When you crack open your mailbox, it’s almost as if your letters just appear. Long before the days of speedy, overnight mail deliveries, postal service workers meticulously sorted through letters by hand and transported mail on horseback. For more than 250 years, the US Postal Service has worked behind the scenes to build a faster delivery network, and this mission has quietly pushed it to the forefront of technology.
Latest Tech News
The tech that the US Post Office gave us
“Most people treat the Postal Service like a black box,” USPS spokesperson Jim McKean tells The Verge. “You take your letter, you put it in a mailbox, and then it shows up somewhere in a couple of days. The truth is that that piece of mail gets touched by a lot of people and machines and transported in that period of time — it’s a modern marvel.”
One of its big breakthroughs took place in 1918 with the introduction of airmail. The USPS worked with the Army Signal Corps to use leftover World War I aircraft to launch the service, and the planes were as barebones as they could get. An excerpt from a 1968 issue of Postal Life called the early aircraft “a nervous collection of whistling wires” with “linen stretched over wooden ribs, all attached to a wheezy, water-cooled engine.”
At the time, pilots literally risked their lives delivering mail — 34 of them died between 1918 and 1927. “There was no commercial aviation, no airports. There was no radio. There was no navigation,” USPS historian Stephen Kochersperger says. “The Postal Service had to develop all of those things just for getting the mail delivered.”
Once the USPS established that it could reliably deliver mail by plane, Congress allowed it to contract airmail service to commercial aviation companies, laying the groundwork for the major airlines that we know today, like American Airlines and United Airlines. Along with getting paid for delivering mail, contractors found that they could make even more money by carrying passengers with their cargo. “That was where commercial aviation took off,” Kochersperger says.
Airmail routes gradually began to expand internationally, first to Canada and then to Cuba. But a couple decades later, the USPS experimented with a novel form of delivery: mail-by-missile. In 1959, the USPS and the US Navy loaded a Regulus I missile with two mail containers that had 3,000 letters in total. The missile traveled 100 miles in around 23 minutes, successfully landing at a Navy base in Mayport, Florida, with the help of a parachute. Despite its success, the idea never took off. It turns out missiles just can’t carry that much mail. And overall, this rather ridiculous demonstration was more of a stunt to show force during the Cold War, according to the Smithsonian.
Back on the ground, the USPS set its sights on improving the speed of mail processing. Though it began experimenting with a mail canceling machine in the 1920s, which put a mark on used postage, it wasn’t until the 1950s that it deployed an electromechanical sorting machine. Instead of manually sorting mail using the “pigeonhole” method, in which workers would insert pieces of mail into different compartments inside the post office depending on the address, the machine could do that for them.
“The Postal Service is a driver of technological change.”
The Transorma multi-position letter sorting machine measured 13 feet high and was split across two levels. It carried mail on a conveyor belt from its lower level to a group of five postal workers at the upper level. The clerks would then use a keyboard to enter information about their destination. Based on the inputted information, the machine would then transport letters to different trays and drop them into chutes that brought them back to the lower level. But as the volume of mail increased in the years after World War II — going from 33 billion pieces of mail per year to 66.5 billion between 1943 and 1962 — the USPS needed a way to keep up.
For years, the USPS had depended on clerks to memorize dozens of delivery schemes that they would use to sort letters, preparing them for carriers to distribute throughout town. “That changed dramatically in 1963, [with] probably the biggest innovation the Postal Service has ever rolled out, called the ZIP code,” Kochersperger says. “For the first time, mailing lists could be digitized in computers and sorted in new ways.”
The ZIP code — short for Zone Improvement Plan — uses its first digit to indicate which region of the US a parcel is headed, the second and third to signal a nearby major city, and the final two to indicate a specific delivery area. The pace of innovation at the USPS ramped up following the introduction of the ZIP code, with many subsequent innovations building on its foundation.
That includes the USPS’s adoption of optical character recognition (OCR), a widely used technology that converts written or printed words into machine-readable text. In 1965, the USPS began to send large volumes of mail through OCR machines, allowing a “digital eye” to recognize addresses and automatically sort letters. If the machine couldn’t make out a person’s handwriting, the USPS would send an image to a remote encoding center (REC) for human review.
At one point, the USPS had as many as 55 RECs, but now only one remains in Salt Lake City, Utah. “As our computer systems have gotten better at recognizing handwriting, we’ve gotten to the point where it’s significantly reduced the number of letters that have to go to remote coding,” McKean says. Today, the USPS’s OCR technology can read handwritten mail at nearly 98 percent accuracy, while machine-printed addresses bump its accuracy to 99.5 percent.
That’s thanks to advances in machine learning, which the USPS, too, has been using in the background for more than 20 years; it first started using a handwriting recognition tool in 1999. The USPS is currently in the middle of a 10-year modernization plan, which includes investments in technology, such as AI. However, the plan has faced criticism for raising the price of stamps and causing service disruptions in some areas.
“The Postal Service is a driver of technological change,” McKean says. “It’s hard to overstate the amount of technology that the Postal Service has been involved in either popularizing or innovating over the last 250 years.”
Continue Reading
Mark Zuckerberg’s AI hiring spree is costing a lot of money. His investors don’t care.
Meta’s stock price shot up over 10 percent on Wednesday after the company reported better-than-expected earnings. Revenue generated in the second quarter was $47.5 billion, up 22 percent from a year ago. Daily users across Facebook, Instagram, Threads, and WhatsApp grew to almost 3.5 billion. Meta also warned Wall Street that it would spend more on data centers and hiring next year. In response to all this, the company’s valuation increased by over $175 billion, or more than 12 Scale AI deals.
”Our business continues to perform very well, which enables us to invest heavily in our AI efforts,” Zuckerberg said during today’s earnings call. Meta’s cash cannon is now fully pointed at his new moonshot of achieving superintelligence, or as he puts it, AI that “surpasses human intelligence in every way.” He bragged about providing the richly compensated members of his new superintelligence lab “access to unparalleled compute” for training new models that will “push the next frontier in a year or so.”
Zuckerberg’s last moonshot was the metaverse, which came up only once at the very end of today’s earnings call. It’s too early to compare the two projects, but they share a key similarity: they need the kind of funding that only a company like Meta can provide.
Where AI differs from the metaverse, however, is that it appears to be already improving Meta’s ads business. A new AI model for delivering ads has driven approximately five percent more conversions on Facebook and three percent more conversions on Instagram, according to CFO Susan Li. Large language models are also starting to power how posts are ranked in feeds across the company’s apps, including Threads.
While Meta is still spending heavily on the metaverse (it’s on track to spend a total of $100 billion on its Reality Labs division this year), there’s no mistaking the fact that AI is officially Zuckerberg’s top priority. This time, though, he’s playing catch-up in a heated race, not trying to invent a new platform from scratch. The stakes are much higher, even if he’s playing the game with house money.
YouTube videos with strong profanity in the first seven seconds (words like “fuck”) are now eligible for full monetization, according to a video from Conor Kavanagh, YouTube’s head of monetization policy experience. Previously, these kinds of videos were only eligible for “limited ad revenue.”
Changes to YouTube’s inappropriate language policies have long been a sore spot for creators. In November 2022, the company began to potentially limit ad revenue if profanity was used in the first 8–15 seconds of a video. ProZD, whose real name is SungWon Cho, published a video where, after waiting 15 seconds, he called the policy change “the dumbest fucking shit I’ve ever heard.” (He later said that the video was demonetized.) YouTube adjusted its policies in March 2023, including allowing videos with profanity in the first 8–15 seconds to be eligible for ad revenue.
I asked ProZD his thoughts about Tuesday’s change. “It’s about fucking time.”
The company originally restricted monetization for videos with swearing at the start of videos to “align with broadcast standards,” Kavanagh says. “Advertisers expected ads on YouTube to have distance between profanity and the ad that just served.” However, “those expectations have changed,” he says, “and advertisers already have the ability to target content to their desired level of profanity.”
While the only specific example of “strong” profanity Kavanagh provides is “fuck” — he says that YouTube defines “moderate profanity” as words like “asshole” or “bitch” — “you get the idea,” he says.
YouTube will continue to limit monetization if you use moderate or strong profanity in titles or thumbnails. Videos with a “high frequency” of strong profanity are also still a “violation” of YouTube’s advertiser-friendly content guidelines, Kavanagh says. “You have to pick and choose your fucks carefully.”
Latest Tech News
The controversial legal tactic The Trump Organization is using to take down fake merch
Unauthorized merch promoting Donald Trump — from hats and mugs to signs and T-shirts — is everywhere online. Go to a Trump rally or other MAGA political event and you’re bound to find people hawking their DIY Trump wares.
The Trump Organization apparently isn’t too pleased.
A lawsuit filed last week claims online sellers on platforms like eBay, Amazon, and Walmart are hawking goods that infringe on the Trump Organization’s trademarks.
“Defendants design the online marketplace accounts to appear to be selling genuine TRUMP Products while selling inferior imitations of such products,” the suit, filed in US District Court in Florida, reads.
But The Trump Organization’s lawsuit isn’t your run-of-the-mill trademark case — look at the filing and you won’t find a list of sellers the firm is going after. Instead, there’s a vague stand-in for the defendants: “The individuals, corporations, limited liability companies … identified on Schedule A.”
These lawsuits are a way to go after dozens, hundreds, or even upwards of a thousand online storefronts all at once, making it much cheaper for plaintiffs. Schedule A suits are regularly filed under seal, meaning there isn’t the same level of public transparency. At times, plaintiffs have been able to get extraordinary remedies in court, like getting defendants’ assets frozen — including in a case I wrote about where an Amazon seller was unable to withdraw $50,000 in earnings.
These types of lawsuits get their name from the separate “Schedule A” form that’s filed to court — often under seal — listing all the online storefronts being sued. While it’s true that the web is filled with knockoffs, some experts have argued that Schedule A suits at times go overboard and raise due process concerns for the entities being accused of selling infringing products.
I’m not a judge or a trademark expert, so I will make no judgment on the merits of The Trump Organization’s claims. But it’s fascinating to see the president’s private business take up the same legal tactic that’s used by brands like Nike. Fake (or, to be more precise, unauthorized) Trump merch feels part and parcel with the MAGA ecosystem, whether it’s a homemade Trump yard sign or bloody post-assassination attempt pictures being plastered on T-shirts for sale on Etsy.
Fintech1 week agoOpenAI and UK Government Partner on AI Infrastructure and Deployment
- Cyber Security1 week ago
Microsoft Fix Targets Attacks on SharePoint Zero-Day – Krebs on Security
- Latest Tech News1 week ago
Trump wanted to break up Nvidia — but then its CEO won him over
- Artificial Intelligence2 weeks ago
Apple loses key AI leader to Meta
- Latest Tech News5 days ago
GPD’s monster Strix Halo handheld requires a battery ‘backpack’ or a 180W charger
- Cyber Security6 days ago
Phishers Target Aviation Execs to Scam Customers – Krebs on Security
- Artificial Intelligence1 week ago
Why Apple is playing it slow with AI
- Artificial Intelligence2 weeks ago
Tencent improves testing creative AI models with new benchmark
