The open-source community is said to be winning the battle for AI development. This is stated by a senior Google engineer in a paper that has come out. “Plainly put, they are lapping us.”
The document reveals that Google has no secret weapon in the fight for AI. Since the release of AI chatbot ChatGPT in late 2022, the arms race around generative AI has flared up. The two superpowers: Google on the one hand, and Microsoft-backed OpenAI on the other. Soon, Google went after ChatGPT with Bard, while the Redmond giant is busy adding artificial intelligence tools to products like Office 365.
Although Meta and numerous smaller tech companies are also developing large language models (LLMs), experts consider the contention between Google and OpenAI to be the main battle in this area. However, the Google employee sees that the open source community has already solved many problems that have yet to be addressed at the two AI giants. For example, LLMs can already be run on phones. OpenAI and Google, with their large models, rely on a huge amount of expensive hardware to handle advanced AI models.
Google and OpenAI are both very cautious about sharing information about their AI models. Since the advent of GPT-4, OpenAI has even stopped sharing the number of parameters the model is based on. It is rumored to be 1 trillion. PaLM, Google’s most advanced model, has 540 billion. According to the Google engineer, this format is precisely a stumbling block for both of them: where the two AI giants take months to develop new versions, open source evolves at lightning speed.
This forces the companies to take a different path: go open-source themselves. OpenAI and Google have not yet reached that point, but Microsoft has enabled organizations to integrate generative AI into an existing application on an open-source basis. In addition, Meta has chosen to make its extensive LLaMA available for research. The open-source community has now appropriated this AI model since it became accessible to all via a leak in early March.
The release of LLaMA led to an “outpouring of innovation,” according to the Google employee’s document. Although the LLM came with no instruction set or chatbot feature, the open-source community filled it out at lightning speed.
This begs the question of what Google and OpenAI/Microsoft have to do in the AI arena. Obviously, we are talking about two tech giants, each with its own ecosystem of applications. Where AI in open source has all kinds of applications, the existing infrastructure of Google and Microsoft apps, respectively, offers countless applications with a gigantic installed base of users. This cannot be easily bridged by open-source projects.
Another point where the big companies are fighting a different battle than open-source revolves around legislation. Where Google and Microsoft can be held accountable for their approach to AI, this is much less practical with the inherently decentralized open-source community. Yet it is also happening in that corner. Recent legislative proposals from the EU, for example, have caused much discomfort at the Python Software Foundation. It would like to hold developers accountable for software vulnerabilities. In addition, AI legislation could cause AI development from the open source community to slow down tremendously.
All in all, Google and Microsoft certainly have assets to continue to fight the AI battle. Both parties will also certainly have to keep an eye on what open source has to offer. Chances are that the innovations in this area are moving so fast that even those kinds of tech giants won’t be able to keep up. By nature, there is always more development power in a (large) open source community than in a single company.