build1 publisher
Open-weight moderation models land inside the proprietary error band on Bluesky posts
A seven-model comparison on real Bluesky posts reports open-weight sensitivity of 81 to 97 percent against 72 to 98 percent for the proprietary APIs, with the error direction reversing between rudeness and threats.
Publishers:arxiv.org
Reality
- Evidence58
- Adoption
- Insufficient
- Hype gap+30
- Incentives40
- Confidence55