ChatGPT won’t cut it; You will have to have your own LLM
Just like operating systems before them, the idea of LLMs will not be confined to private ownership. GPT is head and shoulders above everyone else right now, but the free market will destroy its barrier of entry. Then everyone with the means to do so will be able to train state-of-the-art LLMs for internal use or to offer as a product/service.
This is because LLMs are not a product. They are a technology, and that’s what happens with technologies
One thing before we get started
I will admit that I editorialize the titles a bit to make them jump out in people’s mental data input streams, and that does require massaging the accuracy a bit. So now that you’re here and your interest doesn't need any additional piquing, a couple of corrections in the title:
- When I said “ChatGPT” I meant all the LLM products OpenAI offers. I said “ChatGPT” to leverage the brand it has built around it among normies
- You, the end user, will not have your own LLM; because you won’t need to. What I meant is that the end user will necessarily require that companies and open source initiatives be able to train and host an LLM on their own
Now let me explain what I mean
We’ve already seen this happen
This happened with the operating system.
Long story short, there was a time when the idea of personal computers was stupid. Computers were business machines; they weren’t meant for normal people to use. And even if the general population could find a use case for them, they wouldn’t be able to operate them because the OS was so esoteric.
This meant that general-purpose OSs were something large companies made for their own use and trained specific people to operate. The technology was basically sequestered from the public. You had a few companies like IBM and AT&T making and maintaining these things. And if you wanted to use it, you would have to pay whatever price these companies felt would cover their research costs (just like OpenAI’s pricing right now).
Then the personal computing market started to snowball. Companies started making OSs with the intention that they will be used by normal people in their bedrooms. Soon this market dwarfed and then consumed the business computer market. At one point companies started using personal computers in their professional operations. The barrier of entry became so low that the OS became open source in the mid-90s, in the form of Linux, and now that OS is on every server that drives our society.
Now, almost everyone in the world interfaces with an OS every day and we don't even give it a second thought. We don’t have to worry about the cost or whether we will be able to operate it because the industry opened them up to anyone to redesign with and everyone refined them for personal use.
The operating system went from tech that only a few giants could create and provide, to something you can tinker with and make a new product out of today, for anyone to use.
And the same will happen to LLMs
Soon LLMs will become something anyone will be able to train and build upon.
GPT was trained on large text corpora on the open internet but someone will try to see how they perform when they are trained on data that isn’t available on the internet; and in as well packaged a bundle as Wikipedia or Google News or Twitter. Someone will try to see if Large Language Models are only good with languages or if they will work with other sequential data like time series or activity streams. Someone will try to train an LLM on their financial transactions instead of natural language to see if it can do financial forensics.
These someones won’t be the end users, but their efforts will be extremely beneficial to the end user. There will be a point where users will demand that LLMs be open to the public, and no tech giants will be able to stop them. Just like users demanded more open alternatives to UNIX and other jealously guarded OSs
This isn’t a socialist pipedream; the free market will do this
I understand that my saying that LLMs should be free (as in freedom) is gonna make you think I’m some wide-eyed idealist who doesn't understand why things have to cost money and why can’t we all just get along, man. But that’s not the case. This change will not be driven by some righteous crusade but by the good old invisible hand of the free market.
Don’t get me wrong: I like socialism as much as the next millennial (quite a bit). But what I’m describing will be the stable equilibrium that free market capitalism will move us towards. When I say the user will demand open LLMs, I don’t mean that it would be a conscious demand. The user will simply gravitate towards products that offer the best value proposition, and that will be made by enterprises that train and host their own LLM (or even let the users host the LLM themselves)
We know this will happen because that’s what already happened. Take Linux as an example. Users don’t consciously demand that companies use Linux on their servers. Users wouldn’t even like Linux if they had to use it themselves. But if a company needs to use a server infrastructure so as to be able to offer its product, you can guarantee that it will use Linux. Because if it doesn’t then the users will leave them. The OS won’t be the direct cause but the company’s product will be worse if it doesn't use Linux. That’s because Linux is the best solution for a server OS, and that’s because anyone can just come in and improve it.
The same will happen to LLMs. not a lot of competition in the market right now but OpenAI’s dominance will end far, far sooner than you think and the market will create competition.
So start thinking about what you would do with an LLM-from-scratch
TL;DR
- LLMs are not a product, they are a technology
- The free market has shown us that it opens up technologies for everyone to productize
- The biggest example is the general-purpose operating system
- Users won’t have their own LLM, but they will demand competition in the market