AI Breakthrough: Recent MIT Graduate Uses Early Model Access to Solve 18-Year-Old Quantum Computing Problem

David Turturean, a recent MIT graduate and a member of the technical staff at the startup lab Physical Superintelligence (PSI), has successfully utilized early access to Anthropic’s Fable 5.2 model to solve an 18-year-old quantum computing problem. The achievement marks a significant milestone in the intersection of artificial intelligence and theoretical physics, raising profound questions about the future of academic research and scientific publishing.

Turturean, who chose to defer his plans to pursue a master’s degree at MIT’s prestigious computer science and AI lab to join PSI, reflected candidly on the gravity of the shift. Speaking with TechCrunch, he admitted that he is entirely open to the possibility that this milestone could mark the end of an era for traditional academic publishing. In a matter of a few months, he suggested, the traditional justification for publishing academic papers may fundamentally shift as artificial intelligence systems begin to autonomously accelerate the generation of scientific knowledge.

Cracking the Aaronson-Ambainis Conjecture

The mathematical and computational problem in question is the long-standing Aaronson-Ambainis conjecture, which addresses the fundamental capabilities of quantum computers and evaluates the precise conditions under which they can solve certain classes of problems faster than classical computers.

Scott Aaronson, one of the original researchers who posed the problem nearly two decades ago, shared his perspective on the surprising resolution. Writing on his personal blog, Aaronson noted that while he does not yet completely understand every nuance of the resulting proof, he firmly believes it is correct. His confidence stems from the fact that the solution was autoformalized in Lean—a specialized coding language designed specifically for mathematics—and subsequently explained by a team of top-tier researchers with whom PSI shared the solution prior to general release.

The newly derived solution effectively closes the door on one specific theoretical avenue that quantum computers might otherwise have pursued to solve certain complex problems at high speeds, though it leaves other pathways open. As Turturean explained, it represents a definitive negative result that allows the scientific community to streamline its efforts by knowing precisely where they no longer need to look.

More broadly, the breakthrough reflects a profound upheaval currently rippling through the scientific community as advanced AI systems increasingly demonstrate the capacity to tackle complex problem-solving in theoretical domains.

Aaronson described a complex mix of feelings—including satisfaction, awe, disappointment, and intellectual fatigue—upon learning of the news. He noted his surprise that the full proof of the Aaronson-Ambainis conjecture had not appeared in OpenAI’s recent mass releases, often referred to as their "Mathocalypse dump," pointing out that because other researchers surely attempted it, the problem proved significantly harder than many others that had already fallen to automated systems.

The Role of Specialized AI Infrastructure

This successful resolution serves as a strong validation of PSI’s core operational model, which focuses on solving complex scientific problems through the deployment of specialized model harnesses, sophisticated scaffolding, and robust computational tools.

PSI CEO Matt Pines elaborated on this philosophy in an interview with TechCrunch, emphasizing that an effective, modern research system must integrate a wide array of support structures. According to Pines, researchers need systems deeply connected to customized harnesses, specialized tools, advanced memory systems, and comprehensive simulation infrastructure to tackle massive problems simultaneously across both breadth and depth, allowing the system to thoroughly consider a wide range of potential alternatives.

Currently, PSI derives its primary revenue from leveraging its computational tools to optimize physical data centers. The startup simulates a complex array of environmental variables—including heat, power, water usage, and sound—where the overall performance of a site, as well as its direct impact on the surrounding community, functions as a direct consequence of physical laws. However, the long-term strategic vision for the company involves translating these theoretical physics breakthroughs into tangible, real-world technologies.

Pines acknowledged that just six to nine months ago, convincing investors of this trajectory required a significant leap of faith. He frequently had to ask stakeholders to trust that theoretical physics would soon transition primarily into the domain of machines. While he admits that the vision is not fully realized everywhere just yet, developments like the solution to the Aaronson-Ambainis conjecture demonstrate that the future is rapidly approaching.

Bridging the Gap Between Math and the Real World

Despite the remarkable successes AI has recently claimed in highly abstracted fields like mathematics and computer science—where solutions can be neatly expressed and verified in formal code—translating these capabilities into less abstracted, empirical disciplines like the life sciences or physics has proven considerably more challenging. Highlighting these ongoing hurdles, the startup Discovered Materials recently launched a $1 million grand challenge aimed at researchers attempting to use artificial intelligence to untangle complex physical phenomena such as superconductivity and heat dissipation.

Pines underscored the fundamental difference between abstract math and empirical physics, pointing out that physics deals directly with the real world, where nature ultimately validates or rejects theoretical models. PSI’s ongoing strategy focuses heavily on figuring out how to take computational results generated by AI and validate them through physical experiments in the most efficient manner possible.

If a discovered computational fact represents a genuine truth about nature, Pines noted, it becomes a rule that researchers can exploit to build entirely new devices, allowing humans to perform actions in the physical world that were previously impossible. Building automated feedback loops to capture and apply these insights is a core objective of PSI’s ongoing research initiatives.

Establishing New Standards for AI-Driven Science

As specialized research entities and "neolabs for science" emerge as primary vehicles for potentially groundbreaking discoveries, participants in the field are still actively grappling with the appropriate standards, ethics, and protocols for this new era.

Because PSI had to hold its solution confidentially until Anthropic’s Fable 5.2 model was nearly ready for public release, the startup was uniquely positioned to share the findings with relevant academic experts ahead of time to better understand the implications of the proof.

Jim Halverson, a professor at Northeastern University who serves as an advisor to PSI, emphasized the importance of maintaining academic integrity during this transitional period. As an academic, Halverson felt a strong responsibility to ensure things were handled appropriately. He noted that he will be attending an upcoming summit at MIT where leading physicists will gather to discuss and formulate a collective response to the dramatically accelerated pace of scientific discovery driven by artificial intelligence. According to Halverson, because best practices are still actively emerging across the scientific landscape, transparency and direct collaboration with domain experts represent the best path forward.

Reflecting on the personal impact of the discovery, Turturean described the unique experience of reviewing a completed proof generated by an artificial intelligence system. Having had time to sit with the results, it marked the first time he had taken an AI-elicited solution in print and carefully read through it to determine how it could be reframed and presented more naturally for a human audience.

For Turturean, who recognizes that the academic community will naturally expect him to be capable of explaining the result in detail, the experience served as a striking realization of the immense weight and capability of modern artificial intelligence.