Execution Failures are Still Failures (Calibrate Surprise by Making Weird Excuses)
“Almost only counts in horseshoes and hand grenades” — Frank Robinson
Hello, my name is Kelechi Nwankwoala
I’m a cognitive market-researcher turned AI researcher/engineer.
In the summer of 2024, I completed a 3-month batch at the Recurse Center, a self-directed coding retreat, where I focused on developing research-level AI expertise. During my first 8 weeks at Recurse, I completed ARENA, an intensive program on AI fundamentals and interpretability methods. This enabled me to conduct original research investigating how Sparse AutoEncoders (SAEs) can selectively remove information from large language models—addressing a critical AI safety challenge. After Recurse, I was selected for Neel Nanda’s competitive MATS training phase, where I worked on interpreting mechanisms of refusal in LLMs. Outside of AI research, I’ve also spent time working on a desktop app to visualize personal time data and fine-tuning an LLM to sound like me.
Recent posts
Sparse Autoencoders for Improving Unlearning in Large Language Models
A.K.A: Smallish Large-Language Models: What Do They Know? Can They Un-Know Things?? Let’s Find Out.