Artificial Intelligence Needs Open-Source Models to Reach Its Potential
Johannes Gutenberg’s invention of the printing press transformed the world in the 15th century, allowing ideas to spread rapidly and fueling immense progress for humanity. However, this revolutionary technology was nearly kept as a closely guarded secret. Fortunately, a former investor, Johann Fust, replicated Gutenberg’s press and even poached his top techie, Peter Schoeffer. This turn of events ensured that the printing press was not confined to the control of a few individuals or nations, allowing for centuries of progress.
Drawing a parallel to the current debate surrounding artificial intelligence (AI), it becomes clear that generative AI, which can produce text, images, and other media, possesses the potential to revolutionize the 21st century even more significantly than the printing press. Generative AI’s ability to process vast amounts of human knowledge through large-language-model algorithms empowers it to write, draw, reason, and problem-solve. These powerful tools have the potential to enhance the capabilities of knowledge workers and enhance productivity, ultimately generating substantial economic value. As someone who has been involved in AI for over four decades, I am confident that generative AI will unlock an unprecedented technology revolution.
However, it is essential to ensure that this power is not monopolized by a privileged few. Considering the impending technological advancements, it becomes crucial for individuals from diverse backgrounds to understand and have access to this transformative technology. To foster inclusivity, I have launched 01.AI, a startup that develops foundational large language models, the building blocks of generative AI. One of our models, Yi-34B, with its 34 billion parameters, is available on GitHub for anyone to engage with, enhance, and tailor to their needs.
While 01.AI aims to make its models accessible to researchers, entrepreneurs, and smaller companies, it is important for major players like OpenAI and Google to contribute to the open-source community while pursuing their business goals. By open-sourcing smaller models, these tech giants can provide accessibility to a broader range of users, including researchers, educators, students, hobbyists, and nonprofits. Embracing openness democratizes generative AI and prevents certain groups from being marginalized due to the tools’ proprietary nature.
However, one significant challenge is the American bias in dominant proprietary models. These models, trained primarily on American and English data, fail to adequately perform when it comes to languages less prominent on the internet. As a result, users from smaller or poorer nations experience a vastly inferior AI experience. To address this issue, it is crucial that each country has access to high-quality native language models tailored to its culture, values, religion, and language.
Some describe 01.AI as China’s answer to OpenAI, but we view ourselves as the more open alternative. The true competition lies between open and closed systems, not between nations. Despite limited resources, we are committed to developing high-quality models for multiple languages, ensuring that this transformative technology is accessible to individuals worldwide. Our goal is to prevent AI from leaving anyone behind.
As a firm believer in the potential of artificial intelligence, I am convinced that it can amplify our humanity rather than replace it. However, this can only be achieved through a steadfast commitment to openness. By embracing open-source models, we can unlock the true potential of AI and ensure a future that benefits all of humanity.
