Why you want to use Open-Source AI for work and life
In today's changing business landscape, where efficiency reigns supreme...
Cringy marketing-speak aside...
Welcome back to another post.
Before the year ends, there's something I want to tell you.
And yes, it involves AI. But it's not like anything I have said in previous posts.
This is more on the side of what kind of power you want to wield.
Having said that...
The Battle: Open-Source vs. Proprietary AI
We have all seen the massive development and improvement AI has achieved during the past 12 months.
Most of that development is due to big tech companies like "OpenAI", Google, and Anthropic.
But what the majority of people aren't aware of is the equally impressive development AI has had in the open-source community.
Many of the things that were impossible even a year ago are now within reach.
Either someone else already created something for that need or there's research that can be used to create a new product that helps others.
You no longer have to be "stuck" using one big company API's.
You can now choose if you want to use a proprietary model or one of the many that are available on Hugging Face.
This battle between the 2 fronts will keep on going for the next year 2024. If you want to go more in-depth over the advancements on both sides, take a look at this article.
Now, let's keep going.
Embracing Open-Source AI
For every offering the tech giants come up with in AI, there's an open-source project that offers similar (if not the same) capabilities.
Here are a few examples:
Open Source: LLaMA 2 Chat, Mistral-instruct, OpenAssistant, Falcon Chat, GPT-NeoX, StableLM, etc.
Open Source: Stable Diffusion, Pix2Pix Deep dream generator.
Open Source: LLaVA, Fuyu-8B
Going open-source provides businesses with many different benefits. We can name a few ones like flexibility, customization, privacy, security, and open collaboration (more on this later).
And this is not just about businesses. This is also for the regular person. If you're a developer (or a tech-inclined person) you can start making your own assistants, agents, or AI-powered tools.
I know and I've seen people online that are building prototypes without even knowing how to code.
In the same way that there are no-code/low-code tools for building websites and apps, there are also tools to allow us to work with LLMs in a graphical way.
Tools such as Flowise, Stack-AI, DataRobot, among others.
And when you start building with open-source AI, there's an additional benefit to talk about.
Tapping into the power of community
I mentioned Hugging Face before and I'll do it here again.
It is not only a platform with a suite of tools that are useful for machine learning and AI projects. It's also a thriving community learning, creating demos, and working together.
You go to the site's homepage, scroll down a little, and the first thing you'll see is a section showcasing all the models, datasets, and spaces that are trending.
Here's how that would look:
If you're not so much of a Python coder (like me) you can see what's getting popular and what you can try out with the open demos.
But if you do know Python, you can start trying models, run inference directly on each of their pages, and use them for fine-tuning or as a base for other projects.
Diving deeper you'll find other creators and AI enthusiasts that are building cool stuff with all these tools.
The open-source community is super welcoming and if you show willingness to learn and collaborate, you can be part of it and start building alongside super smart people.
Each one of us is in a different stage of the journey, so we can help and be helped by others. By sharing knowledge, resources, and lessons learned, we all can build better projects together.
In the history of technology, new advancements have been possible thanks to the collaboration of different people in different parts of the world. This is no different with AI.
Many more people and companies have started to turn to open-source AI since Meta announced their model called LLaMA (and then LLaMA2).
Companies going the Open-Source AI route
Companies have benefitted from using open-source software for many years already. But this time, the benefits are not only for the users, they're also for the producers.
It's not difficult to see that pretty much every company can benefit from integrating AI into their products, systems, or processes.
But what's overlooked in the "companies using AI" discussion, is how the ones that train and release models are benefiting from the open-source approach as well.
Stability AI was the first company that became known in the field thanks to Stable Diffusion. Their open-source alternative to DALL-E 2.
I previously mentioned Meta with their model LLaMA. They started making headlines for their different approach of not using a proprietary model on their products but instead releasing that model under a quasi-open license to the public
Other examples include the French company Mistral getting great recognition for the release of Mistral 7B. This is the first time we saw a small model outperform bigger, and more established models like LLaMA2 7B and LLaMA2 13B.
The main difference is that Mistral 7B is a truly open-source model because it was released under the Apache 2.0 license.
Another AI company called Adept appeared and impressed us with their AI that was able to take instructions and carry out tasks using the internet.
It looked like they were following the same path as Anthropic, doing a great demo of their product and then getting people to join a waitlist.
And then, almost a year after their first announcement...
They decided to open-source their first model called Persimmon-8B.
This is a model that powers part of their internal products.
A month later, they announced again that they were open-sourcing a multimodal model called Fuyu-8B.
Knowing that you can tell that going the open-source route has many benefits for companies working on AI projects. So much so that they are releasing their models for the community to use, test, improve, and provide feedback on them.
And it makes sense that those who have benefited from open-source tools like TensorFlow, PyTorch, or JAX, give back to the community.
Sharing their research and models with the community to enable greater innovation and development for all parties involved.
You know what they say, what goes around comes around.
A revolutionary and transformative technology such as AI can't and shouldn't be handled by only one company that decides what is wrong or what is right.
This is a technology that needs to follow a similar path that computers or the internet had. Providing access for every person under a common set of guidelines and restrictions.
The collaboration and knowledge sharing inherent in open-source AI is what drives further advancements in the field. This allows different kinds of businesses to leverage the latest innovations and create greater value with fewer resources.
However, the use of open-source AI is not only to get benefits in the monetary aspect. Or to get a competitive advantage over others.
It extends so much more.
Open-source AI not only benefits big or small companies, indie developers hacking away in their rooms, or researchers working in academia.
Open-source AI can benefit everyone.
From the bootstrapped founders looking to make the most of their limited resources.
Online creators looking to elevate their content game and spend less time creating high-quality content.
Marketers looking to speed up their workflows and get better results.
Knowledge workers looking to get work done at the job and have time for side projects.
Parents looking to spend quality time and have fun activities to do with their kids.
Students looking to study less and learn more to ace their exams.
And so on.
AI is for everyone and through open source we can make sure that we have the greatest technology in history to work for us instead of against us.
Thanks for reading.
PS: Check this to defeat distractions and improve your focus