PSI's AI Proves a 18-Year-Old Quantum Computing Problem, Reshaping Scientific Discovery

Exercise 1
Vocabulary
Repeat each word, its meaning and the example sentence out loud.
autoformalized
adjective/ˌɔːtoʊˈfɔːrməlaɪzd/converted into a formal, coded proof that a computer can check for logical correctness without ambiguity
The AI's solution was autoformalized in Lean, which allowed other researchers to verify its logical validity independently.
to scaffold
verb/ˈskæfəld/to build temporary or supporting structures that help achieve a complex goal; in AI, to provide prompts and systems that guide a model toward solving a specific problem
Instead of relying on the model alone, PSI scaffolds the AI with specialized tools and domain knowledge to tackle physics problems more effectively.
to elicit
verb/ɪˈlɪsɪt/to draw out or bring forth information, ideas, or solutions that are not immediately obvious
The machine learning pipeline was designed to elicit novel hypotheses about superconductivity that hadn't appeared in existing literature.
a negative result
noun phrase/ə ˈnɛɡətɪv rɪˈzʌlt/a finding that rules out or eliminates a possibility, showing what does not work rather than what does; in science, such results are valuable because they narrow the search space
While the conjecture's proof was a negative result, eliminating one quantum computing strategy, it freed researchers to focus resources on more promising approaches.
to hash out
phrasal verb/tu hæʃ aʊt/to discuss thoroughly and work through the details of a problem or plan until reaching agreement or clarity
Leading physicists will gather at MIT to hash out protocols for peer review and publication standards in an era of AI-generated proofs.
Exercise 2
Article
Read the article aloud on your own, or listen to it first.
PSI's AI Proves a 18-Year-Old Quantum Computing Problem, Reshaping Scientific Discovery
Tap any word to see what it means.
Physical Superintelligence, a startup built on the premise that AI can become the dominant engine of scientific discovery, has just deployed an AI model to crack an 18-year-old open problem in quantum computing.
David Turturean, an MIT graduate and researcher at PSI, used early access to Anthropic's latest foundation model to prove the Aaronson-Ambainis conjecture, which concerns whether quantum computers can solve certain problems exponentially faster than classical machines. Scott Aaronson, one of the two researchers who posed the problem decades earlier, confirmed that the proof is logically sound, having been autoformalized—converted into machine-verifiable code in the mathematical language Lean—and reviewed by a team of established researchers.
The solved conjecture constitutes a negative result: it closes off one avenue of research rather than opening doors. As Turturean explains, the finding tells the field, "we're no longer looking there," allowing the field to redirect effort toward more promising directions.
What distinguishes PSI's achievement is its infrastructure. Rather than relying on a general-purpose model alone, the company scaffolds AI with specialized tools, memory systems, and domain-specific prompts.
CEO Matt Pines emphasizes that the system needed "lots of things related to harnesses" and "lots related to tools" to reach this level of discovery at both breadth and depth.
Today, PSI earns revenue optimizing data center operations—modeling heat, power, water, and acoustic effects as coupled physics problems. However, the company's long-term ambition is to identify theoretical breakthroughs and convert them into commercially viable technologies.
Pines notes that while AI has proved formidable in mathematics and computer science—domains where solutions translate into provable code—applying the same approach to experimental physics and the life sciences remains thorny. In these disciplines, computational insights must ultimately reconcile with empirical observation; nature itself must validate the theory.
This distinction matters fundamentally. Unlike a theorem that stands once proven, a physics conjecture requires experimental confirmation to become actionable.
PSI is building what Pines calls "loops" of reinforcement: the AI proposes a hypothesis, experiments test it in the physical world, and successful findings unlock new capabilities.
The breakthrough has also surfaced a practical question: how should the scientific community adapt its peer review and publishing standards? PSI held its result until Anthropic's model release, then shared the proof with relevant experts for scrutiny.
Jim Halverson, a physicist at Northeastern advising PSI, notes that "best practices are still emerging," and leading researchers are convening at MIT this week to deliberate the governance structures necessary for an era where machines generate proofs faster than humans can absorb them.
Turturean himself grappled with the weight of the discovery. He spent time translating the AI's raw proof into a narrative form humans could grasp, recognizing that the academic community would expect him to fully understand and articulate the work—a requirement that raises uncomfortable questions about authorship and accountability.
The broader implication: as AI systems grow more capable at traversing the frontier of human knowledge, the bottleneck shifts. It's no longer the capacity to prove theorems, but the institutional machinery to validate, commercialize, and build trust around those proofs.
Discuss this story out loud with our AI speaking coach.
It asks one question at a time, at your level, and gives you feedback at the end.
Start the discussion Retell it3,000+ learners
Also works inChatGPTandClaude
Exercise 3
Discussion
Have a discussion based on the following questions.
- 1.In your own organization, where do you see AI currently generating valuable but unverified insights, and what frameworks would you need to build in order to act on those insights with confidence?
- 2.Turturean deferred his master's at MIT to join PSI—a major career gamble. How do you weigh the risk-reward calculus when deciding whether to pivot toward emerging AI-driven domains, and what organizational signals would give you confidence in making that decision?
- 3.The article describes PSI as building 'loops' between AI suggestions, experimental validation, and technological capability. In your work, where have you seen feedback loops between computational insights and real-world testing prove most valuable, and where have you seen them break down?
- 4.Aaronson expressed feeling 'satisfied, awestruck, disappointed, and fatigued' simultaneously—a striking mix. What emotional or intellectual responses do you experience when a problem you've worked on gets solved by a machine, and how do those responses shape your view of AI's role in your field?
- 5.If best practices for AI-generated research are still being negotiated, what principles do you think should anchor those practices—reproducibility, human comprehension, novelty, commercial value, or something else entirely?
- 6.PSI is currently monetizing data center optimization while pursuing theoretical breakthroughs. How realistic do you find the company's pivot from services revenue to commercializing theoretical discoveries, and what risks might that transition entail?
Exercise 4
Further discussion
Have a discussion based on the following questions.
- 1.As AI becomes increasingly capable at research, how might the historical prestige hierarchy of academic disciplines shift? Will mathematics and computer science—where solutions are formally provable—remain the domain of machines, while experimental and observational sciences retain more human authority? Or will this distinction collapse as AI tools for experimental design and interpretation improve?
- 2.PSI's model depends on access to cutting-edge foundation models like Anthropic's latest release. What does this create for smaller research institutions or startups in less well-funded domains? Does the commercialization of AI-driven discovery inadvertently consolidate scientific progress within a handful of well-capitalized firms?
- 3.The article hints at a tension between transparency (PSI shared the result with experts early) and competitive advantage (the company withheld publication timing). How should the scientific community and startups balance open inquiry with the financial incentives that drive innovation? Where does that boundary lie, and who should set it?
- 4.If machines can prove theorems and design experiments, does the traditional academic career—built on apprenticeship, gradual mastery, and the slow accretion of reputation—become obsolete? What new forms of intellectual work and credentialing might emerge in its place?
Discuss this story out loud with our AI speaking coach.
It asks one question at a time, at your level, and gives you feedback at the end.
Start the discussion Retell it3,000+ learners
Also works inChatGPTandClaude
Source
This lesson is our own retelling, for English learners, of a story from TechCrunch.
Related stories

A Mathematician and an AI System Both Claim to Have Solved the Riemann Hypothesis
22 November 2026 · C2

Anthropic's IPO: Can a safety-focused AI company survive shareholder pressure?
22 November 2026 · C1

Self-Teaching AI Systems: A Breakthrough or a Risk?
22 November 2026 · B2

AI and sensors help warn communities about dangerous landslides
22 November 2026 · B1



