Google pushes autonomous AI
Google introduced Groundsource to turn unstructured news into historical, queryable datasets and DeepMind unveiled Aletheia, an agentic system aimed at autonomous research discovery — both moves push models from QA tools toward independent data engineering and research agents announced announced.
[Groundsource announced]research.google on March 12, 2026 by Google Research engineers Oleg Zlydenko, Rotem Mayo, and Deborah Cohen and released an open-access urban flash-flood dataset of 2.6 million records spanning more than 150 countries. research.google The Groundsource pipeline ingested roughly 5 million global news articles to produce a geo‑tagged time series and Google trained an LSTM forecast model on that dataset to highlight urban flood risk across 150 countries on its Flood Hub, with spatial resolution reported at about 20 km² and data shared with emergency-response agencies. techcrunch.com [Aletheia described]arxiv.org in an arXiv paper submitted Feb 10, 2026 (revised Feb 12) is an agentic math‑research system built by Tony Feng, Quoc V. Le, Thang Luong and colleagues that uses a generator–verifier–reviser loop, an inference‑time scaling law, and intensive tool use to operate end‑to‑end in natural language. arxiv.org DeepMind’s experiments show Aletheia autonomously produced a fully AI‑generated research paper ("Feng26"), contributed to a human–AI paper ("LeeSeo26"), and delivered autonomous solutions to multiple open Erdős problems during a 700‑problem evaluation; Google also paired Aletheia with Gemini 3 Deep Think and widened Deep Think API access for subscribers. arxiv.org
