Recently, Yann LeCun has argued that Meta is becoming less committed to open research. I was curious whether there was a small, measurable signal that might reflect this idea.

I analyzed every repository in the facebookresearch GitHub. Since GitHub's API doesn't expose archive dates, I scraped the archive banner from repository's page and grouped repositories by the year they were archived.

This isn't a definitive measure of openness —repositories are archived for many legitimate reasons— but archival trends can still provide a useful historical signal.

The results show a striking spike in repository archival during 2023, followed by elevated levels in the years after. On its own, this doesn't prove that Meta became less committed to open research, but it does raise an interesting question:

What happened in 2023?

Was this simply a large-scale cleanup of aging projects, or does it reflect a broader shift in Meta's research strategy and its approach to open source?

Here's the code: https://gist.github.com/mcavdar/85476f804f20ab072fb1a75353d70782