Angel Lim's Blog

Hi, I'm an independent researcher working on mechanistic interpretability.

My background and work experience is in infrastructure, most recently focused on model inference.

I was one of the founding engineers at Voyage AI (acquired for $220M), and worked on autonomy infrastructure at various self-driving startups like Nuro and Parallel Systems.

Posts

Mechanistic Investigation of In-Context Learning
In-context learning (ICL) is when a model learns at inference-time to do a task from examples given in its prompt. You show it a few input-output pairs, and given a new input, it produces the right output. The model learns what to do on-the-fly all within the forward pass.