Nature (journal)

2026-10-19

STARTUPS · 1 SOURCE · TechCrunch AI

Laplace Intelligence develops AI red-teaming platform for robots and self-driving cars

Laplace Intelligence, founded by Howie Sun and Henry Liu out of the University of Michigan, is building automated red‑teaming for physical AI—using a post‑trained Nvidia Cosmos simulation grounded in real‑world map and traffic data (including TomTom) and AI agents that generate corner cases to stress‑test robots and autonomous vehicles. The startup is presenting at TechCrunch Disrupt’s Startup Battlefield 200 and says it has begun generating revenue.

6.0

2026-10-14

RESEARCH · 1 SOURCE · The Verge AI

Microsoft publishes open-source 'Probabilistic Bias Correction' to improve 2–6 week forecasts

Microsoft researchers published a Nature paper and released open-source software called Probabilistic Bias Correction (PBC) that uses machine learning to correct systematic biases in existing weather models; the team reports PBC improved ECMWF model performance for most precipitation and temperature targets and can run on CPUs. The paper lists coauthors from MIT, ECMWF and others and the method is being tested with partners such as Rhiza Research and the Kenya Meteorological Service.

8.0

2026-10-14

RESEARCH · 1 SOURCE · MIT News AI

ML postprocessing doubles 2–6 week weather forecast accuracy, paper finds

Researchers including scientists from MIT developed a machine-learning postprocessing model that learns and corrects errors in existing forecasting systems; when applied to a gold-standard model it doubled forecasting accuracy in the 2–6 week (subseasonal) range, placed first in an international forecasting competition, and for the first time beat climatological averages, the team reports in Nature. Coauthors include Judah Cohen, Lester Mackey, Microsoft researchers, ECMWF contributors, and others.

8.0

2026-10-13

STARTUPS · 1 SOURCE · TechCrunch AI

Laplace Intelligence builds AI-driven red-teaming platform to stress-test robots and self-driving cars

Ann Arbor startup Laplace Intelligence is developing automated red-teaming for physical AI—using a real-world‑grounded, post‑trained version of Nvidia’s Cosmos simulator and AI agents that generate corner cases—to find failures in robots, autonomous vehicles, and other deployed physical systems; the company is presenting at TechCrunch Disrupt’s Startup Battlefield. Founders Howie Sun and Henry Liu base the product on prior University of Michigan research and aim to surface edge cases before real‑world deployment.

6.0

2026-10-01

RESEARCH · 1 SOURCE · The Decoder

Ataraxos (CMU/NYU/Stanford/MIT) defeats top Stratego player Pim Niemeijer

Researchers from Carnegie Mellon, NYU, Stanford and MIT published a Nature paper introducing Ataraxos, an AI that the authors say is, to their knowledge, the first to reach superhuman performance in Stratego; in an official 20-game match it beat Pim Niemeijer 15–1–4. The system trains purely by self-play, uses a belief network and a regularization schedule to maintain strategic diversity, and—according to the paper—was trained in about a week on 16 Nvidia H100 GPUs (plus four days on four GPUs for the belief network) with an estimated compute cost under $8,000 (2025 prices).

8.0

2026-09-30

RESEARCH · 1 SOURCE · MIT News AI

Ataraxos AI from MIT-led team achieves superhuman performance in Stratego

Researchers from MIT, Carnegie Mellon, NYU and Stanford developed Ataraxos, an AI trained with self-play and efficient algorithms that defeated top-ranked human Stratego players by a large margin and outperformed prior models while requiring far less compute; the work is reported in Nature. The system also generalizes to other imperfect-information strategic games and could inform decision-making in real-world hidden-information problems.

8.0

2026-09-30

MODELS · 1 SOURCE · Google

Google AI model ranks #1 in CDC FluSight forecast of 2025–26 flu hospitalizations

The CDC’s end-of-season FluSight analysis found that, of 39 eligible models, Google’s best model most closely matched observed U.S. flu-related hospital admissions for the 2025–26 season (Oct–May). Google says those forecasts were developed using its Empirical Research Assistance (ERA) tool; ERA research was published in Nature and the technology is now available to trusted testers as part of Google’s experimental science tools.

7.0

2026-09-21

MODELS · 1 SOURCE · Microsoft Research

RetroChimera retrosynthesis model published in Nature and open-sourced

A retrosynthesis framework called RetroChimera — which ensembles two sub-models (R-SMILES 2 and NeuralLoc) to improve prediction of chemical synthesis routes — was published in Nature; the authors also open-sourced the implementation and weights and report extensive validation including recall of rare reaction types and zero-shot transfer. In blind tests, expert chemists preferred RetroChimera’s predicted disconnections over those from its constituent models and prior approaches.

7.0

2024-03-20

RESEARCH · 1 SOURCE · Google Research

Google Research paper shows ML improves global flood forecasts for ungauged watersheds

Google Research published a Nature paper, “Global prediction of extreme floods in ungauged watersheds,” reporting that machine learning models (including LSTM-based approaches) can materially improve global-scale flood forecasting where local streamflow data are scarce—extending forecast reliability on average from zero to five days and enabling Flood Hub to deliver up to seven-day river forecasts across reaches in over 80 countries; model evaluation was done in collaboration with ECMWF and the team has open-sourced related hydrology datasets.

8.0