Data is the bottleneck of Open Source AI. What stops you from describing yours? #80
francescobianco
started this conversation in
Open Source AI
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Both Open Source AI Salons, in Raleigh and in Barcelona, ended on the same point: data is the bottleneck. According to OSI, relatively few models meet the OSAID's data information requirements.
The barriers named in Barcelona are concrete ones: privacy, limited resources, few people with provenance expertise, the risk that someone downstream mishandles the data, and no shared agreement on what "open data" even means.
If you've released a model, or tried to: what stopped you from describing your training data? And what amount of data information would you consider honest and realistic: the full dataset, a reproducible recipe, or a detailed datasheet?
All reactions