AI Innovators - Ep 10 - James Verbus, Sr. Staff Software Engineer, Machine Learning at LinkedIn

In this episode of the AI Innovators Podcast by SaladCloud (www.salad.com), James Verbus, a Sr. Staff Machine Learning Engineer at LinkedIn, shares his journey from particle astrophysics to machine learning at LinkedIn. He discusses identifying abuse on social media platforms, the ethical challenges of AI-generated content, the importance of trust and explainability in AI systems, and the ongoing arms race between AI detection systems and bad actors. Verbus also emphasizes the balance between effective detection and maintaining a positive user experience, as well as offering valuable advice for aspiring machine learning engineers on how to advance their careers.

Takeaways

• Transitioning from astrophysics to machine learning for broader opportunities.
• Detecting abuse on social media parallels searching for dark matter.
• LinkedIn's trust team addresses various types of abuse and trust issues.
• AI can be used for both trust and abuse in social media.
• Explainability in AI is crucial for user trust.
• AI tools can enhance creativity and content creation.
• The landscape of AI is heterogeneous and requires customization.
• Bad actors quickly adopt new technologies for scams.
• AI-generated content raises ethical concerns and user distrust.
• Robust detection systems are essential in the arms race against AI abuse. - - Models can detect AI-generated photos with high accuracy.
• Continuous model updates are essential for effective detection.
• Bad actors exploit AI-generated images for fake accounts.
• Training data quality and volume are critical for model performance.
• Detection models operate at various stages of user activity.
• Balancing detection and user experience is crucial.
• AI engineers should focus on solving real business problems.
• Effective communication of results is key for career advancement.
• Being T-shaped in skills can enhance impact in teams.
• A holistic approach to abuse prevention increases effectiveness.

This podcast is brought to you by SaladCloud: www.salad.com