Leapfrogging, is what many African countries did in telecommunications.

‍

By the time that these countries started rolling out national networks they could skip copper based networks and "leapfrog" to mobile directly.

This theoretically saved the capital cost of the copper networks, albeit at the expense of many years of no telecommunication.

If there were a country that embarked on a huge copper network just as mobile technology was becoming cheap and accessible, it would have been a waste of capital.

This is because new technology emerged, it replaced the evolving technology of copper & fibre voice networks.

I see this happening with LLMs.

I believe Large Language Models and their generalized intelligence output are a new technology.

It's difficult to draw an exact line between technology improvement and differentiate between new and evolving.

‍

How could we distinguish between new and evolving technologies?

‍

Let's take the example of telecommunications.

After the invention of the telephone, the basic premise was much the same - speak to another person, one to one.

The large switching centers employing many people got replaced by dialing a telephone number, I’d call this an evolution of the technology.

Another evolution was fax, leveraging the copper network to send documents.

Area codes, international calling, SS7 and all these improvements evolved the technology to the point where you could reach many fixed points on earth by just knowing the telephone number.

‍

Mobile however changed something fundamental. You could now reach a person anywhere on earth, decoupled from the fixed location.

This replaced the need for a home phone in most people's lives. I don't know of anyone that installs a fixed line into a new house anymore, maybe a fibre line for the internet, but not for voice calling.

‍

It became easier to build out communication to more people, requiring a mobile tower close by to serve thousands of people simultaneously.

Fax also became obsolete as mobile phones assumed more and more roles : voice, document handling, calculator, watch & camera, to name a few.

We can see from this example that mobile phone technology made the fixed infrastructure to homes obsolete (for the purpose of voice).

It also was significantly easier to deploy and enabled much more use

Mobile phones themselves also made many dedicated devices obsolete.

‍

Back to LLMs.

‍

I think the Large Language Models are a new technology that will replace many evolving technologies.

Some examples of technologies that will be replaced by LLMs :

• Chat Bot systems, previously painstakingly built with large decision trees and keyword matching
• IVRs, Menus, Dedicated apps : systems that are setup to determine the human’s intent
• OCR, LLMs with vision can understand the document will eventually replace traditional OCR technologies.
• Spam filtering, when tokens become cheap enough, having an LLM decide what email is relevant will be more effective.
• Language translation, dedicated systems for this will be obsolete.
• Speech-to-text, this will become just audio tokens directly into the model

‍

Expliciting realizing when something is a new technology vs an evolution of existing technology is important.

It's important to realize that many existing paradigms need to be questioned.

‍

It's exciting to be part of this new technology in Stubber.

I'm looking forward to building many new applications using LLMs.

What will you build?

What evolving technologies are about to get replaced?

‍