As someone with a business background, I find the concept of open-source fascinating. It’s basically people working for free and sharing it with others expecting that in the future they will also share something for free, and everyone ends up a little better. It goes against every basic business 101 teaching, right?
In my first semester of Business School, I took a class called “Intro to Coding”. Our professor told us that a way to think of Open Source is to imagine two companies going into a price war around a set of codes, and engaging in a race to the bottom, and then suddenly it’s all free, because why not? At the end of the day, it’s just text, and it’s the code that is free, not the data that runs through it, which is what's more valuable.
The reality seems to be that it was all very academic and an underlying spirit that code should be free. I also feel there should be a spirit that first class tickets should be free, anyone else?
So, free code/software means that I can use it and start building on top of it, allowing me to be faster. However, everyone has that code, so it’s not a competitive advantage.
How do they make money you say? Well, they can sell premium services on top of it, but the reality is, most Open-Source projects don't make money, and sometimes rely solely on donations.
If you want to dig deeper, check this resource out.
I am interested in two things when it comes to open source in AI: Hugging Face and Llama, the Meta (prev. Facebook) AI model. Why? Well, because at the time, I felt Meta open sourced their Llama models to become the new standard, and I just keep seeing the Hugging Face emoji everywhere and I was like WTF, is that a face that hugs? or a hugging emoji? What a weird name. Anyways.
Once of the things I keep running into when writing this is that there are several tools that seem to solve a developer issue that I am simply oblivious to. I feel a little of that when looking into Hugging Face. A lot of the description has been that it’s the GitHub of AI, but if you don't know what that is, then that's not really useful. So, GitHub is like Google Docs for devs. It’s a place where the code resides in the cloud and people could collaborate at the same time on it. It also has features that allow for someone to suggest a change in the code, ask permission from their peers to integrate it, and then have it approved. Another key feature is that this was where people could share code with others and host their own portfolio. Meaning, if you needed to solve something, you could maybe find it in GitHub. They make money by selling premium services to companies and was bought out by Microsoft in 2018. So, back to Hugging Face. GitHub for AI as an answer left me asking … why isn’t GitHub the GitHub for AI? Well, GitHub was built to share software products, and usually, that just means code. For newer AI models, the valuable input is not just the code, but the dataset, and the result of running the model. So, this is where Hugging Face comes in. They are platforms where folks can share pre-trained models so that other devs can use them and not start from scratch. As a business student, this would mean that you could use one of these open-sourced models and not have to get all the data and train your own model to have some form of chat interface in your startup. You might be thinking, “Who the hell invests millions (or billions) in training a model and then giving it out for free?” Well, Meta does. This is where Llama comes in. Meta trained their AI model and then open-sourced it, meaning they gave it away for free. Why do this? Well, it’s because Zuck has a huge heart.
No but for real. There is a competitive advantage to doing this. It means that more indie devs will use Llama and Meta tools around it. Think about it, if its free for me to use Llama in my startup or I have to pay for another AI API, then I will probably go with the Llama one. This also means that if my AI team improves something on the base code, Meta sees it and also benefits from it.