Meta Tokenmaxxing, Intel Joins Terafab, Frontier AI vs. China | Diet TBPN

Overview

This episode dives into Meta's internal AI token usage frenzy, the implications of Intel joining Elon Musk's TerraFab project, and the challenges of AI commoditization in the global race against China. The hosts also explore the chaotic dynamics of corporate retreats and the evolving cybersecurity landscape with AI.

Notable Quotes

- Ranking engineers by token spend is like me ranking my marketing team by who spent the most money. We may not have hit our KPIs, but Joe spent $200,000 on a branded blimp that only flies over his own house. – Jordi Hays, critiquing Meta's token-maxing culture.

- Meta could be running a multi-billion-dollar token bill just for internal code-gen tooling. Vertical integration might be their only way to make this sustainable. – John Coogan, on Meta's AI strategy.

- The fire festival of corporate retreats. – John Coogan, describing Plex's disastrous Honduras getaway.

🧮 Meta’s Token-Maxing Frenzy

- Meta employees are competing on an internal leaderboard called Claudinomics to maximize AI token usage, with 60.2 trillion tokens consumed in 30 days.

- Jordi Hays likens this to the eyeballs metric from the dot-com era, where quantity overshadowed quality.

- Critics argue this incentivizes wasteful behavior, such as bots burning tokens in loops, and violates Goodhart's Law: When a measure becomes a target, it ceases to be a good measure.

- Back-of-the-envelope calculations suggest Meta's token spend could range from $55M to $136M per month, depending on usage patterns.

🔗 Intel Joins TerraFab with Elon Musk

- Intel partners with Musk's TerraFab project to design and fabricate ultra-high-performance chips for Tesla, SpaceX, and XAI.

- The collaboration aims to address chip supply bottlenecks and reduce reliance on TSMC, with plans for a facility in Austin, Texas.

- TerraFab chips will support Tesla's AI-enabled products like robo-taxis and Optimus robots, as well as AI compute in space for SpaceX.

- The geopolitical implications of chip manufacturing and the U.S. government's stake in Intel are also discussed.

🔥 The Fire Festival of Corporate Retreats

- Plex's corporate retreat in Honduras turned into a disaster, featuring E. coli outbreaks, military-style drills led by a Navy SEAL, and encounters with fire ants and rogue porcupines.

- Employees endured extreme physical challenges despite being not a super fit group, leading to widespread exhaustion and chaos.

- The retreat, while disastrous, became a bonding experience and a cautionary tale for corporate event planning.

🛡️ AI in Cybersecurity and Frontier Models

- Anthropic's new AI model, Mythos, is being tested by major tech companies to identify and patch software vulnerabilities.

- The rollout highlights the dual role of AI in cybersecurity: both as a tool for defense and a potential enabler of sophisticated cyberattacks.

- OpenAI, Anthropic, and Google are collaborating to combat model copying in China, sharing information through the Frontier Model Forum.

- The hosts discuss the tension between AI safety, commoditization, and the global AI arms race, particularly with China's rapid advancements.

🌌 Space Data Centers and AI Compute

- The feasibility of deploying AI compute in space is debated, with John Coogan noting that Starlink satellites already perform limited compute tasks.

- Challenges include heat dissipation and economic viability, but Musk envisions scaling space-based compute by orders of magnitude.

- The discussion underscores the futuristic ambitions of integrating AI and space technologies.