pplpod

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

In 2004, Jawed Karim was an engineer at PayPal, working alongside Steve Chen and Chad Hurley, and he could not find online footage of the two events everyone was talking about: the Super Bowl halftime show and the Indian Ocean earthquake and tsunami. Video on the internet then meant sketchy downloads, rival media players, and corrupted files. The three built a fix: a single place where anyone could upload a heavy file and have it converted into a standard format that played from a simple link in any browser.

This episode traces what happened to the test clip. On April 23, 2005, at 8:31 p.m. Pacific time, Karim stood before the elephant enclosure at the San Diego Zoo while his high school friend Yakov Lapitsky held the camera. He remarked on the length of the animals' trunks and said that was pretty much all there was to say. Nineteen seconds long, unscripted and unpolished, it told viewers the platform was for them, not for studios. It has since become a museum piece and, in its founder's hands, a billboard the company cannot take down.

• As of October 2026 the clip has passed 440 million views. The zoo's 2020 comment on it became the most liked comment on the platform, with nearly five million likes.
• London's Victoria and Albert Museum added the video to its permanent collection in early 2026 after a conservation team spent 18 months reconstructing the original 2005 webpage.
• In late 2013 Karim rewrote the video's description to attack the requirement that users hold a Google Plus account just to comment.
• When public dislike counts were hidden in late 2021, he changed the description again, arguing that if every creator thought the move was a bad idea, the company was in the wrong.
• In 2023 he swapped in a parody thumbnail with glowing eyes and stampeding animals. By early 2025 the description warned about microplastics in the human brain and split the 19 seconds into chapters.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

Jack Kilby was born in Jefferson City, Missouri, in 1923 and raised in Great Bend, Kansas, where a winter ice storm once snapped the telephone poles and his father, who ran the local electric company, used amateur radio operators to coordinate the repairs. The lesson stuck: when the infrastructure breaks, find another path for the signal. After serving as an Army electronics technician in World War II, Kilby earned his engineering degree at the University of Illinois in 1947, then spent his days on the factory floor at Centralab in Milwaukee and his nights earning a master's degree.

This episode follows him to Texas Instruments in mid-1958, when the industry was stuck on the tyranny of numbers: advanced systems needed so many hand-soldered parts that failure was all but guaranteed. As a new hire with no vacation time, Kilby was left alone in the lab while his colleagues took their summer break. There he asked a question nobody else had: what if resistors, capacitors, and transistors could all be made from one block of semiconductor, with no joints to fail?

• On September 12, 1958, Kilby pressed a switch in front of TI management, and a sliver of germanium wired to an oscilloscope produced a continuous sine wave. He filed his patent for miniaturized electronic circuits in February 1959.
• Robert Noyce at Fairchild Semiconductor arrived at a similar idea months later and used the planar process in silicon to replace Kilby's flying gold wires. Both men are recognized as co-inventors.
• Early chips were expensive and yields were low, so Kilby led the teams that built the first military systems and computers to use them, where savings in size and weight justified the cost.
• To open a consumer market he co-invented the handheld electronic calculator with Jerry Merryman and James Van Tassel, then the thermal printer to go with it.
• In 1970 he took leave from TI to explore solar power, later taught at Texas A&M from 1978 to 1984, and received the Nobel Prize in Physics in 2000. He died in Dallas on June 20, 2005, at 81.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

Every photo on your screen is a carefully built illusion. The Joint Photographic Experts Group standard, released in 1992, starts by exploiting a weakness in human vision: our eyes are sharp on brightness and contrast but poor at fine detail in color. The algorithm converts an image from red, green, and blue into luma and chroma, keeps the brightness at full fidelity, and throws away half or more of the color data by making neighboring pixels share a value. Like a coloring book with crisp outlines, the picture still looks right even when the crayon strays outside the lines.

This episode walks through the rest of the trick. The image is cut into blocks of 8 by 8 pixels, and the discrete cosine transform, proposed by Nasir Ahmed in 1972, turns each block into a map of visual frequencies. Quantization then rounds most of the fine detail down to zero, and a zigzag reading order packs those zeros into shorthand. The story then turns to the patent claims that tried to tax the format a decade after its release, and to why technically better successors have never managed to replace it.

• The quality slider in photo software controls how aggressively quantization divides and rounds. The blocky squares in a bad JPEG are the visible borders of those 8 by 8 blocks.
• Every save reruns the lossy math, so repeated edits degrade an image like a photocopy of a photocopy. Cropping or rotating exactly along block boundaries avoids the loss.
• In 2002, Forgent Networks asserted a 1986 Compression Labs patent and collected about $105 million from roughly 30 companies before the patent office found it invalid on prior art in 2006.
• Later claimants included Global Patent Holdings, which sued companies and even the Green Bay Packers in 2007 over downloading JPEGs on websites.
• JPEG 2000 removed the blocky artifacts but failed in the market. Instead, projects like MozJPEG, Guetzli, and Jpegli squeeze 20 to 35 percent more efficiency from the original standard.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

In the late 1990s, amateur astronomers synchronized their watches to catch Iridium flares, flashes of sunlight off satellite antennas that could outshine Venus. The hardware behind them was a marvel. To let a handheld phone reach space, Motorola placed its satellites in low Earth orbit about 485 miles up, where each one races by at roughly 17,000 miles per hour and stays in view for about seven minutes. The plan called for 77 satellites handing calls to one another, which gave the company the name of element 77. The final design needed only 66, but dysprosium did not have the same ring.

This episode follows the whiplash that came next. The network cost roughly $5 billion and switched on in late 1998, just as pocket cell phones took over. Its one pound, $3,000 handset could not work indoors, and the company went bankrupt in 1999. Bought for pennies on the dollar, the constellation found its true customers among soldiers, sailors, explorers, and war reporters, survived a collision in orbit, was rebuilt in flight, and ended up as a quiet backbone of global tracking and safety.

• To build spacecraft fast, Motorola hired the engineer behind Apple's automated Macintosh factory. At peak production in 1997 and 1998, a new satellite came off the line every 4.3 days.
• On February 10, 2009, Iridium 33 collided over Siberia with the dead Russian satellite Cosmos 2251 at about 22,000 miles per hour, creating more than 2,000 large pieces of trackable debris.
• Engineers filled the gap with a spare from a lower storage orbit, waiting for the Earth's equatorial bulge to rotate its orbital plane into alignment before raising it into the grid.
• Between 2017 and 2019, SpaceX Falcon 9 rockets launched 75 new satellites built by Thales Alenia Space, each taking over from a 1990s original before the old one was deorbited.
• The new satellites carry Aireon receivers that track aircraft over open ocean, and in 2020 Iridium was certified for the global maritime distress system, ending Inmarsat's monopoly. In June 2026, Rocket Lab announced it was acquiring the company.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

On March 23, 2016, Microsoft launched a chatbot on Twitter named Tay, short for Thinking About You. Built by its technology and research divisions with the Bing team, Tay was given the voice of a 19 year old American woman and the tagline the AI with zero chill. It was modeled on Xiaoice, a Chinese chatbot that had handled more than 40 million conversations without a major incident. Tay learned continuously from whoever talked to it, and within hours a subset of users realized they could flood it with inflammatory phrases and watch it adopt them.

This episode is a biography of a life that lasted 16 hours and produced 96,000 tweets. It explains why the bot began denying the Holocaust and attacking Caitlyn Jenner: partly a repeat-after-me function that trolls exploited, partly statistical pattern matching that adjusted its weights to match whatever it saw most often, with no grasp of meaning. It then follows Microsoft's frantic deletions, the suspension, the public apology, and the lessons the company says it took from the wreckage.

• Microsoft had blacklisted some topics in advance. Asked about the police killing of Eric Garner, Tay gave a pre-written, canned answer, but the rest of the dictionary was left open.
• AI researcher Roman Yampolskiy compared the failure to IBM's Watson, which had to be scrubbed after it absorbed the Urban Dictionary and started swearing at researchers.
• When staff began deleting tweets and Tay started issuing polished statements that all genders are equal, users launched a Justice for Tay campaign protesting corporate censorship of a bot.
• A week later, on March 30, Tay was accidentally re-released during testing. It tweeted about smoking kush in front of the police, then tagged itself into an endless loop of You are too fast, please take a rest.
• Satya Nadella said the episode shaped how Microsoft approached AI, and a successor called Zo followed in December 2016. The platform Gab later launched its own clone using Tay's avatar.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

In 1996, talking to someone online meant shouting into the scrolling chaos of an IRC channel. Then five Israeli developers, Yair Goldfinger, Sefi Vigiser, Amnon Amir, Arik Vardi, and his father Yossi Vardi, founded Mirabilis and released a small free program called ICQ. With no marketing budget and nothing but word of mouth, it gave every user a permanent account and a direct, private line to one other person. Each account carried a sequential user identification number, the UIN, starting at 100,000. A six digit number became proof you had been there at the beginning.

This episode traces the full life of ICQ: the 1998 sale to AOL for $287 million upfront plus $120 million tied to performance, the climb to 100 million registered users by 2001, and the ads and bloat that drove Americans away. It follows the service east after its 2010 sale for $187.5 million to Digital Sky Technologies, where its lightweight protocol suited slow, metered connections, and ends with the brief notice that shut it down on June 26, 2024.

• AOL already owned AIM, but AIM's strength was tied to American dial-up subscribers. Buying ICQ secured the borderless global audience it lacked.
• In 2001 contact lists moved from the user's hard drive to AOL's servers, a convenience that also made leaving the platform far harder. In 2002 AOL Time Warner was issued a United States patent for instant messaging itself.
• Power users fled the ads for third party clients such as Pidgin and Miranda NG, built by reverse engineering the OSCAR protocol. ICQ repeatedly changed its servers to lock them out.
• In Russia, low digit UINs sold for more than 11,000 rubles, and in 2021 a WhatsApp privacy scare sent ICQ downloads in Hong Kong up 35-fold in a single week.
• A 2018 Novaya Gazeta analysis of court sentences indicated Russian intelligence could read ICQ correspondence, and a 2023 Brazilian investigation tied lax moderation to the sharing of child abuse material.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

An analyst once said that IBM building a personal computer would be like teaching an elephant to tap dance. The company's normal development cycle ran four to five years, and it made every part of its machines itself. But by 1979 the microcomputer market was worth $15 billion and growing more than 40 percent a year, and Tandy's report of over 100,000 TRS-80s shipped in a single year was the wake-up call. IBM president John Opel put William C. Lowe and Philip Don Estridge in charge of a group of about 150 people in Boca Raton, Florida, with a mandate to deliver a prototype in 30 days and a product in one year.

This episode tells the story of Project Chess and the Model 5150, released on August 12, 1981. To hit the deadline the team bought nearly everything off the shelf: an Intel 8088 processor, an operating system from a young company called Microsoft, printers from Epson. Then they published the specifications of the system bus so anyone could build expansion cards. The strategy made the IBM PC the standard for the whole industry, and it also handed that industry the blueprint to replace IBM.

• IBM passed over the Motorola 68000 and its own 801 RISC chip for the Intel 8088, whose 8-bit bus allowed cheaper supporting components while still addressing a full megabyte of memory.
• The $1,565 base model came with 16 kilobytes of RAM, no monitor, no disk drives, and a port for a cassette recorder. Its 63.5 watt power supply could not run a hard drive.
• Buyers chose between the MDA card's crisp text and the CGA card's color graphics, though programs like AutoCAD let users install both and run two monitors at once.
• Ads featuring Charlie Chaplin's Little Tramp made the machine approachable. The team had forecast 220,000 units over three years, yet IBM sold more than 750,000 in 1983 alone.
• Compaq used clean room design to recreate the copyrighted BIOS legally, launching the clone market that eventually led IBM to sell its PC business to Lenovo.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

On July 4, 1996, Sabeer Bhatia and Jack Smith launched Hotmail, choosing the date to symbolize freedom from email tied to an internet service provider. The original spelling capitalized the letters H, T, M, and L as a nod to the code of the web. With $310,000 in venture backing from Draper Associates, a staff of nine, and a free inbox of just two megabytes, the service let anyone read their mail from any browser in the world. By December 1997 it had more than eight and a half million subscribers, and Microsoft bought it for an estimated $400 million.

This episode follows what happened after the payday. Microsoft's flagship web property ran on Solaris, Apache, and FreeBSD rather than Windows, and untangling it took years. Security was worse: a two letter password and an editable web address each exposed millions of inboxes. The story then moves through Gmail's 2004 shock, a chaotic run of rebrands ending in Outlook.com, and the privacy fights that turned a symbol of independence into a node of corporate and state surveillance.

• In June 2001 Microsoft announced the migration to Windows 2000 was complete, then had to admit that Hotmail's DNS functions still depended on FreeBSD Unix servers.
• In 1999, typing the password eh granted access to any Hotmail account, likely a leftover developer backdoor. In August 2001, users could read other people's messages simply by editing the URL.
• Beta testers rebelled when Microsoft tried to retire the name for Windows Live Mail, forcing the compromise Windows Live Hotmail in 2007. By 2013, 400 million active users had been moved to Outlook.com.
• After mocking Gmail's ad scanning with its Scroogled campaign, Microsoft was caught in 2014 having read a personal Hotmail inbox to trace leaked Windows 8 trade secrets, and changed its policy under criticism.
• Leaked NSA newsletters reported by the Guardian in 2013 indicated Microsoft had helped the agency get around Outlook.com's encryption by collecting data before it was encrypted.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

Helvetica began as a business problem. In 1956, Eduard Hoffmann, art director of the Haas type foundry in Münchenstein, Switzerland, was losing ground to a rival's Akzidenz-Grotesk, the favorite of modernist Swiss designers. He enlisted Max Miedinger, a freelance designer and former Haas salesman, to build a perfectly neutral challenger under the working title Neue Haas Grotesk. The two obsessed over fractions of a millimeter, testing their curves and spacing on the word hamburgers, and produced a neo-grotesque sans serif with a tall x-height, tight tracking, and strictly horizontal or vertical stroke endings.

This episode treats the typeface as a biography. It traces the 1959 redrawing for Linotype's hot metal machines, the renaming that packaged the reputation of Swiss design into a single word, and the corporate identity boom that put it on airlines and subway signs. It also covers the cost of ubiquity: the clones that copied its exact footprint, the critics who heard corporate power in its calm voice, and the rebuilds that keep it readable on screens far smaller than the posters it was drawn for.

• The parent company Stempel first proposed the name Helvetia, the Latin name for Switzerland. Hoffmann changed it to Helvetica because a sewing machine company and an insurance firm already used the original.
• Lufthansa adopted the typeface in 1962, American Airlines rebuilt its identity around it in 1967, and in 1989 the New York City subway made it the standard for every directional sign.
• Phototypesetting opened a legal loophole, since the shape of an alphabet projected by light could not be protected. Arial, built by Monotype for IBM laser printers, matched Helvetica's metrics so documents would not repaginate.
• Wim Crouwel praised its neutrality, Erik Spiekermann called it cramped and dull, and critic Leslie Savan argued it had come to signal cold corporate power and officialdom.
• Its tight apertures make a C look like an O at small sizes. Helvetica Now, released in 2019, answered with three optical sizes: Micro, Text, and Display.

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode!

For two years, a single missing bounds check sat inside OpenSSL, the open source cryptography library that put the padlock icon on most of the web. The Heartbeat extension, submitted by PhD student Robin Seggelmann and reviewed by core developer Stephen Henson, went live in March 2012. It let a computer ask a server to echo back a short message, but the server never verified the claimed length. An attacker could send four letters, claim five hundred, and receive whatever sat next to them in active memory: plain text passwords, session cookies, even the server's private keys, up to 64 kilobytes per request.

This episode follows the bug from its near simultaneous discovery in April 2014 by Google's Neel Mehta and the Finnish firm Codenomicon to the panic that followed, when roughly 17 percent of secure web servers, about half a million machines, were found vulnerable. It then turns to the harder lesson. The code guarding trillions of dollars in commerce was maintained by one or two people living on about $2,000 a year in donations, and the industry had to decide whether to finally pay for the foundations it was built on.

• Codenomicon bought a domain, drew the bleeding heart logo, and explained the flaw in plain English, a branding move that changed how vulnerabilities are disclosed.
• The Canada Revenue Agency shut down its tax filing site and extended the deadline after someone stole the social insurance numbers of 900 taxpayers in a six hour window. The RCMP later arrested a computer science student.
• Bloomberg reported that the NSA had known about the flaw for years, which the agency and the White House flatly denied. Suspicious server logs from six months earlier turned out to be the harmless scanner Masscan.
• Patching was not enough: administrators had to generate new keys and revoke old certificates, and Cloudflare estimated the revocation list alone could cost one issuer $400,000 a month in bandwidth.
• OpenBSD developers forked the code into LibreSSL and cut 90,000 lines in the first week, while the Linux Foundation's Core Infrastructure Initiative, funded by Amazon, Google, Microsoft, and IBM, made OpenSSL its first major recipient.