PSI uses early Anthropic model to resolve 18-year-old quantum computing conjecture
Startup Physical Superintelligence claims solution to the Aaronson-Ambainis conjecture using an unreleased version of Anthropic’s Fable model, validated via autoformalization in Lean.
1 source · single source
- Physical Superintelligence (PSI) reports solving the 18-year-old Aaronson-Ambainis conjecture in quantum computing using an unreleased Anthropic model.
- The solution was autoformalized in Lean and validated by external researchers, including Scott Aaronson.
- PSI’s near-term revenue comes from data center optimization, with a long-term focus on converting physics breakthroughs into technologies.
Physical Superintelligence (PSI), a startup pursuing AI-driven scientific discovery, says it used early access to Anthropic’s unreleased Fable 5.2 model to resolve the Aaronson-Ambainis conjecture, an 18-year-old open problem in quantum computing. David Turturean, a PSI technical staff member and recent MIT graduate, told TechCrunch the solution closes one avenue for quantum speedups but not all, calling it a "negative result" that redirects future research.
Scott Aaronson, who co-posed the conjecture with Andris Ambainis, wrote that while he does not fully understand the proof, he considers it correct because it was autoformalized in Lean and reviewed by a team of top researchers. Aaronson described feeling "satisfied, awestruck, disappointed, and fatigued" by the news, and questioned why the result was absent from OpenAI’s recent Mathocalypse release, suggesting the problem may have been harder than anticipated.
PSI’s approach emphasizes specialized model harnesses, tools, memory systems, and simulation infrastructure to tackle scientific problems at depth. CEO Matt Pines told TechCrunch an effective research system requires breadth, depth, and the ability to explore many alternatives. The company’s near-term revenue comes from using these tools to optimize data centers by simulating variables such as heat, power, water, and sound.
Pines said investors previously required significant explanation to accept the premise that theoretical physics would soon be largely within machines’ domain, but he now sees the field moving in that direction. He added that physics differs from mathematics because "nature gets a vote in your models of reality," and emphasized the need to validate computational results experimentally.
PSI held its result until Anthropic’s model was nearly ready for release, allowing the company to share it with relevant researchers to improve understanding. Jim Halverson, a Northeastern University professor advising PSI, said best practices for AI-accelerated science are still emerging, and that transparency and expert consultation were priorities. Turturean noted the academic expectation to explain AI-elicited solutions in human-comprehensible terms, reflecting the weight of the discovery.
- Nov 11, 2026 · The Verge — AI
Google DeepMind unveils AlphaGenome 2 to predict genetic variant effects on cell types
Trust79 - Nov 11, 2026 · Google DeepMind — Blog
Google DeepMind unveils AlphaGenome 2 to predict DNA variant effects in specific cell types
Trust84 - Nov 11, 2026 · arXiv cs.AI
Researchers demonstrate auditable human-agent teams for autonomous synchrotron X-ray experiments
Trust79