The author also discusses the possibility of an exponential process in AI development, where each new generation of AI inspires more investment and decreases the cost of the next generation, creating a positive feedback loop. However, if each new AI generation fails to inspire enough investment or reduce costs sufficiently, the AI boom could fizzle out. The author interprets Sam Altman's request for $7 trillion as an attempt to centralize and expedite this process, but predicts that it will likely occur in a slower, more decentralized manner. The article concludes with a critique of OpenAI's approach to AI safety, suggesting that the company is primarily interested in scaling up AI as quickly as possible.
Key takeaways:
- The cost of training AI models like GPT is increasing exponentially, with estimates suggesting GPT-6 could cost $75 billion and GPT-7 could cost $2 trillion.
- Training these AI models requires significant resources, including computing power, energy, and training data. The demand for these resources is expected to exceed the world's current capacity.
- There are potential solutions to these challenges, such as synthetic data and algorithmic progress, but these are still in early stages of development and may not be sufficient to meet the growing demands.
- OpenAI's plan to scale up AI as quickly as possible has raised concerns about safety and the potential for a sudden, disruptive leap in AI capabilities.