Stories by adebayoj

Aligning LLMs at inference time by suppressing internal concepts

Extracting 100K concepts from an 8B LLM

Show HN: Steerling-8B, a language model that can explain any token it generates