Executive summary : Open-source AI models are emerging as viable alternatives to paid platforms like OpenAI's ChatGPT, offering users access to powerful tools without recurring fees. While these models present opportunities for independents, small businesses, and developers to reduce costs and increase efficiency, you must consider the fine print. Decision-makers must carefully assess these models' long-term viability and risks before fully committing.
OpenAI's ChatGPT, Google's Gemini, and Anthropic's Claude are in an arms race to be your go-to platform for all things AI. They also want your money, so while you can use many for free, the capabilities have limitations, with all the good stuff starting at $20/month/user.
📈 However, the monthly fees do not stop there. Want to create audio, video, or images with an integrated suite of editing tools like Adobe Creative Cloud? How about information worker tools like Google Workspace or Microsoft 365? If I take the base plans for some of these products, you are paying over $182/user/month.
💸 Have you downloaded an app that has AI built-in? Maybe a personal assistant or a tool to help you write? All those apps have to pay small fees for every interaction you have with the app, which adds up. You might be paying $35/month, but every time you do something in the app, the software developer has to perform micro-transactions with various AI companies that eat into their profit.
AI is a world-changing technology we hope to make available to everyone. But as you can see, the costs can become increasingly limiting for independents, small businesses, and, in some cases, much larger organizations.
🦸🏽 Open source comes to save the day!
Open source promises to give you ChatGPT-like superpowers without the monthly fees.
Free, open-source AI models are here to free you of monthly fees and empower you with the latest AI technology!
Open-source software (OSS) generally refers to software apps or services that are freely available. The code can be inspected and perhaps even changed to meet your needs. OSS encourages collaboration and innovation, as developers worldwide can contribute to improving and customizing the software. OSS is typically free to use, allowing users to tailor the software to their needs, fix issues, and share their changes with others, fostering a community-driven development model. In the collaborative and free nature of OSS is a bit more opaque, but more on that later.
Here are the top 3 open source AI models you will see frequently in the news. I refer to them as foundational models because they contain world knowledge and are not designed for one specific task. However, they are also referred to as Large Language Models (LLMs) or SLMs (Small Language Models):
💡
Editor's note: There are thousands of open-source AI models. I will focus on foundational chatbot-style AI models for the remainder of this article.
🌟 A new world of opportunities
These open-source alternatives allow you to download small, medium, and large versions, each with progressively more knowledge and skills. You can do this today for free (assuming you don't pay for bandwidth) and use them locally on your computer!
These models offer thousands of use cases and opportunities. Here are three:
- Independents and small businesses can use chatbots on their computers and avoid paying monthly fees. I use them to summarize long documents and compose social media posts.
- Companies with internal, custom-built software can add AI features at a fraction of the cost of a paid model.
- Software developers can add these smaller models into their apps without paying fees to the larger companies.
🎉 Rejoice! Open source saves us!
With open source, we, the people, have access to the advanced capabilities only the big entrenched players have. We are unshackled from corporate lock-ins and can generate any content we want, any time we want!
Right?
🤔 Wait a minute...
If you are excited to freely chat, build apps with superpowers, make your business more efficient, and reduce monthly costs, you should be! And you might want to ask yourself these questions:
"Why would Google, Microsoft, and Facebook be our open-source saviors when they have a vested interest in profiting from me?
Are these models just another way to lock me in to their platforms?"
Those are important questions to consider before jumping head-first into using these products. This article could be a very long series, so I am going to give you high-level answers to the following questions:
- What is the definition of free with these open-source models?
- If I like the open-source model, can I use it to create apps for my business or sell on open markets?
- Should I assume these models will receive regular and frequent updates with new and refreshed "knowledge"?
🔍 The definition of "free" with foundational AI models
Is open source software truly "free"?
In what I will term traditional OSS (open-source software), a person(s) creates software to solve a problem and makes the code free of charge on an OSS platform.
For example, if you have ever wanted to play MP3 files, DVD movies, or other media, you may have used the popular VLC Media Player. You can visit GitHub, a site Microsoft owns, to see its code repository (a repo). There, you can read up on its latest innovations and see what the vibrant community of contributors is planning for new versions.
While you often find people writing software code on GitHub, they create and test AI models on another platform—with the cutest name—Hugging Face. Here is OpenAI's ChatGPT (GPT-2) model.
Open source does not mean license-free
💡
Now would be a good time to let you know I am not a lawyer, so while I researched all my statements to the best of my ability, I may be misspeaking.
Now that you know there are popular places to build, contribute, and share software and AI models, you should also know that is not a requirement. Anyone who builds open-source software can share as much or as little as they want. They can even add licenses that protect the software.
Here are three popular licenses you might see from the OSS community:
- MIT License : We call this a permissive license because it allows anyone to freely use, modify, and distribute the software, usually requiring or requesting some attribution to the original author.
- GNU General Public License (GPL): We call this a copyleft license because sometimes people will take your open-source code, modify it, and then sell it for a price. A GNU license ensures that modified software versions are open source and distributed under the same GPL terms.
- Apache License 2.0 : A permissive license similar to MIT but includes more explicit terms for patent rights and contributions, allowing users to use and modify the software freely.
Like the rest of the OSS community, people and companies that release open-source AI models can choose to use any license they like. While the Apache License 2.0 is popular, here are two others the AI community appears to be adopting:
- Creative Commons (CC BY 4.0): This license permits others to share and adapt AI model data and outputs as long as credit is given to the original authors.
- OpenRAIL (Responsible AI License): This license is explicitly designed for AI models; it allows for free use and modification but includes clauses to ensure ethical use and prevent harmful applications of AI technology.
Open-source foundational AI Models and licenses
Always review the terms of use on a digital device for open-source AI models.
As I mentioned at the outset of this article, Meta (the Facebook and Instagram company) offers an open-source foundational AI model called Llama. It allows you to install something equivalent to ChatGPT's text prompting on your computer. As a software developer, you can create apps that use Llama to enhance user interactions.
While people in the AI space refer to Llama (and Google's Gemma and Microsoft's Phi-3) as an open-source solution, I would say they live in a gray area. You will not find the source material used to train the model, all the algorithms they use, or the code used to create the model. In that sense, the "open source" part of the product is the output of Meta's software code, not the code itself.
Peeling the onion back and reading Meta's Llama 3 Community License Agreement, you will find it is a worldwide royalty-free, limited license. Some of the limitations require you (the licensee):
- 1.b.i.(B): Prominently display “Built with Meta Llama 3” on a related website, user interface, blog post, about page, or product documentation
- 1.b.V: Will not use the Llama Materials or any output or results of the Llama Materials to improve any other large language model (excluding Meta Llama 3 or derivative works thereof).
- 2: Additional Commercial Terms. If, on the Meta Llama 3 version release date, the monthly active users of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights.
Google's Gemma is much more specific in its restrictive uses. For example, you cannot:
- 1: Generate any content, including the outputs or results generated by Gemma or Model Derivatives, that infringes, misappropriates, or otherwise violates any individual's or entity's rights[...]
- 2: Perform or facilitate dangerous, illegal, or malicious activities[...]
- 3: Generate and distribute content intended to misinform, misrepresent, or mislead[...]
To my knowledge, Microsoft's Phi-3 model is under a very permissive MIT license, which I mentioned earlier in this article.
Should I be worried about these licenses?
In my humble opinion, the licenses are not too restrictive. I called out the Llama 3 license because Meta restricts the number of users (700M is a considerable number, though). More importantly, the output cannot improve another large language model. That second part is a little disconcerting.
What happens if you prompt Llama with information and then paste that into ChatGPT, and the setting allows your prompts to train ChatGPT? Does that mean you just broke the license? I think, technically, the answer is yes. Will the lawyers come for you? I don't think so. But, if you are a researcher or developer using multiple AI models and need them to interact with–and learn–from each other. I could see that as perhaps breaking the legal agreement.
Or not. Remember, I am not a lawyer.
Okay, Bill, I read the licenses, but is there anything else I should be worried about?
Mind you, this article is not about all AI open-source models, just the big-name ones. Meta, Microsoft, and Google care most about software developers using these tools. While not said aloud, one could assume that these models directly respond to threats by the names of ChatGPT and Claude.
If they do improve these models to the point where they chip away at the market dominance of some of these other models, then what would be the point of giving them away? Why not sell their competitive models (like Google already does with Gemini, by the way)?
Currently, these models are pretty up to date with how we use our language, so their responses feel fresh and "now." But what terms will we use a year from now? Three years from now? If you invested in these models and they do not remain fresh with new versions and content, that great app you built inside your company or for your startup might seem stale and archaic.
With every release of a model, the company building them has the right to create new licenses. Maybe their models remain open source, but their licenses become more restrictive and even require you to pay a percentage of the profits you are making from the use of the model. Maybe these licenses require you to pay the creators a percentage of your earnings. What happens if the model outputs material you never thought would result in someone's injury, and your business is now responsible?
All of this could come true. Or not.
Open-source models as SaaS on the cloud
If you are a software developer and want to offer an app, there is a good chance you will not download a multi-gigabyte OSS model onto every device where your product runs. Instead, you will host the model on so-called cloud platforms, like Microsoft's Azure, Google's Cloud, or Amazon's AWS.
Of course, many more cloud platforms are out there. Still, the idea is these smaller OSS models might be ideal for up-and-coming startups or organizations that need an AI model with reasoning abilities but do not need to pay for high-end services like ChatGPT or Gemini. The good news is Google, Microsoft, and Amazon will offer these models (and more) at a cost. Sometimes, the price can be lower than the more prominent players, so there is an incentive to offer varying price points and capabilities.
The idea that you and I can download a powerful model to our computers and interact with it, just like ChatGPT, is great for everyone. If we want to use them for more advanced applications, we will probably end up paying the entrenched cloud providers, who will offer us options at different price points.
For example, if you compare Microsoft's Phi-3 pricing to Microsoft's ChatGPT pricing, you will notice that a software developer can save a significant amount of money using Phi-3. However, that does not mean Phi-3 is a better choice since ChatGPT offers a wider array of services. Or maybe Phi-3 is lean and mean and meets all their needs. This is why competition is good, and I hope more foundational OSS AI models find their way to the market and become available as services on cloud platforms.
Carefully weigh your open-source AI decisions
Carefully weigh your open-source vs. closed-source decisions.
I love the open-source community and think this new wave of open-source AI models is exciting. I hope the community thrives and there is more engagement in the open development of the tools. I also hope you carefully weigh the decisions you make.
Dr. Deborah Chen provided editing services for this article.
Top comments (0)