Prefer to listen? Here's the narrated briefing on this area.
Listen to Long-term abundance and scientific acceleration
Plus unlocks audio streaming. Pro adds downloadable audio, video, certificates, and more.
Evidence & Reach
Reid Hoffman's 2025 book *Superagency* argues that AI as cognitive amplification creates a positive feedback loop: better tools → better scientists → better tools. The 2024 Nobel Prizes in Physics (Hinton, Hopfield — neural networks) and Chemistry (Hassabis, Jumper — AlphaFold) recognized AI methods themselves as fundamental scientific contributions. Materials science discovery is accelerating: Google DeepMind's GNoME paper identified 2.2 million new stable inorganic materials in one paper. Fusion plasma control is now AI-assisted at multiple research reactors. Drug discovery timelines that took 10-15 years are now seeing AI-accelerated phases of 2-3 years. By 2026 the discoveries themselves were landing: OpenAI's reasoning model disproved a 1946 Erdős conjecture in discrete geometry, Google's Empirical Research Assistance system wrote expert-level scientific code published in Nature, and DeepMind's AlphaEvolve reported measurable algorithmic wins across seven research and industry fields.
Sources:Reid Hoffman — SuperagencyNobel Prize 2024 — Chemistry (AlphaFold)GNoME — materials discoveryReasoning model disproves a geometry conjecture (OpenAI, 2026)Google ERA — from Nature to computational discovery (2026)
A Specific Story
In 2025, a team at MIT used machine learning to screen 39,000 molecular compounds for antibiotic activity against drug-resistant *Acinetobacter baumannii* (a leading cause of hospital-acquired infections worldwide). They identified abaucin, the first novel structural antibiotic class in 60 years — a class of medicines that humanity had effectively stopped discovering in the 1980s. Clinical trials began in 2025. Multiply this story across every major disease category and the scale of what AI-augmented science can unlock starts to come into focus.
What's Next
AI-driven scientific research as a continuous loop (hypothesis generation → simulation → experimental design → result analysis → next hypothesis, with humans supervising at each stage but the loop itself running 24/7), fusion-energy commercialization accelerated by AI-controlled plasma, and the bigger thesis: that the rate of *useful scientific discovery per year* — currently flat or declining across many fields — finally starts climbing again.
Recent Top AI Stories showing progress in this area.
Mistral's open Leanstral 1.5 aces a formal-math benchmark and finds real software bugs
Mistral released Leanstral 1.5, an Apache-2.0 open model for formal mathematical proof in Lean 4. The mixture-of-experts design carries 119 billion total parameters but activates just 6 billion, and it scores a perfect 100 percent on the miniF2F benchmark while solving 587 of 672 problems on PutnamBench. Beyond math, the model flagged five previously unknown bugs across 57 open-source repositories during testing — including a critical integer-overflow flaw in a decoding library — pointing at formal verification as a practical path to safer software.
Mistral acquires Emmi AI to launch Physics AI for engineering, with Airbus, ASML, Safran, and Siemens Energy on board
Mistral acquired Vienna-based Emmi AI and used the deal to launch Physics AI, a class of data-driven models that learn from physics solver outputs and predict the behavior of physical systems in seconds on a single GPU, replacing simulations that traditionally take hours or weeks. Mistral named Airbus, ASML, Safran, and Siemens Energy as launch partners, targeting aerospace, semiconductors, energy, and industrial equipment. The offering pitches three concrete use cases: exploring thousands of design variants for new products, optimizing factory tooling, and running real-time digital twins on live sensor data — engineering acceleration with named industrial deployments rather than future-tense lab promises.
OpenAI's new reasoning model disproves a 1946 Erdős discrete-geometry conjecture
OpenAI says a new general-purpose reasoning model discovered a counterexample to a 1946 Paul Erdős conjecture about optimal unit-distance configurations, a problem mathematicians had assumed was solved by the obvious square-grid construction. Mathematicians Noga Alon, Melanie Wood, and Thomas Bloom reviewed the result and published companion remarks endorsing the disproof. The claim arrives seven months after OpenAI's previous Erdős announcement was shown to be a misrepresentation of prior literature, so the named verifications matter.
Google's ERA writes expert-level scientific code, published in Nature
Google Research published **ERA — Empirical Research Assistance — in *Nature* on May 19**, an AI system using tree-search over thousands of candidates to write and optimize scientific code across genomics, public health, satellite imagery, neuroscience, and time-series forecasting. Concrete wins: ERA-built forecasts ranked at or near the top of the CDC's leaderboards for flu, COVID-19, and RSV; a California water-runoff model beat the state's official Bulletin 120 outlook; and a retail forecasting variant met or exceeded both commercial consensus and Chicago Fed estimates. Built on Gemini.
SandboxAQ wires its quantum-chemistry models into Claude for drug discovery and materials science
SandboxAQ integrated its Large Quantitative Models for quantum chemistry, molecular dynamics, and microkinetics directly into Claude, letting computational and research scientists at pharmaceutical and materials companies query simulation-grade physics models in natural language without their own digital infrastructure. "For the first time, we have a frontier quantitative model on a frontier large language model that someone can access in natural language," said Nadia Harhen, SandboxAQ's general manager of AI simulation. Anthropic has not yet detailed the underlying integration mechanism.