Power to the People, Through My Lens: An Interpretive Take on LLMs

Review of Andrej Karpathy’s X, who is Previously Director of AI @Tesla, founding team @OpenAI

Summary: The Unique Diffusion Path and Impact of LLMs

Historically, technology has followed a top-down diffusion path, spreading from government and military sectors to corporations, and eventually reaching the general public. This progression occurred because early technologies were capital-intensive, rare, and required specialized knowledge to develop and use.

Electricity, Cryptography, Computers, Flight, Internet, GPS, those trends feel very natural. New and powerful technologies are typically scarce, capital-intensive, and require specialized technical expertise in their early stages.

What the LLM is..?

So, the fact that LLM is radically reversing this traditional pattern is so unique and remarkable. While it disproportionately benefits regular people, its impact on corporations and governments is significantly more muted and delayed.

The barrier to entry is too low. The models are extremely affordable, even free, fast, and accessible through just a URL. They can even run on local devices. And they understand and communicate in each user’s native language, slang, tone, and even emojis. This is the first time in history that such a large portion of the general public has experienced a technological leap this powerful, this quickly.

So then, why is the benefit so much more limited in the corporate and government sectors?

Why Are Corporations and Governments Slow?

Corporations and governments are surprisingly slow to adopt LLMs, especially compared to individuals. One reason is that LLMs offer broad but surface-level knowledge. For experts, this means they are useful as assistants for helping draft legal clauses, summarize documents, or write boilerplate code but they can not transform the entire capabilities of specialists. On the other hand, individuals with limited domain, low level expertise benefit much more from LLMs’ wide ranging abilities.

Large organizations also face more friction. They operate through complex systems, legacy codes, brand guidelines, strict security protocols, regulatory compliance, and more. Even one flawed output can be risky. Simply, LLMs are fast and flexible, but institutions are not.

Looking Ahead

Currently, even state-of-the-art LLM performance is highly affordable and accessible. Beyond, spending more money does not directly go to better performance. We cannot buy a better ChatGPT with more money, **Bill Gates is chatting with the same GPT-4o as us. **

Conclusion

We are living through a so unique situation in the history of technology. What we’re witnessing now would have once been considered a top secret super intelligence project managed for governments.

“The future is already here, it’s just not evenly distributed.” - William Gibbson

And yet, surprisingly, the future is not only here but also, it is been distributed with shocking equality. Power to the people. Personally, I love it.

My Point of View, Individual VS Democratization of Experience

“Experience,” Once Exclusive to Experts, Now Belongs to Everyone

In the past, developing an app or reading and analyzing academic papers was the domain of experts alone. Complex tools, knowledge, time, and even degrees were required. But now, things have changed. Thanks to LLMs, students, professionals, and ordinary people can venture into realms they never imagined possible. Even without expertise, anyone can fully engage with a slice of that experience.

The Turning Point: When Abilities of Paying Create Disparities

Right now, everyone uses similar models. However, if premium AI models like GPT-8 Pro or Ultra start leading real productivity gaps, we could go back to an era where capitalists and elites dominate once again.