The Rise of Open-Source Generative AI: Power to the People - Part 1

Artificial intelligence has long been the exclusive playground of tech giants and well-funded research labs, especially when creating the most advanced models. Building sophisticated AI systems, especially large language models (LLMs), demands immense resources—access to vast datasets, cutting-edge computing power, and deep technical expertise. However, a significant shift is occurring: the tools and models emerging from these resource-intensive efforts are increasingly being shared as open-source projects. This means that while the initial creation might be in the hands of a few, the ability to innovate, modify, and build upon these foundations is now available to many.

In this two-part series, I’ll explore how open-source generative AI models are shaping the future of artificial intelligence. In Part 1, I’ll dive into the true meaning of 'open source' in AI, guided by the OSI’s newly defined standards. I’ll discuss what it takes for a model to be genuinely open and why these standards are crucial for fostering innovation. In Part 2, I’ll take a closer look at some of the most prominent open-source large language models (LLMs), evaluating how they measure up against these standards, along with their capabilities and the challenges they present.

What Does 'Open Source' Really Mean for AI?

Ah, the term "open source"—that phrase that conjures images of community-driven innovation, where developers collaborate freely and code flows like water. But what exactly does "open source" mean in generative AI? As it turns out, not everything that claims to be "open" is as accessible as it seems. Enter the Open Source Initiative (OSI), the organization dedicated to upholding the principles and integrity of open-source development.

Founded in 1998, the OSI is responsible for maintaining the official definition of "open source." Historically, this has applied to software, but in 2023, they expanded their scope to include AI models, crafting the first-ever definition of open-source AI. And they didn’t do it alone—they brought in a diverse group of 70 experts, including researchers, lawyers, policymakers, activists, and representatives from major tech companies.

This definition is significant because, as outlined in the image above, which highlights the eight key features of open-source AI, some companies have been overly generous with the term "open source." The real question is: how many of the existing open-source LLMs meet all these criteria? To address this, the OSI plans to implement an enforcement mechanism to ensure that models marketed as open-source truly adhere to these standards.

OSI also intends to release a list of AI models that pass the test—though we’re still waiting for that rollout.

Why does this matter?

Clearer standards make it easier to distinguish between models that genuinely embrace the principles of openness and those that merely claim to. As the image above outlines, the more a model adheres to these key features—transparency, accessibility, and shareability—the more it encourages innovation. However, as anyone working on building and deploying LLM-based apps and products knows, not all models live up to these ideals.

In Part 2, I’ll take a closer look at some prominent large language models (LLMs) and evaluate how well they meet the OSI’s definition of open-source, along with their capabilities and the challenges they present.

Stay tuned as we dive deeper into which models truly live up to the open-source promise—and how these models could impact the future of AI development.