Follow the latest coverage, related explainers and connected technology stories.
Google and Google DeepMind launched a new institute bringing together perspectives and research on artificial general intelligence, beginning its work with four articles addressing transparency, economic policy, human well-being, and the evaluation of advanced models. One of the articles proposes a U.S. framework for evaluating leading artificial intelligence models before their release.
OpenAI confirmed that discussions about AI safety have been ongoing for weeks with Anthropic and Google DeepMind, alongside calls to establish an independent standards body and a mechanism for verifying the safety of advanced models. This comes amid warnings about antitrust-law risks and opposition within the U.S. administration to tighter regulation out of concern for maintaining competition with China.
Swiss company Fusionality has raised $3.7 million, or CHF 3 million, to develop control systems and simulation environments that help fusion companies build power plants capable of supplying electricity to the grid more quickly. The company is initially focusing on magnetic-confinement reactors, using artificial intelligence to improve specific parts of the control system rather than manage it entirely.
Google DeepMind announced AlphaGenome Atlas, a database that predicts the effect of every possible single-nucleotide change in the human genome, based on the AlphaGenome model. The database allows researchers to search a one-petabyte dataset and use the AVI score to rank the most impactful variants.
Google DeepMind has selected 16 startups, nonprofits, and a research team to participate in the first cohort of the AI for the Planet Accelerator in Asia-Pacific. The teams will receive three months of mentorship and technical support to use AI models for biodiversity, sustainable agriculture, and carbon solutions.
OpenAI’s Astra model is expected to use “recurrent depth,” a technology that processes queries through repeated loops instead of the usual sequence, potentially making reasoning traces less clear to observers. The company says the technology’s use will be limited and that it remains committed to preserving readable chains of thought, but AI safety researchers warn that its use could expand.