So it’s no surprise that the open-source-is-what-OSI-says-it-is group (note: I’ve historically been in this camp) criticizes Meta for labeling its LLM large language model (LLM) as open. source, the restrictions nevertheless do not meet the OSI definition of an open source. The industry response was a collective shrug. See, for example, “Why LLaMA Meta Models Are Open Source” – a title that must drive the OSI people crazy. Part of this stems from, as one HackerNews commenter puts it, the idea that “Meta, through the Llama models, has done more for open source LLM than anyone else.” Going back even further, open sourcers can look at Apache Cassandra, React, GraphQL, PyTorch, and other Meta projects that raise the OSI bar for open source.
It’s hard to be too grumpy about a company that has created some of the most important open source projects in the industry.
And yet some people are very grumpy, despite the fact that there was (and is) no fixed definition of open source in AI. Yes, OSI has finally released an open source definition for AI, Open Source AI Definition 1.0, but as with the cloud, OSI is playing catch-up, and its definition has disappointed some of its staunchest supporters by not dictating it. training data also open.