AI Safety Fundamentals: Alignment

83 Episoder

1. Constitutional AI Harmlessness from AI Feedback

   Publisert: 19.7.2024
2. Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback

   Publisert: 19.7.2024
3. Illustrating Reinforcement Learning from Human Feedback (RLHF)

   Publisert: 19.7.2024
4. Deep Double Descent

   Publisert: 17.6.2024
5. Chinchilla’s Wild Implications

   Publisert: 17.6.2024
6. Eliciting Latent Knowledge

   Publisert: 17.6.2024
7. Intro to Brain-Like-AGI Safety

   Publisert: 17.6.2024
8. Toy Models of Superposition

   Publisert: 17.6.2024
9. Low-Stakes Alignment

   Publisert: 17.6.2024
10. ABS: Scanning Neural Networks for Back-Doors by Artificial Brain Stimulation

    Publisert: 17.6.2024
11. Two-Turn Debate Doesn’t Help Humans Answer Hard Reading Comprehension Questions

    Publisert: 17.6.2024
12. Gradient Hacking: Definitions and Examples

    Publisert: 17.6.2024
13. Imitative Generalisation (AKA ‘Learning the Prior’)

    Publisert: 17.6.2024
14. Discovering Latent Knowledge in Language Models Without Supervision

    Publisert: 17.6.2024
15. Least-To-Most Prompting Enables Complex Reasoning in Large Language Models

    Publisert: 17.6.2024
16. An Investigation of Model-Free Planning

    Publisert: 17.6.2024
17. Empirical Findings Generalize Surprisingly Far

    Publisert: 17.6.2024
18. Compute Trends Across Three Eras of Machine Learning

    Publisert: 13.6.2024
19. Worst-Case Thinking in AI Alignment

    Publisert: 29.5.2024
20. How to Get Feedback

    Publisert: 12.5.2024