AI Research
Research that leaves the lab: the results with consequences outside the paper they were published in.
14 stories
Stanford Team Uses AI to Design First Living Viruses
Stanford scientists used AI models to design working bacteriophage viruses that killed drug-resistant bacteria in lab tests, opening a path to faster infection treatments while raising fresh concerns about controlling AI-designed genetic code.
August 6, 2026 · 2 min read
OpenAI's Astra Solves 10 Decades-Old Math Problems
OpenAI says an unreleased internal version of its next model, called Astra, solved ten previously unsolved math and computer science problems for about $2,000 in computing costs, and the bigger story is what kind of work AI is now good enough to attack next.
August 4, 2026 · 2 min read
Google Data: AI Used in 21% of Tasks, Not Full Jobs
A Google study of 15 million real Gemini conversations found AI touches most jobs but only handles a small slice of daily tasks, with full automation still rare, even as companies keep blaming AI for layoffs.
July 28, 2026 · 2 min read
First AI-Designed Drug Enters Final Human Trials
Insilico Medicine has started Phase III trials for rentosertib, a drug built entirely by AI software, marking the first time a medicine discovered and designed from scratch by artificial intelligence has reached the final stage of human testing required for regulatory approval.
July 7, 2026 · 3 min read
Google Study: AI Helps People Name Skin Conditions
Google published peer-reviewed research showing AI tools nearly triple ordinary people's ability to correctly name a skin condition from a photo, but the harder problem, knowing how urgent the situation is and what to do next, remains largely unsolved.
June 12, 2026 · 2 min read
Google DeepMind Spinoff Raises $2.1B to Find Undruggable Proteins
Isomorphic Labs, built on Nobel Prize-winning AI protein research, has raised $2.7 billion total and signed deals with Novartis, Eli Lilly, and Johnson & Johnson to use AI to find drug targets on proteins that traditional chemistry could never reach.
June 11, 2026 · 3 min read
Harvard Study: AI Agents Cut Task Time by 87%
A joint Harvard and Perplexity study, using real data from over 84,000 sessions, finds that AI agents complete tasks 47 times faster than conversational AI search, at a fraction of the cost, and that users hand over more complex, higher-value work as they get comfortable with the technology.
June 9, 2026 · 3 min read
Bezos Backs $500M Bet on Brain-Inspired AI
A startup called Flourish, backed by Jeff Bezos and valued at $2.5 billion, is trying to build AI systems that work more like the human brain: learning continuously, adapting quickly, and consuming a tiny fraction of the electricity that today's AI systems require.
June 10, 2026 · 3 min read
Biohub Releases Free AI Tool to Design Cancer-Fighting Proteins
Mark Zuckerberg and Priscilla Chan's nonprofit Biohub released a free, open set of AI models that can design proteins targeting cancer and immune diseases, compressing research that once took years into days and giving any lab in the world access to tools previously available only to the best-funded research teams.
June 10, 2026 · 3 min read
NVIDIA Publishes Research on Robots That Train in Simulation
NVIDIA presented eight robotics research papers at the world's top robotics conference, showing that robots trained entirely in virtual environments can now perform reliably in real warehouses, factories, and labs, which matters to any business operator considering automation.
June 1, 2026 · 2 min read
OpenAI Model Solves Geometry Problem Unsolved Since 1946
An OpenAI reasoning model independently produced a verified mathematical proof that overturned an 80-year-old geometry belief, marking the first time an AI has autonomously solved a prominent open problem in mathematics, with real implications for how AI may soon contribute to science beyond pattern-matching.
June 2, 2026 · 2 min read
Datadog's Toto 2.0 Brings AI Forecasting to Practical Business Scale
Datadog has released Toto 2.0, a free, open AI forecasting model that, for the first time in its field, reliably gets better as it grows bigger, and this matters to any business that depends on predicting numbers over time.
June 2, 2026 · 3 min read
Nous Research Cuts AI Training Time by 2.5x With No New Hardware
A small open-source AI lab has published a training method that makes building AI models up to 2.5 times faster without buying more computing power, which matters because the cost of training AI is the single biggest reason most industries still cannot build their own models.
June 2, 2026 · 3 min read
NVIDIA Bets on Ineffable Intelligence to Build Post-ChatGPT AI
NVIDIA has partnered with Ineffable Intelligence, a London AI lab that just raised $1.1 billion at a $5.1 billion valuation, to build the computing infrastructure for a new kind of AI that learns from its own experience rather than human-written data, a shift that could eventually produce AI capable of genuine discovery rather than just sophisticated pattern-matching.
June 2, 2026 · 3 min read