Quick Hits — August 16, 2026
A quick sweep of the smaller stories moving the AI world: Meta's face-recognizing glasses patent, a viral AI teacher answering her critics, and Fei-Fei Li's thesis on what AI can't touch.
Fei-Fei Li took the Huberman Lab stage with a thesis worth a bumper sticker: AI is a magnifying glass for human ability, not a replacement for it. In the wide-ranging episode, the Stanford professor and World Labs founder argued that models can never reach the private emotional experience that never makes it online, called Socrates history's greatest "prompt engineer," and rejected both extremes of the education debate — banning students from AI outright, or letting it quietly dissolve their drive to learn. Her sharpest warning was aimed at the industry: a handful of tech companies shouldn't get to make society's decisions, and pure market logic must not override social ethics.
A Meta patent application made public this week lays out facial recognition for smart glasses — and privacy advocates are calling it an escalation. Filed in February and viewed by CNET, the 44-page application describes AI identifying people and creating media, including videos, informed by location data and online services. The Center for Democracy and Technology said the tech could put stalking and domestic-violence victims at risk, and the timing is awkward: Meta spent the summer quietly removing dormant face-recognition code that Wired found hidden in its AI app. Patents aren't products, but this one lands as "pervert glasses" creeps into the media lexicon.
Beihang associate professor He Jing, whose plain-spoken AI tutorials made her a Bilibili star, answered her critics in a long interview — and revealed her comment section runs partly on AI. The 35-year-old, whose MOOC clip blew past two million views and sparked debate over her credentials, defended her cross-disciplinary path ("why assume fast people are abnormal, not that slow people picked the wrong track?") and explained why she made an AIGC short film to rebut rumors: in an algorithm-driven feed, clarifications can't outrun them. She also disclosed that the agent fielding her replies once cost about 200 yuan an hour to run before she switched to cheaper models. Take: China's AI-education story is increasingly being told by its teachers, not its platforms.
