Dwarkesh and Noah Smith on AGI and the Economy

Unknown Source August 04, 2025 61 min

artificial-intelligence ai-infrastructure investment generative-ai openai microsoft anthropic meta

🎧 Listen to Original

50 Companies

117 Key Quotes

4 Topics

4 Insights

🎯 Summary

Podcast Summary: Dwarkesh and Noah Smith on AGI and the Economy

This 61-minute discussion between Dwarkesh Patel and Noah Smith (author of No Opinion) centers on defining Artificial General Intelligence (AGI), assessing the current capabilities of AI models against that definition, and exploring the profound economic implications of achieving true automation.

1. Focus Area

The primary focus areas were:

• Defining AGI: Moving beyond raw intellect to an economic definition based on job automation capability.
• Current AI Limitations: Specifically, the lack of continual learning and context retention in modern LLMs compared to human workers.
• Economic Impact of Automation: Analyzing the debate between AI as a substitute versus a complement to human labor, and the potential for unprecedented economic growth (galaxy-scale growth).
• Future of Work and Demand: Debating who will consume the massive output generated by near-total automation.

2. Key Technical Insights

• The Continual Learning Gap: A major bottleneck preventing current AI from achieving AGI is the inability to engage in continual learning—the process where humans interrogate failures, build context over time, and integrate feedback over months to become better at a specific task. Current models have their understanding “expunged by the end of a session.”
• Capability vs. Economic Value: While current models exhibit reasoning capabilities, they often fail to translate this into the massive economic value ($10k/month per human) seen in human labor because they lack the necessary contextual integration and long-term adaptation skills.
• Alien Intelligence: AI is expected to remain an “alien intelligence,” meaning its reasoning and problem-solving methods may fundamentally differ from human intuition, emotion, and context-building, even if it achieves high performance.

3. Business/Investment Angle

• Complementarity vs. Substitution: The prevailing narrative treats AI as a perfect substitute, but historically, technology acts as a complement. The long-term economic impact hinges on whether AI labor complements or replaces human labor, though AI labor has the advantage of extremely low subsistence/marginal costs (e.g., electricity vs. human upkeep).
• Underestimation of Job Complexity: Past predictions of mass job displacement (e.g., truck drivers, radiologists) have failed spectacularly because they underestimated the sheer number of capabilities required to fully automate a job beyond core reasoning (e.g., the long tail of physical or contextual tasks).
• Galaxy-Scale Growth Potential: If AGI is achieved and capital/labor become functionally equivalent (buildable/scalable), the potential for explosive growth (speculated at 20% annually) exists, driven by agents pursuing massive goals like space colonization, regardless of immediate consumer demand in the traditional sense.

4. Notable Companies/People

• Noah Smith: Author of No Opinion, providing an economic perspective on technological progress and labor markets.
• Dwarkesh Patel: Host, framing the discussion around the “scaling era” and historical context of technological shifts.
• Sam Altman/Ilya Sutskever (Implied): Mentioned in the context of defining “superintelligence” and observing that current high-level AI hasn’t immediately transformed the world as predicted (the “1973 test” observation).
• Waymo/Uber: Used as an example where user preference for seamless, personalized service (even with glitches) suggests humans may value human interaction or specific service qualities over pure automation in some sectors.

5. Future Implications

The conversation suggests the industry is heading toward a critical inflection point where the technical hurdle of continual learning must be overcome before AGI unlocks true economic transformation. If this hurdle is cleared, the bottleneck to growth shifts from human population limits to political/regulatory constraints or the goals of the dominant economic agents (whether human or autonomous AI). The structure of GDP itself may change if autonomous agents drive production for non-human consumption goals (like galactic expansion).

6. Target Audience

This episode is highly valuable for AI Strategists, Venture Capitalists, Economists, and Technology Executives interested in the long-term, macro-level implications of AGI, moving beyond near-term product cycles to fundamental shifts in labor economics and growth theory.

🏢 Companies Mentioned

Matt Bruenig ✅ economist_theorist

Miles Kimball ✅ economist_advisor

Mark ✅ big_tech_figurehead

BYD ✅ ai_adjacent_industry

Elon Musk ✅ ai_leader/executive

And Leopold ✅ unknown

Because I ✅ unknown

Can I ✅ unknown

If Elon ✅ unknown

Steven Spielberg ✅ unknown

Mark Andreessen ✅ unknown

So I ✅ unknown

Matt Bruenig ✅ unknown

Miles Kimball ✅ unknown

Costa Rica ✅ unknown

💬 Key Insights

"That will teach you like how big is the model? like you learn a lot just from publicly using a model and knowing"

Impact Score: 10

"But at some point, you can't keep this like 4x trend going a year. And after that point, then it has to just like come from new ideas of like, here's a new way to get trained in model."

Impact Score: 10

"The long time when people will say, I don't know, there's a sort of long argument. I don't know how much to bore you with this, but basically, the things we think of as very difficult and requiring intelligence have been some of the things that machines have gotten first. So just adding numbers together, we got in the 40s and 50s. Reasoning might be another one of those things where we think of it as the apogee of human abilities, but in fact, it's only been recently optimized by evolution over the last few million years, whereas things like just moving about in the world and having common sense and so forth and having this long-term memory, evolution spent hundreds of millions, if not billions of years, optimizing those kinds of things. So those might be much harder to build into these AI models."

Impact Score: 10

"Now, if in future, your population is, your effective labor supply is like largely AIs, this dynamic just means that like your inference capacity is literally your geopolitical power, right?"

Impact Score: 10

"technology has already destroyed the human race, and basically UBI is just like keeping us around on life support for a little while while that plays out."

Impact Score: 10

"So if an A100 costs a couple thousand dollars a year to run, the value of an extra year of intellectual work is still like a hundred thousand dollars. So you're like, look, we've saturated all the A100s, and we're going to pay a human a hundred thousand dollars because there's still so much intellectual work to do. In that world, the return on buying another A100, like an A100 costs $40,000, just like in a year that A100 will pay you over 200% return, right? So you'll just keep expanding that supply of compute until basically the A100 plus depreciation plus running cost is the same as an extra year of labor."

Impact Score: 10

📊 Topics

#artificialintelligence 169 #aiinfrastructure 11 #investment 10 #generativeai 8

🧠 Key Takeaways

💡 do a UBI today, but like, if all human wages go below subsistence, then the only way to deal with that is through some kind of UBI rather than if you happen to see you open AI, you get a trillion-dollar settlement, otherwise your group