Can open-weight LLMs can predict the effects of weight edits without running the model (9B–30B)
I’ve been experimenting with weight-space reading using Qwen and Muse models (9B–30B). The goal: can an LLM inspect raw model weights and predict the behavioral effects of weight edits without running the target model?