We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
back
“Interpreting Complexity” by Maxwell Adam
01:12:35
Share
2025/3/14
LessWrong (30+ Karma)
AI Chapters
Transcribe
Chapters
Introduction: Neural Networks as Gardens
What Can We Learn from Prior Work?
What's Our Contribution?
What is Singular Learning Theory?
A Parameters Journey Through Function Space
Quantifying Degeneracy with Volume Scaling
Basin Volume and Its Scaling
Regular Versus Singular Landscapes
A One-Dimensional Intuition
Calculating the Local Learning Coefficient (LLC)
Stochastic Gradient Langevin Dynamics (SGLD)
A Synthetic Memorization Task
Beyond Averages: A Gradual Noising
Interpretable Fragility
Finding More Complex Circuits
Spirals in the Machine
Renormalization With Temperature
Conclusion and Future Work
Appendix: Scaling Mechanistic Detection of Memorization
Shownotes
Transcript
No transcript made for this episode yet, you may request it for free.
