allsnsrank
회원가입

AI Trained on AI Collapses. Here’s Why Human Mistakes Matter | Dartmouth's AI at 70

Dartmouth
구독자 11.9만명
조회수 577회 · 좋아요 16 · 2026-11-16 📊 채널 정보 보기 ▶ 유튜브

지금 보고 있는 영상

AI Trained on AI Collapses. Here’s Why Human Mistakes Matter | Dartmouth's AI at 70
조회수 577
영상 설명 · 눌러서 펼치기
What happens when AI learns from AI? In this talk from Dartmouth’s conference AI at 70: What Must Remain Human?, @ThayerSchool assistant professor Bijan Mazaheri explains why generative models lose the rare, anomalous cases that make human data valuable, and why that matters for security, science, and engineering.

Part of AI at 70, marking 70 years since the 1956 Dartmouth workshop that launched the field of artificial intelligence.

Explore all the AI at 70 coverage: https://bit.ly/4iN18nW

In this talk:

- Why training AI on AI-generated data leads to “model collapse”
- Why AI struggles with “what if” questions and counterfactuals
- How anomalies reveal cause and effect, and why aggregation hides it
- Why more data can make causal reasoning worse (Simpson’s paradox)
- Why humans move fluidly between trends and individual cases, and AI doesn’t (yet)

Follow us:
https://www.facebook.com/Dartmouth | https://www.instagram.com/dartmouthcollege/ | https://www.linkedin.com/school/dartmouth-college | https://www.tiktok.com/@dartmouthcollege | https://twitter.com/dartmouth