Can a general LLM read a LoRA weight diff zero-shot and say what it implants? Evaluated on DIT's hidden-topic and news-summary benchmarks.
lora ai-safety interpretability mechanistic-interpretability llm-agents qwen3 model-diffing weight-diff
-
Updated
Nov 6, 2026 - Python