Dexa/This Week in ML & AI

Interpretable Code

The challenge of interpreting code is highlighted, revealing that human-written code often lacks clarity. However, discrete representations of algorithms can enhance interpretability, especially with tools designed for mechanistic interpretability. This opens up new avenues for understanding the inner workings of models that are typically opaque.

In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Learning Transformer Programs with Dan Friedman - 667
Related Questions
- Can neural networks be explained?
- Why are neural networks hard to explain in the episode Demystifying LLMs with Mechanistic Interpretability Researcher Arthur Conmy and the clip Automating AI Interpretability?