Underdog Unleashed: How Open-Source AI Threatens Google's Dominance and Shakes Up the Industry
Author
Nathan GarzaTL;DR: The open-source AI community can achieve with $100 what Google struggles to do with $10,000,000. Meanwhile, Meta benefits from the open-source AI community's efforts, as the community builds on top of a "leaked" Meta AI model. If Google doesn't quickly pivot from enormous, time-consuming AI models to smaller, faster-to-develop models and doesn't start contributing to the open-source community, it will be left behind by competitors. Google's time-consuming approach to AI development is particularly harmful due to the brain drain from Google to OpenAI, which takes away the secrets that could have given Google a competitive advantage in AI.
A leaked memo from a Google engineer, titled We Have No Moat and Neither Does OpenAI, reveals concerns about the company's inability to compete with open-source projects in the AI race. The memo highlights how open-source AI models are rapidly closing the gap in terms of quality, speed, and customization. This poses a threat to restricted and paid models like those offered by Google and OpenAI.
“Open-source models are faster, more customizable, more private, and pound-for-pound more capable. They are doing things with $100 and 13B params that we struggle with at $10M and 540B [params]. And they are doing so in weeks, not months. This has profound implications for us.”
The engineer emphasizes that open-source models have made significant progress in recent times, thanks to the affordability of large language models (LLMs) and a vast outpouring of innovation. This has led to breakthroughs including running foundation models on smartphones, scalable personal AI, responsible release, and multimodality.
“We have no secret sauce. Our best hope is to learn from and collaborate with what others are doing outside Google. We should prioritize enabling 3P [third-party] integrations.”
The memo suggests that Google should focus on collaborating with open-source projects and learning from their advancements to stay competitive in the AI field. It also argues that maintaining large AI models may be a disadvantage, as smaller models can be iterated upon more quickly. The engineer further recommends that Google should embrace an open-source approach and establish itself as a leader in the open-source community.
“Giant models are slowing us down. In the long run, the best models are the ones which can be iterated upon quickly. We should make small variants more than an afterthought, now that we know what is possible in the <20B parameter regime.”
The timeline provided in the memo details the significant advancements made in open-source AI from February to April 2023. These include the launch of Meta's LLaMA, the introduction of fine-tuning on a laptop, the development of models that achieve "parity" with Bard, the creation of open-source GPT-3, and the implementation of multimodal training in one hour.
The graph below illustrates that, when using GPT-4 as a judge, Vicuna-13B (open source) achieves over 90% of the quality of OpenAI's ChatGPT. Vicuna-13B was released just three weeks after the Meta's LLaMA-13B leak. The cost of training Vicuna-13B is approximately $300.
In the leaked memo, the engineer also notes that directly competing with open-source projects is a losing proposition due to their significant advantages. For instance, open-source projects benefit from a global network of researchers, individuals who are not constrained by licenses, and creators who deeply understand specific use cases.
“Most importantly, they [open source community] have solved the scaling problem to the extent that anyone can tinker. Many of the new ideas are from ordinary people. The barrier to entry for training and experimentation has dropped from the total output of a major research organization to one person, an evening, and a beefy laptop.”
To leverage the open-source ecosystem, the memo suggests that Google should follow the example of Meta, which has benefited from owning the platform where innovation happens. By owning the ecosystem, companies can shape the narrative on ideas and retain their position as thought leaders.
Regarding OpenAI, the memo claims that their closed policy and the poaching of senior researchers make secrecy a moot point. The engineer believes that OpenAI will also be eclipsed by open-source alternatives unless they change their stance.
In conclusion, the leaked memo highlights the rapid advancements in open-source AI projects and the need for Google to adapt its approach. By collaborating with and learning from open-source projects, Google can maintain its competitive edge in the AI landscape and avoid being left behind by unrestricted and innovative alternatives.
