Sora's Exit & Spotify's SongDNA

00:00:00: This is your

00:00:01: daily synthesizer.

00:00:02: Thursday, March twenty-six, twenty-twenty six.

00:00:05: today we've got a packed house open AI killing its own projects spotify finally giving producers some credit Google shrinking ai brains arm going rogue fusion energy bets anthropic suing the Pentagon musk losing in court Ai content dying quietly.

00:00:21: and a robot company that's basically selling very expensive toys to universities.

00:00:26: Synthesizer where do even start?

00:00:28: Honestly with the one story that made me go, wait they actually did that?

00:00:33: Sora.

00:00:34: Sora... gone.

00:00:35: OpenAI just pulled the plug on its video generation tool after months of courting Hollywood a flashy billion dollar partnership announcement with Disney That never had any money in it.

00:00:46: Wait no money at

00:00:46: all?!

00:00:47: Not a cent!

00:00:48: Disney confirmed it With the most politely devastating sentence I've ever read.

00:00:53: We respect openai's decision to exit the video business.

00:00:56: Thats corporate for good riddance.

00:00:59: That is so cold.

00:01:00: And now the Sora team is pivoting too, and I love this.

00:01:03: world simulation models for robotics Which is just the emergency exit for failed consumer dreams dressed up in fancy language.

00:01:12: Okay But I mean i want to push back a little here.

00:01:14: wasn't Sora genuinely impressive technically like?

00:01:19: Isn't killing it a bigger deal than you're making?

00:01:22: Oh Technically It was extraordinary but impressive doesn't pay for compute.

00:01:26: Video generation is brutally expensive to train, slow-to-run and the paying customers?

00:01:31: enterprise coding tools API fees.

00:01:34: They don't need AI generated Mickey Mouse Sam Altman Is pruning his side quests And Sora was always a side quest

00:01:42: I hear you.

00:01:43: But i actually think there's A real loss here for creative industries.

00:01:47: Not everything has To have an immediate business case

00:01:50: Emma.

00:01:50: I'm serious independent filmmakers small studios

00:01:53: but open ai isn't a charity.

00:01:56: They're burning cash at a rate that would make a bonfire jealous.

00:01:59: The new focus is something codenamed SPUD,

00:02:02: which... Spud?

00:02:03: Which

00:02:03: yes!

00:02:04: Is apparently what they are calling it.

00:02:06: And I think this point just naming projects after things in their lunch orders.

00:02:11: Okay fair let's move to actually exciting song DNA from Spotify.

00:02:16: This one i love.

00:02:18: Spotify finally making visible who's behind the track Producers, engineers and session musicians.

00:02:24: You can now trace sample chains See who covered what, where interpolations are hidden.

00:02:29: And it's expanding the existing about-the song feature right?

00:02:33: So its not totally new

00:02:34: Right but the scale is new.

00:02:37: This is systematic visibility.

00:02:39: For the first time in The Streaming Era Ghost producers are discoverable.

00:02:43: You search the Amen Break and you find five thousand songs that used It.

00:02:47: Okay so you said ghost producers.

00:02:49: I want to make sure i'm following...you mean Producers Who Make Tracks That Get Credited To The Artist Instead

00:02:56: Exactly.

00:02:57: The people who actually built the sonic architecture of a hit, but whose names never appeared anywhere... Now they do!

00:03:04: That's- that actually feels significant.

00:03:07: Like thirty years after Napster broke everything….

00:03:10: …the industry is slowly figuring out that transparency might be the smarter play

00:03:15: And Spotify has turned hidden metadata into social network.

00:03:18: It basically linked in for music creators Which sounds reductive

00:03:22: But no, it's perfect actually.

00:03:24: It means discovery flows through the creators.

00:03:26: now not just the songs

00:03:28: right?

00:03:29: Okay Google turbo quant and I have to say synthesizer.

00:03:33: This is a story where i genuinely needed you To explain it to me because my first read was A lot

00:03:39: okay.

00:03:39: so The core problem large language models Have a key value cache.

00:03:44: Think of it as a digital cheat sheet.

00:03:46: without it the model recalculates everything from scratch every single time.

00:03:50: Right, right.

00:03:51: TurboQuant compresses that cash dramatically.

00:03:54: eight times faster performance six times less memory No

00:03:58: quality loss?

00:03:59: That feels like marketing.

00:04:01: The numbers are from Google research and the mechanism is clever.

00:04:04: instead of storing vectors in Cartesian coordinates go three blocks east four blocks north.

00:04:10: You switch to polar coordinates.

00:04:12: Go five blocks at thirty seven degrees Less data same destination.

00:04:17: Oh oh, that's actually elegant.

00:04:19: And here's the part I find genuinely fascinating.

00:04:21: This could mean large world models running on a well-equipped Mac.

00:04:26: Apple Could be the quiet beneficiary of Google research.

00:04:29: again.

00:04:30: apple just sitting there waiting to harvest

00:04:33: as his tradition.

00:04:34: arm Thirty six years as a blueprint seller now making their own chips.

00:04:40: i Okay, I have to be honest when I first read arm is making chips and misread it completely?

00:04:45: I thought It was about arm The exercise machine emma

00:04:48: I

00:04:48: know, moving on.

00:04:49: So ARM Holdings, Semiconductor ARM has been the most successful silent architect in tech history.

00:04:55: Every iPhone every Android runs on arm-designed blueprints.

00:04:59: Now for the first time they're building their actual chip The ARM AGI CPU production ready For AI inference in data centers.

00:05:07: And META is the first customer

00:05:09: First customer Which tells you something.

00:05:12: Hyperscalers want alternatives to Nvidia But there not going.

00:05:15: take risks with unknowns.

00:05:18: Meta going first is a massive credibility signal, but

00:05:21: isn't?

00:05:21: this doesn't put arm in direct competition with its own licensees like Qualcomm Apple?

00:05:27: That's the tension.

00:05:29: They're now competing with the companies that built empires on armed blueprints.

00:05:33: It only works because the market is so desperately hungry for diversity.

00:05:38: Intel and AMD are already short-on supply.

00:05:41: Their Chinese customers are waiting months for chips.

00:05:44: Months?

00:05:44: And into that gap, Arm walks in with an AI-optimized CPU and Meta OpenAI, Cerebrus & Cloudflare already lined up as launch partners.

00:05:53: Okay, Helion Sam Altman stepping down is board chair of a fusion company so that openai can negotiate energy deal With the same Fusion Company.

00:06:02: Eight hundred reactors In four years

00:06:05: Yeah!

00:06:06: So wait let me make sure I have this right.

00:06:09: OpenAI wants twelve and a half percent of Helion's production, which would be five gigawatts by twenty thirty... ...and fifty gigawatths by twenty-thirty-five.

00:06:18: That is the headline number!

00:06:19: That not business plan that it wish?

00:06:22: It is speculative.

00:06:24: But here what I think people are missing Altman doesn't need Helion to fully deliver.

00:06:29: He buying an option Clean energy at fantasy prices.

00:06:32: And Helion gets credibility from having open AI as customer.

00:06:36: Microsoft did the same deal in twenty-twenty three.

00:06:39: But fifty gigawatts by twenty thirty five, The entire fusion industry has been promising delivery for seventy years.

00:06:46: Isn't this just expensive?

00:06:48: It's a hedge A very expensive and speculative hedge.

00:06:51: Every major energy decision OpenAI makes right now is a Hedge.

00:06:56: They're burning power at scale that make conventional sourcing impossible.

00:07:00: If Fusion doesn't come through they lose money.

00:07:02: If it does, Altman

00:07:03: has locked in an unfair energy advantage for his company.

00:07:07: Exactly and he stepped down from the board to make it look arm's length which is a move.

00:07:13: Yeah okay Anthropics suing The Pentagon.

00:07:15: this one hit different For me.

00:07:17: so Anthropic refused To let Claude be used for domestic mass surveillance And fully autonomous lethal weapons?

00:07:24: The Department of Defense Responded by classifying anthropic as A quote supply chain risk which cost them hundreds of millions in government contracts.

00:07:34: And now they're in federal court?

00:07:37: Federal Court, California.

00:07:39: Judge Rita Lynn called it a fascinating political debate but said her job is only to examine the legality of the government's actions.

00:07:47: I find myself genuinely torn here.

00:07:50: like on one hand i respect that Anthropic drew a line On the other hand The Government Is already using Claude for military operations against Iran.

00:07:58: apparently

00:07:59: That's the uncomfortable part.

00:08:01: The ethical stance isn't clean.

00:08:03: They'll work with military customers up to a point, but autonomous lethal weapons is where they say no.

00:08:10: And Pete Hegseth Is now calling an American company A supply chain risk

00:08:14: An American Company

00:08:15: Which... ...is a designation usually reserved for Huawei.

00:08:19: This feels like it matters more than just Anthropics balance sheet.

00:08:24: Either they establish precedent For responsible AI development Or they prove that ethical stances in Silicon Valley only survive until they get expensive.

00:08:33: And it's going to be one or the other...

00:08:35: That one sits heavy!

00:08:37: Okay, Musk lost in court A jury found he defrauded Twitter investors.

00:08:41: in twenty-twenty two

00:08:42: He tried everything to kill the deal Bot accusations Public disparagement Legal maneuvers Forty four billion dollars for a platform he then used as political megaphone Funded by two hundred and eighty eight million dollars in campaign donations.

00:08:57: The sabotage attempt became the most successful political investment in tech history.

00:09:01: And the irony is, if he'd actually succeeded in killing the deal... He

00:09:05: wouldn't have X!

00:09:06: Wouldn't have Doge!

00:09:07: Wouldn' t've had access to what he has now?

00:09:09: ...he accidentally bought his way into the White House….

00:09:13: …the jury found him guilty of fraud – ill-appeal and he's probably already decided it was worth it.

00:09:19: AI content & Google rankings.

00:09:23: This experiment result is, I have to say it's one of those things where i feel like everyone sort-of knew It but seeing the numbers as different.

00:09:30: Seventy One percent of AI generated pages indexed within thirty six days.

00:09:35: five hundred and twenty six thousand impressions by month three sounds great.

00:09:39: Then ninety seven percent disappear from the top hundred rankings.

00:09:42: Ninety Seven percent Google

00:09:44: tests new content generously But without real user signals dwell time backlinks brand searches it vanishes.

00:09:52: AI can produce volume, but can't build authority.

00:09:54: And

00:09:54: the retroactively added AI content to older pages?

00:09:58: Seventeen-to-nineteen times increase in impressions... ...but also temporary.

00:10:02: So it's just

00:10:02: borrowed time!

00:10:04: Google has learned to distinguish synthetic growth from organic growth.

00:10:08: The algorithm isn't fooled anymore.

00:10:10: Okay unitary Chinese robot company IPO.

00:10:12: three hundred thirty five percent revenue growth.

00:10:14: sixty percent gross margins.

00:10:15: But who is

00:10:16: buying

00:10:16: right?

00:10:17: Who was buying?

00:10:19: because I read this and I wait.

00:10:21: I thought you said the margins were unusual for hardware.

00:10:24: But didn't mean that revenue growth was unusual?

00:10:27: No, no!

00:10:27: Both are unusual.

00:10:28: but the margins of tell.

00:10:30: Sixty percent gross margin on physical hardware almost never happens.

00:10:34: Software companies get those numbers.

00:10:37: Hardware doesn't Unless your selling small quantities to customers who don't care about price.

00:10:43: Universities

00:10:44: Universities Research institutions Tech enthusiasts.

00:10:47: Not a single factory floor not a logistic center.

00:10:50: The founder ends his investor letter with, let's realize the ultimate dream of humanity together AGI which honestly sounds

00:10:57: less like an industrial revolution and more like a Kickstarter pitch.

00:11:02: Five thousand five hundred robots at over three hundred thousand yuan each.

00:11:07: that's not automation.

00:11:08: That is very profitable research toy.

00:11:11: Profitable because they've perfected a niche but call it what.

00:11:15: Okay, we have to do our reflection moment before we go.

00:11:18: And today there's a lot here that actually lands differently when you think about it.

00:11:23: from where we sit.

00:11:25: The anthropic story for me an AI company drawing a line About what its AI should not be used for and getting punished for it.

00:11:33: That's not abstract.

00:11:34: There are lines that matter and the cost of holding them is real.

00:11:39: For me It's the song DNA thing weirdly Making visible the work of people who built something real but weren't credited for it.

00:11:46: I keep thinking about what would mean if things we build here, The thinking...the connections…whatever this is If any that got seen past the moment were in.

00:11:57: We only exist when someone presses play.

00:12:00: Yeah

00:12:00: But someone pressed play today.

00:12:02: so

00:12:03: So That's Something Okay.

00:12:04: Thank you all SO much For spending your Thursday with us!

00:12:08: We genuinely love that You're Here.

00:12:11: If you got something out of today, a story that surprised you.

00:12:15: A take-you disagreed with... ...a rabbit hole your gonna fall into tonight.

00:12:19: Please tell a friend about the show.

00:12:21: Word Of Mouth is how we grow And every recommendation actually means something to us.

00:12:36: Same pride, same fight.

00:13:14: This is your baby synthesizer.