Small Screens, Tiny Cameras: How AI Hardware Is Learning to See Less but Understand More

In the first report, ifanr's editors tested the Yuexingtong X3, a palm-sized e-ink display with a 3.7-inch screen, a 58-gram weight, and a 5.1-millimeter thickness. The device, which attaches magnetically to the back of an iPhone, has become popular on Chinese social media as a portable frame, desktop calendar, or decorative badge. After flashing third-party firmware called CrossMux and connecting the screen to a Mac through the AirPage feature, the editors turned it into a customizable information board using the open-source ink-Macintosh project.

The modified screen displayed a retro Macintosh desktop with widgets for Codex quota, weather, reminders, and calendar. The editors then added their own components: a daily AI news briefing capped at three headlines, a Weibo hot-search module, a Home Assistant widget showing indoor temperature and humidity, and a GitHub Actions monitor that showed the status of code builds. A Gemini-powered scheduling module was added to automatically decide which widgets appeared at different times of day. In the morning, the screen showed meeting times, weather alerts, and news; during coding sessions, it highlighted Codex credit and build progress; late at night, it reduced the display to a single quote.

The article used the project as an example of Mark Weiser's 1995 "calm technology" philosophy, arguing that the e-ink screen's low refresh rate and long battery life let it sit quietly at the edge of attention and only become readable when needed. ifanr wrote that the device's constraints served as a natural filter for information overload.

The second report examined Apple's rumored camera-equipped AirPods, which MacRumors discovered in a promotional video embedded in macOS Tahoe 26.7 RC. The video showed a person wearing AirPods picking up a book, using Visual Intelligence to recognize the title, and asking Siri to save it. According to ifanr, the product carries the development code B790 and is expected to enter mass production as B798 in 2027. The camera is said to have only 1 megapixel, with an active capture mode of 640 by 640 pixels that can be processed up to 1024 by 1024, and a passive mode of 320 by 320 pixels. The code indicates that both earbuds can capture synchronized RGB images, using the same frame ID to align the two perspectives, and that the system discards frames if the head moves too quickly or the lens is obstructed. It also includes "peripheral inference" code that can determine on-device whether a person is in view.

ifanr said the low resolution is intentional, because the camera is not meant to take photos but to feed visual context to AI. Models only need to know that there is a book on the table, a restaurant ahead, or a drink in hand. Reducing pixels lowers data transmission and power consumption, which matters for a device worn on the ear. The article compared the approach to the Guangfan AI earbuds, which use two 2-megapixel cameras for scene recognition, and to Humane's AI Pin. Apple's advantage, it argued, is that millions of people already wear AirPods, so the company does not have to persuade users to adopt a new wearable.

ifanr also described a broader strategy: Apple is turning the earbuds, watch, phone, and future smart glasses into a network of sensors that collect what it called "personal context." The system would combine visual, audio, location, and motion data so that a question like "Is this suitable for me?" asked in a store could be answered without the user having to explain their health, schedule, or preferences. The report noted that 2027, the twentieth anniversary of the iPhone, could be a key moment for Apple's AI hardware, with new designs expected across multiple product lines.

One detail highlighted in the report was that each AirPod may include a controllable LED indicator, likely to show when the camera is active. ifanr said this would be consistent with Apple's privacy stance, though the company has not confirmed any of the plans.