DeepSeek-V4-Flash, a local model competitive with low-end frontier models for agentic coding, makes LLM activation steering newly practical for engineers. The post explains how steering works — extracting concept vectors from model activations and boosting them during inference — and explores why it hasn't seen wider adoption: big labs don't need it, API users can't access weights, and basic use cases are outcompeted by prompting. The author is cautiously skeptical: most simple steering gains can be replicated with prompts, while ambitious goals like steering for 'intelligence' or codebase knowledge likely require full fine-tuning. However, with a capable open-weights local model now available, the open-source community may finally start exploring steering seriously, potentially producing libraries of boostable features alongside model releases.
