Google DeepMind watermarked AI-designed proteins without weakening them, an MIT-led team beat the world’s best Stratego players at a fraction of the training cost, and frontier models keep getting cheaper to run. The…
Three recent developments show how new capabilities are being prepared and measured: the National Aeronautics and Space Administration (NASA) advances a far-infrared telescope mission into design, a shared-propellant satellite completes ground tests, and an open benchmark broadens testing of artificial-intelligence responses in mental-health conversations.
The Mars food challenge shifts the design problem from resupply alone to integrated production, a satellite sequence makes lava movement trackable over time, and a new benchmark lets forecasting agents be tested on the same dated evidence—with clear limits on what each result shows.
The edition covers a more detailed record of water on early Mars, a United States clarification on non-animal drug safety methods, Albania’s entry into the Artemis Accords, and seven state hubs for aerospace technical training.
A new partnership puts evaluators inside an artificial-intelligence lab, the National Aeronautics and Space Administration adds three crew missions, and mouse experiments identify a long-range cortical sleep circuit—each with a clear evidence boundary.
A new X-ray astronomy observation, a semiconductor translation program, a family coordination experiment, and a scroll-screening proof of concept show progress while the evidence remains bounded.
A gene-therapy approval, a refined measure of Earth’s center of mass, a new federal physical-world artificial intelligence funding topic, and guided workshops for older adults—with the evidence limits that determine what changed and what remains unproven.
A patient-specific imaging method, a United States pilot for early drug-trial preparation, a new lunar crater, and a planet still forming—four recent reports with practical or scientific consequences and explicit evidence limits.
A commercialization pilot, a method for steering generative models toward hard constraints, new space-biology investigations, and Djibouti’s Artemis Accords signing—all recent developments whose practical results remain to be demonstrated.
