Andrej Karpathy, former AI director at both OpenAI and Tesla, recently posted an insightful analysis, categorizing human engagement with large language models (LLMs) into four distinct tiers. These tiers, he noted, exhibit a sharp decline in scale and significant disparities in cognitive experience. Globally, only around 5,000 individuals have access to the most advanced internal AI systems, leaving the majority of people with minimal to no real-world exposure to this transformative technology.
Karpathy explained that the core reason for these vast differences in experience among different groups lies in the fact that the true power of cutting-edge LLMs only becomes evident when tackling large, complex problems that require a long-term perspective. In contrast, small or routine tasks fail to reveal their full capabilities. Additionally, he highlighted that the potential of LLMs will initially be unleashed in purely digital, verifiable, and high-economic-value fields, where their impact can be most clearly measured and leveraged.