E122: Is AI the next great computing platform? ChatGPT vs. Google, containing AGI & RESTRICT Act

David Sacks said ChatGPT cut a blog post from roughly a week of research and writing to one day, while humans still did substantial editing and review. He now plans to use it routinely as researcher, first-draft writer and editor, especially to eliminate the blank-page bottleneck.

Sacks’s more durable startup idea was “gift to get”: reward users with credits for contributing scarce proprietary data, then let them spend those credits using the AI. He argued this could seed vertical models in fields such as architecture or medicine, where plans, lab results and reports are difficult to obtain.

OpenAI’s plugins changed the strategic question from “AI API” to “AI destination”: ChatGPT could browse, call services and move from a natural-language request to an actual transaction. Friedberg’s key point was that this closes the gap between search and commerce, threatening interfaces built around lists of links rather than completed actions.

The group disagreed on whether OpenAI’s early lead would compound or be overwhelmed by incumbent distribution. Sacks emphasized ChatGPT’s user attention and developer ecosystem; Chamath emphasized Google, Meta, Apple and Microsoft’s built-in distribution, proprietary data and economic incentive to keep users inside their own products.

Friedberg’s most specific Google diagnosis was institutional, not technical: regulatory scrutiny and a defensive approval culture had kept strong AI capabilities focused internally rather than turned into consumer products. He argued Google had the talent, data, hardware and cost structure to compete, but would need to accept short-term revenue risk and organizational disruption.

The panel did not expect an AI pause to hold because capital, talent and competitive pressure were accelerating the race, not slowing it. Sacks noted YC was already heavily AI-focused, while Friedberg argued software models could be copied, forked and improved too easily for regulation or intellectual-property barriers alone to contain them.

The clearest labor example was a 3D game artist who said Midjourney cut character work from weeks to two or three days, leaving him mostly prompting and editing. Chamath expected displacement to show first at Accenture, TCS and Cognizant, where margins directly reward doing the same coding work with fewer people.

The final policy concern was that the 2023 RESTRICT Act reached beyond TikTok, giving Commerce broad authority over foreign-adversary-linked technology transactions. The bill carried maximum penalties of $1 million and 20 years for willful violations, but the transcript’s claim that ordinary VPN use alone could trigger them was not explicit in the bill.