The Winner of Artificial Intelligence is Open Source

The Open Source Ranking is a parameter that shows the rapid involvement of developers in the field of large-scale language models. According to the data from the Hugging Face ranking, different versions of Vicuna and Meta's Llama-2 have been downloaded millions of times in the past month. Meta's Llama-2 is a set of pre-trained and fine-tuned generative text models, ranging in size from 7 billion to 700 billion parameters, and has been downloaded over 3 million times. The Large Model Systems Organization (LMSYS) has released 17 different versions of the Vicuna model (7B and 13B), with slightly over 1 million downloads. This indicates that the biggest threat faced by major suppliers like OpenAI, as opposed to large tech companies, comes from the open-source community. Meta has taken an astonishing lead in embracing open-source compared to Google and OpenAI, which either keep most of their technology behind closed doors or make it available only through partnerships with specific companies. The significant improvement in Meta's reputation has been mentioned multiple times recently. One reason why the aforementioned companies are reluctant to expose their technology to the public is that they consider these technologies as products rather than just internally developed infrastructure. Their ultimate consideration in building this technology is the interests of shareholders, not developers. Meta's statements When the initial version of Llama was leaked, Meta sent takedown requests to GitHub and Hugging Face to regain control. However, as the code spread widely on the internet, Meta abandoned its attempts. Instead, the company accepted this fact and chose it as a direction for future development, subsequently releasing more models. The power behind the company's open-source awakening may be its AI leader, Yann LeCun. Although AI is firmly rooted in open-source principles, Llama represents a milestone as the first major open-source benchmark model. Meta's Simon has been actively advocating for open-source in online networks and communities, paying attention to this topic every day. From retweeting tweets about "keeping AI open" to responding to discussions with other AI experts, LeCun does not hesitate. Just a few hours ago, LeCun replied to MIT professor Max Tegmark, saying, "Like many others, I strongly support open AI platforms because I believe it's a combination of many forces: people's creativity, democracy, market forces, and product regulation." In addition to providing access to the Llama model, Meta also shares its weights, which other major language models do not. Weights represent the parameters the model acquires during the training process, simplifying the development and execution of AI algorithms. In contrast, other GPT models are only accessible through application programming interfaces (APIs). Meta may have surpassed OpenAI and Google. Recently, an internal Google memo circulated online, in which a Google AI engineer referred to the open-source community as the "third faction quietly shutting us out." Economic value Now even investors are discussing open-source. Wei Lien Dang, Managing Partner at Unusual Ventures, said, "If you go back nine months, you wouldn't see open-source alternatives challenging OpenAI and some leading proprietary solutions." "This phenomenon has significantly increased." Clearly, investors have chosen open-source as a reliable competitor. Even James Currier, Managing Partner at NFX, noticed the significant cost-saving advantage of transitioning from closed-source to open-source models. In his portfolio, a company previously had to spend $150,000 per month to access a specific model. However, after adopting an open-source alternative, the startup's operating costs were greatly reduced, with the same model costing only $4,000 per month. To some extent, Meta and LeCun should be recognized for their inclination towards open-source language models. A month ago, LeCun tweeted that AI systems are rapidly becoming part of infrastructure. He also pointed out that historically, infrastructure has often existed in open-source form, such as the software infrastructure of the internet, Linux, Apache, and JavaScript browser engines. Although the company continues to invest in its technology, the open-source community is more likely to emerge victorious in the end.