Noah Smith
· Noahpinion
· March 02, 2026 at 08:02
· ⏱ 28 min read
| Read on Substack ↗
Summary
Noah Smith argues that current AI, combining human-level reasoning with computer superpowers (speed, memory, tirelessness), already constitutes superintelligence and is actively accelerating scientific discovery across math, physics, biology, and social science. This means the bottleneck for scientific progress is shifting from human cognition to AI scale, with profound implications for productivity gains in research-intensive sectors and for companies supplying AI tools and automated labs.
•AI has already solved ~100 Erdős problems since October 2025, many via automated literature search and some with original proofs, per Terence Tao.
•Tao describes AI enabling a new 'scaling' approach to math: sweeping 1,000 problems with 20 techniques to pick low-hanging fruit, something humans never do.
•Google DeepMind's internal model 'Aletheia' autonomously solved four Erdős problems and contributed to publishable papers in algorithms, economics, and physics.
•AI connected to an autonomous lab (Google with Ginkgo) reduced protein production cost by 40% and compressed 150 years of traditional lab work into weeks.
•AI tools (Claude Code, Gemini Deep Think) are acting as 'tireless research assistants' — checking proofs, finding errors, doing literature synthesis, and automating grunt work for scientists.
•Social science is seeing radical cost reduction: AI can generate publishable-quality papers in hours, with researchers like Paul Novosad and Yascha Mounk reporting top-journal-quality output with minimal human input.
•The 'burden of knowledge' — the growing time needed for humans to master existing literature — is being shattered by AI's ability to synthesize across disciplines instantly.
•Smith notes that AI's superintelligence comes from combining roughly human-level reasoning with pre-2022 computer superpowers (speed, memory, scale) — not from surpassing humans in taste/judgment.
The article explicitly describes a collaboration between Google and Ginkgo Bioworks where GPT-5 connected to an autonomous lab reduced protein production cost by 40% and compressed 150 years of work i
The article explicitly describes a collaboration between Google and Ginkgo Bioworks where GPT-5 connected to an autonomous lab reduced protein production cost by 40% and compressed 150 years of work into weeks. This validates Ginkgo's platform-as-a-service model and signals growing commercial adoption of AI-driven synthetic biology.
Risk: Partnership dependency on Google; Ginkgo's revenue still small and path to profitability uncertain.
The article repeatedly highlights Google DeepMind's AI breakthroughs (Gemini Deep Think, Aletheia solving math problems, autonomous lab partnership with Ginkgo) as evidence that current AI is already
The article repeatedly highlights Google DeepMind's AI breakthroughs (Gemini Deep Think, Aletheia solving math problems, autonomous lab partnership with Ginkgo) as evidence that current AI is already superintelligent and driving scientific acceleration. Google's position as the AI leader in this narrative benefits from increased enterprise/cloud demand and positive scientific reputation.
Risk: Regulatory scrutiny on AI safety and potential competition from open-source models or other hyperscalers.
This newsletter, published March 02, 2026,
features Noah Smith
discussing DNA, GOOGL.
2 trade ideas extracted by AI with direction and confidence scoring.