TBPN
TBPN

Open-Source AI Battle, Google Throttles Meta, Micron Margins Moon | Edward Coristine & Tai Groot, Chad Rigetti, Pim de Witte, Yadin Soffer, Jack Morris, Neil Movva, Jakob Diepenbrock, Chris Altchek

June 29, 2026

AI Summary

5 min read

The Wall Street Journal led its front page with a story on China's Z.ai releasing GLM 5.2, an open-weight model that security researchers found matched or exceeded Anthropic's Claude Opus 4.8 at identifying software bugs when given additional instructions.

Open-source model performance and policy questions

GLM 5.2, released June 13, ranked among the top ten models by usage on OpenRouter and drew attention for running without API restrictions. Reviewers noted strong coding results alongside high token consumption, which raised questions about whether its per-task cost stayed competitive with closed frontier systems. Hosts and guests revisited earlier arguments from John Luttig and Dario Amodei that open models would lag on the most capable capabilities while still creating downstream risks in cybersecurity and bio domains. They observed that the gap between closed and open models had not widened dramatically, forcing policy discussions in Washington to focus on how quickly fixes could be deployed before open weights made new exploits widely available.

Capacity limits and pricing pressure in AI infrastructure

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

What you'll learn

  • 1 (00:23) **WSJ Front Page AI Story** - Lead story on China's GLM-5.2 resetting the US-China AI race
  • 2 (02:23) **GLM-5.2 Open-Weight Debate** - Performance claims, national security implications, and open vs closed source trajectory
  • 3 (07:05) **ELO Benchmark Chart Analysis** - Casey/CAISI chart comparing US and Chinese model progress since 2023
  • 4 (09:25) **John Luttig Closed-Source Thesis** - 2024 post revisited on why foundation model value stays closed
  • 5 (12:35) **Dario Amodei 2023 Testimony Clip** - Resurfaced warnings on open-source bio and cyber risks
  • 6 (14:30) **Tyler GLM-5.2 Review** - Distillation concerns, creative writing performance, and token hunger
  • 7 (21:40) **Google Throttles Meta Gemini** - Capacity limits disrupt Meta internal projects

+ Full timestamped outline available in the app

Show Notes

  • (02:21) - Open-Source AI Battle
  • (14:33) - GLM-5.2 Review
  • (21:38) - Google Throttles Meta
  • (27:20) - 𝕏 Timeline Reactions
  • (34:38) - Micron Margins Moon
  • (38:48) - Comcast Splits in Two
  • (43:04) - Europe Meets Suburbia
  • (48:18) - Edward Gorberstein, head of engineering at the National Design Studio, discusses the launch of Ramparts, a local-first privacy model that allows users to control the data they share with AI by keeping personal information on their devices. He explains that existing AI models are too large to run in browsers, preventing in-browser PII removal, and emphasizes that Ramparts is open-source, with weights available on Hugging Face, enabling technical users to create custom applications. Gorberstein highlights the studio's mission to improve the American digital experience by developing user-centric software, citing previous successes like Trumper X, which saved users over $500 million in drug costs.
  • (57:53) - 𝕏 Timeline Reactions
  • (01:03:16) - Chad Rigetti, founder of Rigetti Computing, discusses his journey from developing quantum computing at IBM to establishing his own company, which went public in 2022. He highlights the importance of integrating quantum technologies into data centers to enhance AI capabilities, emphasizing the need for a multimodal approach to quantum hardware. Rigetti also addresses the challenges of transitioning from private to public markets and the significance of long-term strategic planning in the evolving quantum computing landscape.
  • (01:28:42) - Pim de Witte, CEO of General Intuition, discusses the company's unique approach to AI development by leveraging extensive datasets of action-labeled video game footage to train models capable of spatial-temporal reasoning. He emphasizes the competitive nature of the AI industry and highlights General Intuition's distinct advantage: a proprietary dataset that enables their models to predict actions in both virtual and physical environments. Additionally, de Witte announces a recent $320 million funding round, bringing the company's valuation to $2.3 billion, which will support further advancements in their AI research and applications.
  • (01:36:39) - Yadin Sofer, co-founder and CEO of Tracer, discusses the company's emergence from stealth with the launch of a subterranean defense technology firm. He highlights the challenges of underground operations, such as unpredictable geology, and emphasizes the importance of small-diameter, long-length designs for efficiency. Sofer also mentions Tracer's $25 million seed round aimed at collaborating with the military to establish a U.S. subterranean strategy for warfare.
  • (01:42:52) - Jack Morris, co-founder and head of research at Engram, discusses the company's recent emergence from stealth with $98 million in funding from investors like General Catalyst, Kleiner Perkins, and Sequoia. Engram
TBPN

More from this podcast

TBPN →