0:00 New court filings show OpenAI executives knew they were engaging in mass book piracy to train their models, and were specifically worried about the “optics” of it on Hacker News.0:11 Last week we talked about AI gone wrong, but this...0:14 this is different.0:15 This isn't an algorithm making a mistake; this is an alleged strategy session on how to get away with theft, with the receipts to prove it.0:24 So, let's get into the headlines everyone's talking about.0:28 First, the other big platform drama.0:30 Just two weeks before the election in Brazil, Meta temporarily blocked President Lula’s Facebook page and his campaign ads.0:38 The page came back online the same day, but the damage was done.0:42 The incident immediately blew up into a massive debate about censorship, foreign interference, and the sheer, unchecked power these platforms have over national politics.0:53 The official line is that it could have been an automated error, but when an "error" can silence a world leader during a critical election window...1:03 it starts to look a lot more like a feature than a bug.1:06 It’s a stark reminder that the moderation policies of a company in California can have profound, real-world consequences for a sovereign nation's democracy.1:17 Then we have a story that is pure, classic Hacker News.1:20 Ken Shirriff, who you probably know for his incredible deep dives into old hardware, has done it again.1:27 He reverse-engineered the floating-point coprocessor from 1980, the Intel 8087.1:32 And he figured out exactly how it calculated the tangent function so ridiculously fast.1:37 It turns out it was using a clever combination of an algorithm called CORDIC and polynomial approximations.1:44 The result?1:45 It could compute a tangent in ninety microseconds.1:48 The main 8086 CPU, by itself, would take thirteen THOUSAND microseconds to do the same thing.1:54 It’s just this beautiful piece of micro-archaeology, peeling back the silicon to find the ingenuity that powered the first generation of PCs.2:03 Speaking of deep dives, here’s one that will make you question every statistic you’ve ever read.2:09 A blogger who goes by TheBorys on Substack was just idly looking at maps and realized something was off about the size of Yemen.2:18 The officially cited number everywhere — Wikipedia, the CIA World Factbook, everywhere — is 555,000 square kilometers.2:25 But it just didn't look that big.2:27 So he did the work.2:29 He meticulously traced the borders in Google Maps, included all the islands like Socotra, and did the math.2:35 The actual size?2:36 Closer to 456,000 square kilometers.2:39 That's a nearly one-hundred-thousand square kilometer error.2:43 And he traced the bad number all the way back to a statistical yearbook from 1991.2:48 A typo, or maybe a bit of national pride, has been propagating as fact for over thirty years.2:54 It’s a perfect parable for the internet age: a single, determined person can uncover a truth that massive institutions have missed for decades.3:03 And on the developer front, a couple of new tools for drawing things with code got a ton of attention.3:10 The first is called Reladraw.3:12 If you've ever used a tool like Mermaid or Graphviz, you know the frustration.3:17 You write the code for your diagram, and the tool just...3:20 puts things wherever it wants.3:22 Reladraw’s creator wanted the best of both worlds: define the diagram in code, but still have manual control over the placement of every single element.3:32 It’s for people who want their diagrams to be both reproducible AND look good.3:37 The other one is called Drawgent.3:39 This one is a bit more futuristic.3:41 It's a coding agent that's integrated with a live Excalidraw canvas — that's a popular online whiteboard tool.3:48 So you can literally draw a diagram, and the agent, which supports Rust and JavaScript, will help you code it in real time.3:56 It's this wild, collaborative space where the visual diagram and the underlying code are constantly in sync.4:03 Finally, in the world of AI research, two papers made a splash.4:07 One is a proposal from DeepSeek for something they're calling "Elastic Compute." It's basically a new kind of sandbox infrastructure designed to make it easier to train those big, complex "agentic" AI systems at scale.4:21 It’s all about solving the gnarly distributed computing problems that pop up when you're trying to get thousands of AI agents to work together.4:30 The other paper was a clever hack.4:33 Researchers took a large language model, GLM-5.3-Flash, and figured out how to make it act like a much smaller, faster "decision model" just by crafting the right prompts.4:44 They got it to produce its final decision in a single token, which makes it super fast, and it even works with vision input.4:52 It’s another example of how much untapped potential is still hiding inside these existing models, if you just know how to ask the right questions.5:01 So, that’s the sweep.5:02 Now, let’s go deeper.5:04 Okay, let's really dig into the OpenAI story, because it’s so much more than just another lawsuit.5:10 The Authors Guild, representing writers like George R.5:13 R.5:14 Martin and Jonathan Franzen, dropped new filings in their case against OpenAI and Microsoft.5:20 And what they contain is...5:21 damning.5:22 The filings allege that OpenAI executives knew, in their own words, that their practice of scraping massive datasets of books to train GPT was illegal.5:32 They were fully aware they were using copyrighted material without permission.5:37 And here’s the part that really gets me.5:39 They were worried about the "optics." Specifically, they were worried about how it would look on Hacker News.5:46 It’s this incredible moment of self-awareness.5:49 They knew what they were doing was wrong, and their primary concern wasn't the ethics or the legality, but the potential for bad PR within the very tech community they belong to.6:01 The CEO of the Authors Guild, Mary Rasenberger, called it a "shocking disdain for writers." And it's hard to argue with that when you see the evidence laid out.6:11 So, where have we seen this before?6:13 The pattern is almost a cliché at this point.6:16 It’s Napster.6:17 It’s the early days of digital music.6:19 A new technology emerges that can do something revolutionary, but its power is built on the unauthorized use of existing creative work.6:28 In the late nineties, it was MP3 compression and peer-to-peer file sharing.6:33 The music industry was completely unprepared.6:36 Suddenly, their entire catalog was available for free, and their business model was shattered overnight.6:42 The tech guys, led by people like Sean Parker, had this ethos of "information wants to be free." They saw themselves as liberators, breaking down the old gatekeepers.6:53 The parallels with OpenAI are obvious.6:55 You have a disruptive technology — generative AI.6:59 You have a legacy industry — publishing.7:01 And you have the tech company deciding that its mission is so important that the old rules, like copyright, don't really apply.7:09 They'll just take the data, build the future, and sort out the legal mess later.7:14 It's the "move fast and break things" philosophy applied to the entire corpus of human literature.7:21 But here’s where the analogy starts to break down, and where this gets even more complicated.7:27 Napster was about distribution.7:29 It took a song and made it available to more people.7:32 It didn't change the song itself.7:34 OpenAI is doing something fundamentally different.7:37 It’s not distributing the books; it's ingesting them.7:41 It’s grinding them down into statistical patterns to create something entirely new.7:46 The lawsuit even calls GPT an "existential threat to those who write and publish books." And that’s not hyperbole.7:53 If an AI can generate a novel in the style of any author, what does that do to the profession of writing?8:00 This isn't just a business dispute; it's a philosophical one.8:04 What does it mean to "read" a book when you're a machine?8:08 Does training an AI on a copyrighted work constitute "fair use"?8:12 The Napster case was ultimately pretty straightforward: you can't give away someone else's song for free.8:19 This is...8:19 murkier.8:20 OpenAI will argue that training is a transformative use, that they aren't reselling the books, but are using the knowledge within them to build a tool for human progress.8:31 The authors argue that their work IS the product.8:34 You can't build a model that writes like George R.8:37 R.8:37 Martin without feeding it George R.8:40 R.8:40 Martin's books.8:41 And that's the core of it.8:42 The tech industry, for twenty years, has operated on this assumption that data is just a raw material, like oil or iron ore, free for the taking.8:52 This lawsuit is the moment where the people who create that data are standing up and saying, "No.8:58 That's not data.8:59 That's our life's work.9:00 And you have to pay for it." Now let's switch gears to the other big story: Meta blocking Brazilian President Lula's Facebook page.9:09 On the surface, it seems like a straightforward case of a platform messing up.9:14 The page was taken down, there was an outcry, and it was restored the same day.9:19 Meta can just say "oops, our automated systems made a mistake," and move on.9:24 But that’s exactly the problem.9:26 The "where have we seen this before" here isn't just about tech.9:30 It’s about media power.9:31 For over a century, the kingmakers in politics were newspaper barons.9:36 Think William Randolph Hearst in the United States.9:39 He could use his chain of newspapers to start a war, to elect a president, or to destroy a political career.9:46 The power was immense, but it was also, in a way, transparent.9:50 You knew who owned the paper.9:52 You could read the editorials.9:54 The bias was out in the open.9:56 What Meta and other social media giants have created is a new kind of media power that is vastly larger in scale, global in reach, and almost completely opaque.10:06 When Hearst wanted to attack a candidate, an editor had to write a headline.10:11 A press had to run.10:12 Trucks had to deliver the papers.10:14 When Meta's algorithm decides a post violates its policy, a world leader can be silenced instantly, across the globe, with no human intervention.10:24 And the defense that "it was a mistake" is where the analogy to old media completely falls apart.10:30 Could you imagine the editor of the New York Times accidentally running a headline saying "PRESIDENT INDICTED" and then just saying, "oh, sorry, typo"?10:40 Of course not.10:40 There are layers of accountability.10:43 But with platforms, the scale is so massive that "automated errors" are accepted as a cost of doing business.10:50 But an error that silences a presidential candidate two weeks before an election isn't just a technical glitch.10:57 It's a political event.10:58 As one commenter on Hacker News pointed out, this kind of "election interference" has been happening in Europe for years, where it often seems like parties outside the establishment are held to a much stricter standard.11:12 It creates a "two-tier system." So what's the real issue here?11:16 It's not just about whether Meta is biased against Lula.11:20 It's about the fact that a private company, with its own commercial interests and a set of rules written in secret, has the power to make these kinds of decisions at all.11:31 The call by some to just ban all political ads is a tempting one, but it's a blunt instrument.11:37 What counts as "political"?11:39 Is a post about climate change political?11:41 Is a post about healthcare?11:43 The real problem is the concentration of power in a system that lacks any real public accountability.11:50 The decision to block Lula's page, whether by a human or an algorithm, had a real impact.11:55 It injected chaos and suspicion into an already tense election.11:59 And the fact that it was reversed so quickly doesn't fix the problem.12:04 It just highlights how arbitrary the power is.12:07 One moment you're silenced, the next you're back.12:10 Who made the call?12:11 Why?12:11 We'll probably never know.12:13 And that opacity, that lack of a clear answer, is more corrosive to democracy than any single bad decision.12:20 So, what does all this add up to?12:22 You have OpenAI's secret book piracy, Meta's opaque political moderation, a thirty-year-old error in a country's official size, a hidden algorithm inside a vintage chip.12:33 The common thread this week is about what happens when the black box gets pried open.12:38 Sometimes it's a court filing that reveals a company's cynical internal monologue.12:44 Sometimes it’s a dedicated engineer with a microscope revealing the secrets of old silicon.12:49 And sometimes it's just one curious person with Google Maps who refuses to believe the number everyone else takes for granted.12:58 This week wasn't just about new tools or corporate drama.13:01 It was about the fight for transparency, and the surprising things you find when you actually look under the hood.13:09 What this all sets up is a future where "trust, but verify" becomes the default stance for everything.13:15 We're moving into an era where we're going to demand to see the receipts — on the data used to train our AI, on the rules that govern our public square, and maybe even on the size of a country.13:28 The age of the unexamined black box is ending.
