Ricoh announced the ¥259,800 edition in September for an October 21 release, allocated entirely by lottery through its own store and a dedicated Tokyo space rather than sold first-come. The camera itself is unchanged apart from a heritage-inspired finish and a redesigned lens-ring cap; the story is the allocation method, not the hardware. Two independent camera retailers ran their own separate lotteries for the same release, so this reads less as a pure brand-exclusive sale and more as scarcity-by-lottery spread across a channel — a useful precedent for selling heritage and scarcity without simply raising the price.
Read the source →The festival, running October 30 through November 8 for its 10th edition, is staging "AUTOBAHN," designed by architecture studio ALTEMY, on the old Tokyo Metropolitan Expressway between Sukiyabashi and Ginza. It hosts roughly 150 creators from 13 countries, including the design studios nendo, Jin Kuramoto and Kwangho Lee. It's a scale precedent worth watching for presenting design-forward objects at a temporary public event rather than inside a conventional retail space.
Read the source →The Bee Frontier ($74.99, worn on the wrist or clipped to clothing) and Bee Eclipse ($99.99, a magnetic phone attachment) both record and summarize everyday conversations, free for 14 days of memory, unlimited under an $11.99-a-month Pro plan, with new purchase integration into Amazon's store and Alexa. Earlier reporting treated Bee as discontinued after Amazon's acquisition of the company; it's alive and actively expanding. It's a clear demonstration that the obvious competing strategy to premium, subscription-free hardware is cheap devices paired with recurring revenue.
Read the source →It tracks heart rate, HRV, sleep and activity, and plugs into third-party AI agents — Meta's Muse, Claude, ChatGPT and others — rather than running its own assistant. Preorders are expected next month, with shipping targeted for December or January. Alongside Bee, this is a second concrete subscription comparable landing in the same two weeks, both well below the price band a premium, non-subscription instrument sits in.
Read the source →It works by suppressing peripheral visual clues — signage, architecture, landscape features — that let a model guess where a photo was taken, without touching the subject the image is actually about. In testing, it cut city-level geolocation accuracy from 41.2% down to 9.9%. That's directly useful for any always-on personal camera: a path to removing exactly the detail that turns a private working photo into a location trail, without losing the content that makes the photo worth keeping.
Read the source →An earlier deep-dive corrected its own trigger story: a widely repeated claim that a major tech company "paid $10 million" for a failed airline's internal employee data turns out to be an unapproved bankruptcy-auction bid, still awaiting a US court's sign-off, with a higher $12.5 million cash bid sitting on the same docket and no limits written into the proposed sale order on how the buyer can use the data. The more consequential finding: Japan's privacy law permits this kind of employee-data sale more easily than the US bankruptcy process does, with a new consent-free route for AI-training data arriving in 2028. It also found that storing data only on-device doesn't by itself prevent a vendor from locking a device's owner out of their own records — a hardware security feature on at least one popular single-board computer burns an irreversible key at manufacture, so "local-only" and "locked out" turn out to be fully compatible unless a product is designed against that deliberately.
Two open questions. First, whether to start tracking industrial and factory-floor process-capture research — built for an employer's own monitoring, not an individual's own use — as a new comparison area, alongside creator tools, archive software and museum practice. Second, whether cheap ambient-capture wearables like Bee and Natura, which pair inexpensive hardware with a subscription, deserve tracking as a named category, rather than being noted case by case.
Answer on GitHub →