In the past few months, I’ve written about the need for regulation for the companies involved in building the artificial intelligence models that (will) power most of the apps you use today. That was from an ethical use perspective so that AI models are not used for nefarious purposes such as undressing people using their photos and videos. It was also a call for making AI models open-source so that there is community oversight on the development process, the data sets used and the capabilities being developed. Big tech companies are mostly not interested in open-sourcing their own AI models. There appears to be a philosophical divide among the big tech companies on the need for open source.
On one end, we have companies such as OpenAI which does not open-source their main AI models which drive the product that we all know as ChatGPT. They do open-source some models such as Whisper, which is used for transcribing and translating, and Evals, a software framework for evaluating the performance of AI models. Their entire valuation is based on the GPT-3, GPT-4 and upcoming models. So there is no way they will open-source those models because OpenAI as a company has nothing else to fall back on. Microsoft, which uses OpenAI’s models for powering Copilot and even key products of their subsidiaries, would also not want OpenAI’s models to be open-sourced.
Meta (previously Facebook), on the other hand, has partially open-sourced their latest and most powerful Llama 2 AI models because it believes in opening up the models to the community and reaping the benefits of community innovations. I say “partially” because you have to see a license from Meta if your user count goes above 700 million. Essentially, big tech is prevented from using Llama 2 models but small companies can use it. The other rider is that you cannot use Llama 2 to improve other models. So only Llama 2 can be improved using derivatives of Llama 2. It’s not completely open-source but it is way better than the approach taken by OpenAI. It’s very understandable that Meta chose to partially open-source Llama 2 because its cash cow is the advertising business that relies on Facebook. OpenAI, unfortunately, has nothing else that it can rely on.
Then there are companies such as Mistral which have completely open-sourced a lot of their models without any riders. Their newer models are not available for download but only via their own API which isn’t exactly open source. So we can see a varying degrees of the open-source philosophy being applied by different companies. There are big tech companies that publicly talk about open-source as the best thing since sliced bread but behind closed doors they hate it for the fact that it threatens major revenue sources. This has resulted in the formation of two blocs of parties duking it out over open-sourcing. This tussle between AI companies has been going on for a while but it has not reached a pivotal point because they have started lobbying various government agencies. It’s mostly the big tech companies which are leading the lobbying efforts since they have the deep pockets to move pillars within various tiers of the government and it’s bad for the open-source community. And in extension, for the AI industry itself.
Open-source software form the bedrock of innovation in Silicon Valley and that holds true for the AI companies as well. They allow researchers, developers, and enthusiasts to collaborate, iterate, and enhance the AI technologies. Meta is literally banking on the community to help innovate and improve Llama 2 models. So when big tech lobbies against open-source AI models, it threatens to centralise all the innovation within their little walled gardens. Moreover, should they actually prevail, then cutting-edge AI becomes available only to the privileged few. So in the coming months, you will hear horror stories of how AI is being used with malicious intent. In fact, you’ll see more and more stories till it peaks and major governmental bodies have to sweep in and set things straight. That’s when you’ll know that these lobbying efforts have borne fruit. That’s when people need to raise their voices or forever bid goodbye to the idea of democratised AI.
Leave a Reply