Online Safety Bill: How Global Platforms Use MLOps to Keep People Safe
summary
Ofcom commissions Winder.AI to map how major platforms use MLOps and AI governance to moderate content, finding most work focuses on data pipelines and annotation rather than novel models, heavy use of A/B experiments in production, and massive scale: Facebook reports ~6.4 billion automated actions and 400 removals every second. Key hurdles are video annotation, GDPR-driven data deletion that prevents model repro, vendor integration costs (one mid-size platform spends ~$10M/year), looming legal pressure with 10% global turnover fine and foundation models included under EU rules.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:10Social media
- 2:20Is speech really free?
- 3:23Other laws
- 5:25Online safety
- 9:27Ofsted
- 13:01Online safety bill
- 13:50What is "illegal content"?
- 14:42OSM - Ramifications
- 18:06Social media - Scale
- 19:46How? - AI of course!
- 20:36What's MLOps?
- 21:35Report methodology
- 23:25General findings
- 33:42EU AI act
- 39:59What's next?
- 41:26Outro
related talks
13mHuman-in-the-Loop MLOps: Boost Model Performance by 40%
15mML Algorithms for Trust and Safety
01h 11mSupervising the Mass Adoption of Algorithmic Technologies
16mProduction MLOps for Spam Detection: Real-Time Multi-Modal AI
26mAre you ready for MLOps? 🫵
42mMLOps Under Attack: Threat Modeling Modern AI Systems
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
01h 00mWelcome to the machine