OSI readies controversial Open AI definition
AI summary of the linked articleThe Open Source Initiative's board was scheduled to vote on the Open Source AI Definition (OSAID) on October 27, with version 1.0 slated for publication on October 28, and critics say the draft sets the bar too low. The current draft requires model parameters and detailed information about training data, but not the training data itself. Tom Callaway argues that making data optional devalues the meaning of "open source" in all contexts, while OSI executive director Stefano Maffulli says the definition reflects input from a wide range of stakeholders. Stephen O'Grady argues that the open source term should not be extended to AI at all. |
 |
Recommended by 2 curators |
2 comments  Wow nature is healing!! This is what I said too, when I argued that Mark Zuckerberg is lying to the public, when he calls LLama open source. Because Meta isn‘t sharing the training data/process with the public and an open LLM is vastly different from open source code. If they just publish the model weights, you‘re still reliant on Meta to continue training the model and releasing the new weights for free in the future  |
