Yesterday I listened to a podcast interview about AI, and it left a deep impression on me. The guest was the person in charge of Replit Agent, who talked about the experience of using the latest cutting-edge large models, the development of AI technology, and many thoughts on AI.

After listening, I felt that whether you are an outsider or an insider in AI, you could learn something from it. It's an interview that appeals to both the elite and the masses, and I highly recommend everyone go listen to it.

The guest mentioned several points that left a strong impression on me.

The first point is that Self Evolving Agent is the AI direction he is most optimistic about in the next 12 months.

This is similar to the concept of Recursive Self Improvement that I talked about in a previous post, and it is also what I believe is the next biggest direction in the AI application layer after Coding Agent.

An Agent that forms a self-closed loop could theoretically run 24 hours a day, 365 days a year. Once this direction truly lands, the token demand behind it will be several orders of magnitude larger than it is now, and you can imagine the investment opportunities behind it.

However, currently the real-world resistance to RSI still outweighs actual progress. I think we can keep observing.

The second point is that currently 99% of people worldwide have not yet boarded the AI train.

The guest mentioned a case from his own travels: when he was taking a taxi in Houston, USA, he deliberately talked to the drivers about AI. It turned out that none of them had heard of Vibe Coding, and half of them had never even used ChatGPT.

Coincidentally, last night I had dinner with a friend who is planning to start an AI business. He said that the vast majority of people around him have only used Doubao (a Chinese AI assistant), and many have never heard of ChatGPT or Claude. This aligns exactly with the viewpoint from the interview.

I instantly realized that I had fallen into an information cocoon. The audience who often watch me on Twitter and YouTube must have all heard of these large models, and many have paid the $20 subscription fee to use them daily. I often mistakenly thought that cutting-edge large models have already become popular worldwide, but the truth is completely the opposite.

If in major countries like China and the US, many people still haven't used large models and don't know what Vibe Coding is, it's even less so in other countries. The potential of the AI market is far, far larger than we imagine.

The third point is to open up the ceiling of imagination for AI.

The guest mentioned that human imagination is currently the biggest bottleneck of AI. When Edison invented the light bulb back in the day, he would never have imagined that humans could invent computers based on electricity.

I have always felt that compared to previous technological revolutions, the biggest characteristic of AI is that it is a completely new species, not a tool. The two are entirely different concepts. Throughout history, humans have invented many tools, but creating a new species is the first time, and behind this lies infinite imagination.

If you lack imagination and only focus on existing AI, the most intuitive feeling is that AI is about to reach its end. This is typical pessimism: being right in the short term but absolutely wrong in the long term.

Elon Musk often says, it's better to be optimistically wrong than pessimistically right. I strongly agree with this statement. The development of AI is accelerating faster and faster. We may already be at the next technological singularity. Why should we not be optimistic?