Search engines like Google have become so critical in our day-to-day lives, especially for the desk working class. Engineers, scientists, and almost any profession that utilizes the internet have grown to become dependent on search engines.

It’s hard to imagine how one can be productive without search engines these days. On the other hand, if you take a closer look at how these search engines work, you’ll notice that there’s much more room for improvement.

The general concept of Human-Computer Interaction (HCI), and how we access information on the internet is changing with the progress of AI development.

Foundation Models

The next Google isn’t likely to be powered by page-rank like algorithms, or any traditional algorithms for that matter. In the software 2.0 world, the next Google will likely use a mixture of huge foundation models, and some advanced engineering techniques like continual learning.

I firmly believe that foundation models have enormous potential to unlock more human talent and empower people by not only tools to express their ideas like stable-diffusion, but something much more powerful.

This mind-blowing exponential curve of github stars from Emad Mostaque, founder of Stability.ai says it all:

Stable diffusion as a text-to-image model, was able to see such a wide adaptation rate of almost a vertical exponential curve, imagine what else can we build that would be useful.

Right now, we’re currently in the prompt-engineering era. With time, we’ll see easier ways to interact with these models, and the experience will eventually converge to something more human like the science-fiction movie Her.

LLM for Browsers

A current trend that some companies are following is using Large Language Models (LLM) to manipulate web browsers. Adept.ai is a very interesting company working on this, with their latest release ACT-1.

Although we’re still at the very early stages of LLM for browsers, there’s a lot of potential to make a big impact on how we currently interact with computers.

This is a small example that Nat Friedman has built, natbot that controls a browser using GPT-3 to do some caffe reservations. Another cool example was Sharif Shameen’s hack to buy AirPods using GPT-3:

These are simple yet inspiring examples of the potential LLMs have to change how we use computers and the internet.

Continual Learning

Why continual learning you might think? We’ll we might get away with working products using enormous foundation models trained on the whole internet, but eventually, these models will need retraining and improvements.

Now I’m not saying that continual learning will be sufficient or the only solution these models might need, but I do believe will need some very smart engineering solutions to bring these models into production and give people the best experience possible!

Conclusion

The next generation of search engines and the way we use computers are improving with the advancements in AI, empowering more people every day and changing the world for the better.