Piloting the world's first double-blind AI evaluations
Google’s double-blind evaluation pilot treats benchmark contamination as a systems-security problem, not merely a testing etiquette issue.
Desk
Google, DeepMind, and Gemini developments.
Google’s double-blind evaluation pilot treats benchmark contamination as a systems-security problem, not merely a testing etiquette issue.

Google is turning speech recognition into an intent-aware interface layer, raising the product value—and the risk—well beyond literal transcription alone.
Persistent social worlds are a harder test of useful intelligence than games with fixed scores and clean resets. The limits of the evidence are as important as the headline announcement.

Google is using football sponsorships to make consumer AI and phones recurring match-day habits. The limits of the evidence are as important as the headline announcement.

Google is combining measured coding gains with promotional pricing, making agent economics central to the model release. The limits of the evidence are as important as the headline announcement.
Sign-language translation reaches phones through multilingual training, on-device pose extraction and community involvement rather than word-by-word transcription. The limits of the evidence are as important as the headline announcement.
AMIE’s move from text to live video targets the observational core of telemedicine, but simulated consultations cannot establish safety in messy clinical practice.

Google is moving interpretation and action closer together inside its ad stack, saving analyst time while deepening dependence on the platform being measured.
WeatherNext’s real advance is unifying cyclone track and intensity forecasts at coarse resolution, potentially giving forecasters more warning without enormous local simulations.

Google’s July catalog shows AI becoming a connective layer across its products, with account access and cross-app context now more strategic than any model release.

Huge enrollment validates demand for agent-building skills, but the drop from registration to capstone work reveals the limits of headline participation metrics.

Gemini Robotics ER 2 is Google’s attempt to make a robot’s planner watch video, time its next step and coordinate other machines without a stop-and-think pause.