About a year ago, I watched this 3Blue1Brown LLM tutorial (https://www.youtube.com/watch?v=eMlx5fFNoYc&t=367s) on how a model’s self-attention mechanism is used to predict the next token in a sequence, and I was surprised by how little we know about what actually happens when processing the sentence “A fluffy blue creature roamed the verdant forest.”
A year later, the field of mechanistic interpretability has seen significant advancements, and we're now able to "decompose" models into interpretable circuits that help explain how LLMs produce predictions. Using the second iteration of an LLM "debugger" I've been working on, I compare the hypothetical representations used in the tutorial to the actual representations I see when extracting a circuit that describes the processing of this specific sentence. If you're into model interpretability, please take a look! https://peterlai.github.io/gpt-circuits/