Implications: Despite the hype about using large language models (LLMs) to diagnose and treat disease, LLMs are fundamentally trained on language—so it’s not surprising they excel at expressing complex information in easy-to-understand terms. I wouldn’t be surprised if LLMs transform how doctors talk to patients before they revolutionize treatments.

"The shape of the shadow of the thing." Ethan Mollick, a professor at The Wharton School, is one of the most consistently provocative thinkers on how businesses can do real work with generative AI. In a recent post, he pauses to reflect on a transitional moment in AI's development, as we pass the initial hype wave and begin incorporating AI into our workdays: What is this "thing" we call AI, anyway? Which of its limits are transient, and which are fundamental? And what will AI turn into as it grows ever-more powerful?

A new AI tool can diagnose brain tumor subtypes in just 90 minutes. Researchers in the Netherlands announced new deep-learning algorithm that can quickly diagnose brain cancer subtypes from fragments of a tumor's DNA. Using the tool, surgeons can sample a tumor at the start of surgery and, while the patient is still on the operating table, update their surgical plan to target the patient's specific cancer. The algorithm's ability to recognize shreds of DNA is "akin," the New York Times reports, "to someone recognizing an image based on only one percent of its pixels, and from an unknown portion of the image."

The Atlantic's deep dive into OpenAI CEO Sam Altman. I can't beat The Atlantic's copywriters, who tease this piece as covering Altman's "ambitious, ingenious, terrifying quest to create a new form of intelligence." If you're a regular reader of AI news, you won't find much new here: It includes a 101-level explainer of ChatGPT's technology, a rundown of Altman's interactions with world leaders, and some existential dread about AI's long-run implications. But few articles have so effectively captured Altman's blend of hubris, humility, and humanity — illuminating a personality that, for better and worse, is changing our world fast.

AI peers inside 2,000-year-old scrolls buried by Vesuvius. This one has nothing to do with healthcare, but it's just so cool. Spurred by a $1 million prize, computer scientists are making fast progress at using machine learning to read papyri buried in the ancient eruption of Mount Vesuvius. The scrolls are so delicate they crumble if unrolled, and their ink is all but invisible — so competitors must work from 3D scans of unopened scrolls taken in a particle accelerator. If successful, the effort could double humanity's corpus of text from antiquity. (In case you're wondering, the first word deciphered was "ΠΟΡΦΥΡΑϹ," which means "purple.")