You're tuned in to Tech Beat. Here are the stories shaping the conversation today.
Researchers at Berkeley are making the case that a global, hardwired pause on frontier AI training isn't just a good idea — it may actually be technically feasible. The new report argues that physical-level controls on the most powerful training runs could be implemented across borders, shifting the pause debate from philosophy to engineering.
On a related note, the terminology around that same technology is getting a political makeover. President Trump's push to rebrand artificial intelligence as "super intelligence" was widely met with skepticism, but some of Silicon Valley's most prominent figures are quietly adopting the language. When the people building the systems start changing what they call them, the framing of the entire public conversation tends to follow.
And over at the National Institutes of Health, PubMed is deploying BiomedBERT — a large language model trained specifically on biomedical literature — to surface hidden connections between research articles. The goal is to help scientists discover relationships across studies that might otherwise take years to find through manual review. It's a quieter application of AI, but arguably one of the more consequential ones.
Keep surfing. Tech Beat out.
