Future of Client Devices

I am quite certain that with the emergence of AI, we have a new responsibility to transform the current client hardware to match the pace and the capabilities and even the unknown possibilities that we are yet to discover.

For decades we have been using legacy hardware and software to create new frontiers in technology, but now we seem to have reached a limit. This is not a limit of compute on the client side nor a limit of feasibility overall.

We have hit a wall that shows we need new inventions, new innovations to make the client devices and software be more native to the AI technologies that are evolving non-stop.

There have been announcements for new devices from various startups such as Rabbit and also other tools like VR goggles that are more advanced than their previous versions or AI powered glasses.

Even though they are all amazing, there is still some things missing and it cannot be filled with existing approaches.

Privacy and data security aside, we have to figure out new ways of building technology and serve it to clients.

A neural network is not bound to the limitations of a standard client's computer or a cell phone or even goggles.

Look at the robots for example. There have been a lot of different innovations in robotics to merge the more capable AI software with the hardware. They are all amazing, mostly the humanoid robot progress is impressive. But we still lack something in the daily lives of humans, in the daily lives of people who are working in both technology-related fields and non-technology fields.

My guess is that we will always have gigantic data centers, compute clusters and workstations that we will keep in places to power our client devices. I assume this because every time we create a new technology or expand, we always create a gigantic version of it that is too ugly for clients and daily use, but too powerful to ignore. A tiny rock in nature, generally has a mountain connection.

Today's gigantic data centers will eventually become irrelevant because we will transfer that power into the client devices themselves. But then we will realize that we can create the more gigantic plants that do a million times more than what their previous generations did and what new client devices can.

This means that future data centers most probably will have millions of more magnitude in terms of capability than today's data centers. And the capabilities of today's data centers will be transferred into tiny client devices. And this loop will go on for maybe forever.

I say forever because anything humans or even the nature does has a gigantic version of it, then a small mini version of it that are interconnected. The moment we get rid of cell phones or laptops or workstations on daily use, we can be sure that we have transferred all their power and use into data centers. And this is overall a good thing for progress. There is no evidence so far that it is a better option to keep all the compute and data on a local client device and still benefit from emerging technologies with full capacity. You can not.

My laptop has 1TB of storage in it. It has 16GB of RAM. It has an M1 Pro processor that, for those of you who don't know, is a combined chip that has the GPU, CPU, RAM, and some other processing capabilities. All in a single chip.

This used to be an amazing power when it was released. But a couple of years forward today, I look at it and say, "Well, I cannot train a 13 billion parameter language model, not even use LoRA tuning on it. I cannot run a lot of 3D simulations on it. I cannot run AI agents in parallel. I need more compute."

But I'm pretty sure that in the following years, we will be able to do these current tasks on this good old M1 Pro, thanks to never-ending optimizations. Same old hardware, but more capabilities. But this doesn't change the core problem.

We will still have the same amount of computing power, which now will be able to do more and maybe even tremendously more, but when that comes, at the same time, the ability to have more of it will also be there.

So why not have new inventions to make the clients more capable, even as capable as the old workstations, and transfer the workstations into a million times more capable powerhouses?

In the meantime, for today, the approach I embrace is using cloud platforms for the heavy lifting. For example, I mentioned that my laptop has 1TB of storage. It is okay for most daily uses. But I have cloud subscriptions to store my data in the cloud. I do not use the full capacity of 1TB just because I can. Also, when I'm creating a software that is not just the processing algorithms but contains media in it, it easily fills the 1TB storage and I cannot add more. This is the reason cloud storage is the only near future.

The future client devices will have much more storage capabilities, more intelligent storage methods, and we will easily store gigantic amounts of information in just one little tiny chip. But by that time, we will be craving much more than that, so we will go back to storing our gigantic data in gigantic data warehouses and using our tiny client devices to connect. We won't have built-in screens, keyboards or multiple cameras.

We will connect to those only when we need to. We will keep a smart keyboard at home, one at office, one HQ camera at home, one at office for only when we really need to connect to them via our tiny clients and are very very small and portable let alone energy efficient. Why have a phone screen when you can use any screen?