anthropic, code production by researchers is nearly ten times higher than what was observed during the company's first four years. These advancements have also increased performance in cybersecurity and cyberattacks: actors are rushing to employ these capabilities to strengthen defenses, while malicious groups attempt to exploit them for hacking, with both positive and negative effects.

Innovation and the Global Economy: An Announced Turning Point

Over half a century, most countries have seen research productivity erode despite an increase in the number of researchers. The speaker believes that AI capabilities are already extending to science and robotics and will spread further. He infers possible changes in macroeconomic parameters such as the pace of innovation and the frequency of the emergence of general-purpose technologies. In 2026, early indicators were noted regarding the ability of AI systems to produce results at the scientific frontier.

At the IMF, a Call to Prepare and Share Benefits

Addressing an audience of economists, bankers, and decision-makers, the speaker asserts that the singularity has begun and urges anticipation of broader changes than expected. His message emphasizes the potential benefits of rapid AI advancement and the need to distribute them prudently and thoughtfully. He argues that AI-guided scientific invention can influence the global economy, transform infrastructure planning and construction, and improve health and personal time. This call was made during the Per Jacobsson conference held in Bangkok on Saturday, October 17, as part of an IMF meeting.

The Singularity Thesis and the Speaker's Journey

The singularity is presented as a phase where progress accelerates beyond human tracking capabilities. The speaker claims it began this year and cites Vernor Vinge's 1993 prediction about the acceleration of progress due to superior intelligences. He envisions both wealth and dangers and announces that he will detail the risks later. He recalls that part of his speech was improvised, describes himself as a narrator, an actor in events, and a translator between communities, and emphasizes his responsibility and willingness to be honest about his own role. He highlights the support of his newsletter and thanks his readers. He describes himself as a unique human, having spent nearly 40 years acquiring the experience for this presentation, and outlines his journey: journalist on data centers, a stint at OpenAI in 2016, and then co-founding Anthropic. He asserts that the anticipated moment has arrived: AI is already affecting national security, economic trajectories, and contributing to its own development.