Today's Hallucination HQPanic Sells Better Than Nuance, Apparently
Silicon Valley is having a serious, necessary conversation about AI risk. Unfortunately, so is everyone else — and the everyone-else version involves considerably more horsemen of the apocalypse. Semafor traces how measured concern from researchers became incendiary headlines, bypassing the "complicated but important" stage entirely. The story of AI risk isn't wrong, exactly. It's just been through the blender, poured into a skull-shaped glass, and served with a little umbrella.
Source: Semafor
The Entire AI Industry Agreed on Something. Washington Promptly Disagreed
Dario Amodei published a long open letter calling for slower AI development. Then — in what may be the most surprising sentence ever written — Sam Altman, Elon Musk, and Demis Hassabis all publicly agreed with it. A unified AI industry asking for restraint should carry some weight. Trump and House Speaker Mike Johnson, however, believe the sector is overreacting, which is a bold position to take when the people building the thing say it needs watching.
Source: The Verge
On the Question of Whether AI Will Kill Us All
An outgoing Anthropic researcher warned that AI could "kill all humans" by 2030, which is the sort of resignation letter that tends to generate coverage. CBC does the sensible thing and asks actual experts to unpack it — distinguishing between near-term harms (real, documented) and extinction-level risk (contested, speculative, but not dismissed by serious people). The answer, characteristically, is "it's complicated," which is less satisfying than apocalypse but considerably more useful.
Source: CBC News
Xi Jinping Arrives at BRICS With a Gift Basket and a New Club
At the BRICS summit, Xi Jinping proposed a China-led open-source AI community and invited members into the World AI Cooperation Organization — 29 countries, no EU members, founded in Beijing. The EU, having written its own AI rulebook, declined to join someone else's. Xi pitching open-source cooperation while Beijing considers restricting its own models abroad is a tension the press release did not address. Geopolitics, like AI, works best when you don't read the fine print.
Source: The Next Web
The AI That's Actually Helping Doctors Make Better Decisions
Amid all the extinction discourse, a quietly remarkable study: a multimodal AI model — one that combines different types of patient data — outperformed standard biomarkers in predicting which lung cancer patients would respond to immunotherapy. Tested across a large international real-world dataset, it also measurably improved physician decision-making. No one is being replaced. Doctors are just getting a very well-read second opinion that doesn't bill by the hour.
Source: Nature
The robots may kill us all, but first they'll help with the oncology.
|