Keep pulling the thread on Dwarkesh Patel, Noah Smith.
Dwarkesh Patel predicts that AI will eventually develop the necessary capabilities to automate human labor, unlocking trillions of dollars in economic value.
Dwarkesh Patel predicts that once AGI can be used to build more AI, global economic growth could exceed 20% annually.
According to Noah Smith, intense, government-subsidized competition in China's EV market is causing financial strain for leading companies like BYD, which has reportedly taken loans from suppliers to stay afloat.
The compute used for training frontier AI systems has grown at a rate of 4x per year for the last decade.
Dwarkesh Patel argues that the 4x annual growth in AI training compute is physically unsustainable and can only continue for approximately five more years before hitting economic and resource constraints.
Dwarkesh Patel believes there is a clear technological path for AI to automate 95% of white-collar work.
OpenAI's annual revenue is approximately $10 billion, which is less than the annual revenue of companies like McDonald's and Kohl's.
A key capability distinguishing current AI models from human intelligence is their lack of continual, long-term learning, as their context is expunged at the end of each session.
Dwarkesh Patel believes that because there is no obvious solution for continual learning in AI, achieving AGI is many years away.
In cities where Waymo's autonomous ride-hailing service is deployed, it has experienced high consumer demand, leading to wait times of up to 20 minutes.
Dwarkesh Patel asserts that the marginal cost of operating an NVIDIA H100 GPU is significantly lower than the subsistence cost of a human worker for a year, giving AI labor a fundamental economic advantage.
Contrary to predictions made around 2015 that self-driving trucks would eliminate trucker jobs, a decade later there is a trucker shortage and employment in the sector is higher than ever.