Keep pulling the thread on Alex Imas and Phil Trammell.
The cost to rent an NVIDIA H100 GPU is currently higher than it was three years ago, reversing the long-term trend of decreasing computation costs.
A recent report from the Budget Lab at Yale University found little to no evidence of mass automation or unemployment caused by AI, even in highly exposed sectors like software engineering.
Alex Imas contrasts two potential economic models for AI: one like electricity, where downstream users like ComEd captured most benefits, and another like social media, where platforms captured most of the value.
If AI's economic gains are concentrated, it will be hard to index; if they are diffuse like electricity, developing countries like Nigeria can gain exposure by buying a broad market index.
Alex Imas suggests a potential scenario where developing countries like Nigeria are left behind by the AI revolution if automation allows developed nations to produce commodities they previously imported.
The U.S. government's threat to use the Defense Production Act against Anthropic highlights the political risk created by having a few highly concentrated AI labs.
The widespread adoption of mobile banking in countries like Nigeria demonstrates a "leapfrogging" effect, which could potentially be replicated with transformative AI technology.
Despite a recent increase in the privatization of returns, private companies still account for well under 20% of the total market capitalization of non-tiny companies in the U.S.
Phil Trammell predicts that AI will alleviate frictions keeping companies private, such as disclosure requirements, making it easier for them to go public.
Phil Trammell argues that a leading AI company could be a widely-held public entity, which would provide a safety buffer without concentrating wealth, mitigating the perceived trade-off between AI safety and commoditization.
The current prime age employment rate in the United States is at its second-highest level ever, just below the peak reached in the year 2000.
The O*NET database, a U.S. government resource for tracking job tasks, is rarely updated and considered to be of low quality.