The Rise of Open-Source AI Models: Google and OpenAI Face a New Challenger

Highlights

---

• A leaked document reveals that Google and OpenAI are facing competition from the open-source community in developing faster, cheaper and smaller AI models.

• The leaked document titled "We Have No Moat and Neither Does OpenAI" was written by a senior Google engineer, Luke Sernau, who warns that the open-source community is quietly eating their lunch.

• Google and OpenAI have been too focused on competing with each other, while ignoring the third faction that is the open-source community.

• Sernau argues that the open-source community is using publicly available data and tools to create AI models that are faster, more customizable, more private, and more capable than Google's and OpenAI's offerings.

• LLaMA, a 13-billion-parameter model that was leaked to the public in March 2023, has emerged as a smaller but efficient alternative to Google's Bard (540-billion-parameter) and OpenAI's ChatGPT-4 (560-billion-parameter) models.

• Sernau suggests that Google and OpenAI should learn from and collaborate with the open-source community and prioritize third-party integrations to make their AI models more accessible and user-friendly.

---

Introduction

Google and OpenAI are two of the most prominent players in the field of artificial intelligence (AI), especially when it comes to large language models (LLMs) that can generate text, images, and other content from natural language prompts. However, a recent leaked document from a senior Google engineer reveals that both companies are facing a serious challenge from the open-source community, which is rapidly developing and improving smaller, faster, and cheaper AI models that rival or surpass the quality and capabilities of Google's and OpenAI's products.

The leaked document and its author

The document, titled "We Have No Moat and Neither Does OpenAI", was written by Luke Sernau, a senior software engineer at Google, and circulated internally in early April 2023. It was then leaked by the consulting firm SemiAnalysis on May 5th 2023. In the document, Sernau warns that Google and OpenAI have been too focused on competing with each other for the AI arms race, while ignoring the third faction that is "quietly eating our lunch": the open-source community.

The challenge from the open-source community

Sernau argues that Google and OpenAI have no "secret sauce" or competitive advantage that sets them apart from the open-source developers, who are using publicly available data and tools to create AI models that are faster, more customizable, more private, and more capable than Google's and OpenAI's offerings. 

The lack of competitive advantage for Google and OpenAI

He cites several examples of open-source models that have emerged and improved after Meta Platforms Inc. LLaMA (Large Language Model Meta AI), a 13-billion-parameter model that was leaked to the public in March 2023. LLaMA is much smaller than Google's Bard (540-billion-parameter) or OpenAI's ChatGPT-4 (560-billion-parameter), but it can still generate high-quality content with less data, computing power, and time.

Lowered barrier to entry for AI model training

Sernau says that LLaMA lowered the barrier to entry for training and experimenting with AI models from the level of a major research organization to one person, an evening, and a beefy laptop. He also says that LLaMA spawned several iterations and variations from the open-source community, such as Alpaca-13B and Vicuna-13B, which improved on LLaMA's performance and features. 

Open-source models outperforming Google and OpenAI

Sernau claims that these open-source models can do things with $100 that Google struggles with at $10 million, and that they can run on devices like Google Pixel 6, which Google and OpenAI still consider a major open problem.

The need for collaboration and adaptation

Sernau concludes that Google and OpenAI are not positioned to win the AI race, and that their best hope is to learn from and collaborate with what others are doing outside their organizations. He also suggests that Google should prioritize third-party integrations and make its AI models more accessible and user-friendly. He warns that if Google and OpenAI do not adapt to the changing landscape of AI, they will lose their relevance and market share to the open-source community.

Conclusion

The leaked document by the Google engineer sheds light on the growing influence of the open-source community in the field of artificial intelligence. The document serves as a reminder that even the largest players in the AI industry are not immune to disruption from smaller and more nimble competitors. It highlights the need for companies like Google and OpenAI to be more open and collaborative with the wider community, and to focus on making their products more accessible and user-friendly. The AI landscape is rapidly evolving, and those who fail to adapt risk being left behind.