@nicolas17@social.treehouse.systems
2026-10-01 02:10 UTC
@androcat@toot.cat @markmetz@sfba.social which part of mammography involves "language"? 🤨
Replies (1)
-
@gsuberland@chaos.social 2026-10-01 02:26
@nicolas17@social.treehouse.systems @androcat@toot.cat @markmetz@sfba.social the language part isn't a factor. the idea is usually to use a VLM, or at least the part that turns an image (or video) into a sequence of tokens, then train a classifier model on that token sequence. it's a genuinely workable design in theory, but the externalities of gaining training data, giving it to the orgs with sufficient equipment and capital to build the model, them having sufficiently trustworthy stewardship of the project, etc. make it a non-starter.