Zuck's strategies behind open source

Posted on Edited on

Random words:

Music teacher never answered my question: why should Triangle be included in a piece of music while there are already 10+ instruments and sounds loud in there? I accidentally got the answer from my dance teacher. She said: “different people have different hearing capabilities and thus different understanding of music, what dancers are doing is actually to interpret or reproduce music.”

Back to the topic:

There is always a lot to learn about strategic thinking from Zuck. Here are some of his smart strategies behind open source I’ve learned:

  1. According to this interview, Zuck’s point of open source is to avoid concentration of AI while he didn’t ignore the harmful consequences of open source saying that it’s our responsibility to do a good job of reducing harm. There are several benefits of open source, one is that people could figure out cheaper ways to develop models so it won’t cost too much resource. The other benefit is that they could enable more efficient developments and vertical use cases in a lot of different systems. Take Google and Apple for example, their mobile ecosystems restricted what developers could build or what features they could launch on them.
  1. For companies like Meta with well-established network effect, they really don’t need to have the best model. AI’s content creation potential benefits Meta’s platforms, even if the models are not exclusively theirs. This is the most reasonable reason from business perspective and was stated in an earnings call.
  2. By open-sourcing models, Meta started developer communities which can contribute to whatever the ecosystem Meta built and help solidify the advantage of it. Most recent example is the open model of Horizon OS which powers its VR headsets. It allows developers and creators to take advantage of these technologies to create MR experiences and grow business on it. Then Meta Quest Store can be quickly established.
  3. Models themselves are not a moats. Moats are built through data and habits. Open source eventually makes economic value of foundational model disintegrated. There will be no value in foundational model economically and there is probably less point for VC to plow billions of dollars into a foundational model development startup. The potential economic values are in 100K+ developers iteratively and quickly training and deploying the open source models for specific business use cases. Inference will be way more important than training. So attention and money will be less concentrated to products like OpenAI GPT series and Nvidia training GPU but more on Cloud platforms with inference GPU for personal and business usage.