Stewart Squared
Episode #70: From Twitter to Threads: Escaping the Training Data Mines of Late Capitalism
- Money & Sovereignty
- Geopolitics & Power
- datatraining
- techplatforms
- socialmedia
- semiconductors
- aiinfrastructure
- chipmanufacturing
- foundry
- supercomputers
Listen
Download MP3About this episode
In this episode of the podcast, host Stewart Alsop III engages in a wide-ranging conversation with Stewart Alsop II about data training, social media competition between X and Threads, and the broader technological landscape from semiconductors to AI. The discussion covers everything from Taiwan's dominance in chip manufacturing through TSMC, the evolution of supercomputers from Seymour Cray's innovations to modern GPU clusters, and the challenges facing early-stage companies trying to scale specialized technologies like advanced materials for semiconductor manufacturing. The conversation also touches on the complexities of cryptocurrency adoption, the changing nature of work in an increasingly specialized economy, and the implications of AI data centers on power consumption and infrastructure.
Timestamps
00:00 The Rise of Threads and Competition with X
03:01 The Semiconductor Landscape: TSMC vs. Intel
06:03 The Role of Supercomputers in Modern Science
09:00 AI and the Future of Data Centers
11:46 The Evolution of Computing: From Mainframes to Clusters
14:54 The Impact of Moore's Law on Semiconductor Technology
17:52 Heat Management in High-Performance Computing
31:01 Power and Cooling Challenges in AI Data Centers
33:42 Battery Technology and Mass Production Issues
35:33 The Importance of Specialized Jobs in the Economy
38:54 The Evolution of ARM and Its Impact on Microprocessors
42:49 The Shift in Software Development with AI
46:50 Trust and Data Privacy in the Cloud
49:45 The Democratization of Investing and Its Challenges
53:52 The Regulatory Landscape of Cryptocurrency
Key Insights
- TSMC's foundry dominance stems from strategic focus, not outsourcing. Taiwan Semiconductor Manufacturing Company became the global chip leader by specializing purely in manufacturing chips for other companies, while Intel failed because they couldn't effectively balance making their own chips with serving as a foundry for competitors. This wasn't about unions or cheap labor - it was about TSMC doing foundry work better than anyone else.
- Scale economics have fundamentally transformed computing infrastructure. The shift from custom supercomputers like Seymour Cray's machines to clusters of networked mass-produced computers represents a broader principle: you can't compete against scale with handcrafted solutions. Today's "supercomputers" are essentially networks of standardized components communicating at extraordinary speeds through fiber optics.
- AI infrastructure is creating massive resource bottlenecks. Sam Altman has cornered the market on DRAM memory essential for AI data centers, while power consumption and heat dissipation have become national security issues. The networking speed between processors, not the processors themselves, often becomes the limiting factor in these massive AI installations.
- Trust is breaking down across institutions and platforms. From government competence to platform reliability, trust failures are driving major shifts. Companies like Carta are changing terms of service to use customer data for AI training, while social media platforms like Twitter/X are being used as training data farms, prompting migrations to alternatives like Threads.
- Personal software development is becoming democratized while enterprise remains complex. Individuals can now build functional software for personal use through AI coding assistance, but scaling to commercial applications still requires traditional expertise in manufacturing, integration, and enterprise sales processes.
- Cryptocurrency regulation is paradoxically centralizing a decentralized system. Trump's GENIUS Act forces stablecoin issuers to become banks subject to transaction censorship, while major Bitcoin holders like Michael Saylor introduce leverage risks that could trigger broader market instability.
- User experience remains the critical barrier to technology adoption. Despite decades of development, cryptocurrency interfaces are still incomprehensible to normal users, requiring complex wallet addresses and multi-step processes that prevent mainstream adoption - highlighting how technical sophistication doesn't guarantee usability.
Episode transcript
Everyone is wrong about AI, but if you know the history of technology, you can be wrong and still be right.
I know the history of technology since I started with the personal computer in 1980. So listen in while we explore what is right about AI.
Welcome to the Stuart Squared podcast. Got another exciting episode. We're going to get deep into data training with Threads. As the competitor to X elon Musk is building his Neuralink castle where he's going to try to create brain controlled interfaces. And he needed a lot of training data for that. And he's also got his robots. And so Twitter is basically, he's figuring out how to train on all of our data. And then out of nowhere, Mark Zuckerberg comes in and creates this, a much more authentic version of the original Twitter that has now basically given me an outlet to get off of Twitter. I actually have no problem with Musk's politics. I think a lot of what he did during the 2024 election was actually probably smart, just because the opposite was just such a horrible thing.
Yeah, well, no, no. I mean, but I'm not. I'm like, I was temporary. Allied. Allied with him during that crazy time in 2024.
Yeah, Doge. I mean, like, everything that went on from 2020 to 2024, I was strongly opposed against, as were a lot of, I think, the rest of the country as well. But it's like, it's just not the fact that there's. It's just we had only two options and both of those options were not necessarily great. And so I allied with one that I thought was better. And it's true that it is better, but it's still. Still really shitty, basically.
Your first full transcript is free. After that, an email opens every transcript in the index — a list of readers we can write to, not a guest book.