Ep. 023 - Everyone Leaves Google, Elon Forecasts 1T ARR, Reflecting On GPT-5, Building Personalized Software (Roundtable) | Jon Y, Doug O'Laughlin, Jordan Nanos

Summary

This episode connects three threads: model quality, supply chains, and organizational decay. The panel says GPT-5 itself was underwhelming, but later versions and coding-focused capabilities got substantially better, making AI much more useful for multi-step software tasks. They then dig into export controls on 1.6T transceivers, arguing the Chinese supply chain is so deeply embedded that bans may hurt Western companies more than China. The biggest management debate is Google: Jeff Dean and other high-profile departures are treated as a symbol of bureaucracy, shifting priorities, and a possible move from frontier-model leadership toward TPU/infrastructure monetization. The conversation closes on personalized software and AI safety, with John describing custom tools built around his workflow and Doug warning that malicious AIs could make the internet hostile enough to require hard perimeter defenses.

Notes

• Jordan frames the episode around Chinese component bans, Google departures, and Elon-linked AI capex forecasts.
• GPT-5 is treated as conceptually right but not especially good in practice.
• Later coding-focused model improvements are described as a major capability jump.
• Doug says 5.2 was poor, 5.6 is good, and six is imminent.
• John is broadly hawkish on chip export controls, but says firms can evade the spirit of the law.
• Doug argues the 1.6T transceiver supply chain is mostly in China.
• Jordan says the issue is that bans are now hitting best-in-class products, not cheap knockoffs.
• Jeff Dean’s departure is presented as a symbolic loss for Google.
• Doug calls Jeff Dean the 'Chuck Norris of Google' and says the symbolism matters.
• Doug says the serious AI talent at Google no longer views Google as the best place to build frontier models.
• Doug argues Google has mostly absorbed external innovation rather than creating it internally.
• The Bell Labs/Kodak analogy is used to argue that inventing important technology does not guarantee capturing the value.
• Doug extends the analogy to Google inventing the transformer and then being displaced by the technology it helped create.
• Doug says Google could turn into an infrastructure or TPU-centric business instead of a frontier-model company.
• John proposes forcing Google to license its internal science while keeping the search monopoly.
• The speakers caution that R&D departures may matter less when work is deeply team-based.
• Doug says Jeff Dean is a rare system-level talent across the full stack.
• Jordan says the logic of the new AI labs implies that compute and data alone are not enough.
• Doug casts AI lab competition as a Protestant Reformation-style split.
• Doug warns that future AI competition could become hostile and destructive.
• Doug expects malicious AIs to make the internet unsafe for ordinary browsing and transactions.
• John says a great firewall may be the practical fix for AI-enabled internet risk.
• John jokes that humanoid robots cannot swim long, so physical barriers would help too.
• Jordan connects the AI demand story to Elon’s terafab and a pulled-forward $1T revenue forecast.
• John says Taiwan prioritizes fabs over data centers because power is scarce.
• John says AI adoption in Taiwan is still relatively behind and mainly used for translation.
• John built a custom video editor around his keyboard-centric workflow.
• John recommends rolling your own software and says he plans to build a text editor next.