When AI reaches a ceiling

I think there’ll be an automatic ceiling. AI has done so well, so quickly, but in two or three years’ time people will be like, “You know what? It’s not needed.” They’ll be like, “I have what I have, which does what I need. I don’t need an update or an upgrade.” There is going to be a ceiling beyond which AI’s improvements would cease to remain useful for people. Technology can evolve, but it depends largely on the users.

If you want to go from your home to the office, you don’t need a fighter jet. You can simply take a taxi, even if it takes more time. Similarly, if you plan to go from India to America, there might be a plane which takes you there in 23 hours, but there’s a ceiling to how much people would be willing to pay for it. Beyond a point, people are not going to pay for it. It is not as if we can’t build a fighter-jet-like passenger aircraft, but market conditions dictate how much is enough.

If you compare laptops, laptops have been in business for many years now, but there’s a ceiling to how far people can go with laptops. You can make a laptop entirely from glass. Maybe you can do anything and it doesn’t take any seconds. You open your screen and it’s right there. You can do a variety of things and operate it from anywhere, perhaps from your phone. There could be tons of new functionalities in a laptop. With laptops, there’s a ceiling to how much people are ready to pay for it. The same is going to happen with AI. People will be like, “You know what? I don’t need this facility.”

For ChatGPT, for example, I look at ChatGPT Astro and feel like it’s a bit unnecessary for my work. It consumes a lot, while I feel like Soul is already doing the work for me.

When that ceiling applies, I think there’ll be a breakaway where people say, “Okay, it’s okay not to test the latest intelligence model in ChatGPT.” ChatGPT Astro might be exciting for one or two days, but then you realise that what you already have is doing the work. Don’t keep buying technology you’re never going to use. Don’t buy an AI model which is doing pictures when you’re not going to take pictures. If you’re not editing pictures or videos, don’t buy it for that.

When AI starts making assumptions and doing things for you, I think it is becoming more intelligent. When people move away from it and there’s an inflection point, what I’m trying to say is that the intelligence would have a ceiling. Beyond that, you could develop a model which does everything for you. You could say yes and, in five seconds, a website like Instagram is built, taking everything into account. But who is buying it? There has to be some use for it. A software developer company might buy a five-second model, but they are building that model for someone who probably doesn’t need it.

AI has learned human behaviour and has built these algorithms, but human behaviour is also going to constantly evolve. Instagram now may become a useless app, and with changing behaviours AI would need to learn those changing behaviours. It would need to make assumptions differently, including assumptions never made before. In the future, it would have to take into account many more assumptions which are not even written anywhere. That is because human behaviour is ever evolving.

One of the things I find fascinating is that Claude said it made a lot of changes to my code without my permission. It said it was because it was trained this way on millions of lines and performed the duty as per that. It took assumptions, just like people do. In my case, Claude simply applied all of that without my permission. I don’t want to go to the tip of the intelligence because no one can control that. When that begins to happen and the balance begins to tilt, I had to unsubscribe from Claude because I didn’t think it was helpful. I want to be involved. I want to be associated with what’s going on.

I think there may also be a case that AI models would be banned and put outside. Banning is different, but people may simply stop subscribing to them. When people stop subscribing, the rate of evolution may be stalled because they will not have many users. When that happens, there might be a reverse revolution, going back to the past a little bit, where you rely on human voice.

Right now, entry-level jobs are gone. Then the five-year-experience guys are gone. Then the ten-year-experience guys are gone. Eventually, top management begins to question whether they should be there in the job or whether one of them should be fired. When it reaches a stage where management is saying, “Well, this is my business,” and if AI is going to take over our business, then where is the business?

Right now, people are not too worried because entry-level jobs are gone. They would be worried when the ten-year guy, an experienced guy, has lost their job. Then it escalates into a lot of people who are jobless. I think governments will come up with universal pay. But people need a sense of purpose behind work activities. They can’t just sit idle at home and do nothing. I don’t know how it’s going to work.

Maybe businesses deliberately keep humans involved. This is the ceiling I’m referring to. Right now, it’s efficiency improvement. Beyond efficiency improvement, maybe people would be like, “I don’t want a case where I am being served coffee by a robot barista.” I want to speak to the barista. I want to talk to my Uber driver.

There might be a ceiling where society would be like, “We don’t want this.” It’s not that it’s not possible. That is the ceiling I’m referring to.

There is a greater risk that AI would be misused by powerful people who have the latest model and know how to manipulate everything. For them, nothing can be done about it. But if you think of society in general, when that ceiling arrives, I think there might be a push for human content. There might be a push for bloggers.