AI Safety Fundamentals: Alignment

83 Épisodes

1. Yudkowsky Contra Christiano on AI Takeoff Speeds

   Publié: 13/05/2023
2. Why AI Alignment Could Be Hard With Modern Deep Learning

   Publié: 13/05/2023
3. Least-To-Most Prompting Enables Complex Reasoning in Large Language Models

   Publié: 13/05/2023
4. Measuring Progress on Scalable Oversight for Large Language Models

   Publié: 13/05/2023
5. Supervising Strong Learners by Amplifying Weak Experts

   Publié: 13/05/2023
6. Summarizing Books With Human Feedback

   Publié: 13/05/2023
7. Robust Feature-Level Adversaries Are Interpretability Tools

   Publié: 13/05/2023
8. Debate Update: Obfuscated Arguments Problem

   Publié: 13/05/2023
9. High-Stakes Alignment via Adversarial Training [Redwood Research Report]

   Publié: 13/05/2023
10. AI Safety via Debate

    Publié: 13/05/2023
11. Takeaways From Our Robust Injury Classifier Project [Redwood Research]

    Publié: 13/05/2023
12. Introduction to Logical Decision Theory for Computer Scientists

    Publié: 13/05/2023
13. AI Safety via Debatered Teaming Language Models With Language Models

    Publié: 13/05/2023
14. Toy Models of Superposition

    Publié: 13/05/2023
15. Understanding Intermediate Layers Using Linear Classifier Probes

    Publié: 13/05/2023
16. Acquisition of Chess Knowledge in Alphazero

    Publié: 13/05/2023
17. Feature Visualization

    Publié: 13/05/2023
18. Discovering Latent Knowledge in Language Models Without Supervision

    Publié: 13/05/2023
19. Progress on Causal Influence Diagrams

    Publié: 13/05/2023
20. Careers in Alignment

    Publié: 13/05/2023