back
5 comments
Meanwhile there is the 'Foundation model transparency index' from Stanford https://crfm.stanford.edu/fmti/
"In 2024, the OSI will release a new draft of the open-source AI definition monthly, based on bi-weekly virtual public town halls. 'Our goal is to have a 1.0 release by the end of October', Maffulli said. Everyone is welcome to partake in the discussions regarding the drafts in OSI's public forum."

Excellent, I am so happy to hear this.

I feel like to be truly considered OSI-compatible open source, you’d need three things: 1. Model code license (MIT, GPL, etc.)

2. Model weights license (CC?)

3. Model training code license (MIT, GPL, etc.)

I don’t think anything new needs to be invented, since AI is simply data and code, and we always have licenses for these. But an “Open Source AI” really needs proper open licenses on every critical component otherwise it’s only partially open source.

gpt 4, a closed model that exhist only as an api, is ahead of stable diffusion 2, a model with the traning code, dataset and weights downloadable, ok.

I understand there are some other metrics in the mix, but the flat average makes absolutely no sense here.

Many "AI"s won't be small to be interesting. Most will need, very probably, huge hardware + training data access.