pku-yuangroup / hallucination-attack Goto Github PK

LLMs (e.g., GPT-3.5, LLaMA, and PaLM) suffer from hallucination—fabricating non-existent facts to cheat users without perception. And the reasons for their existence and pervasiveness remain unclear. We demonstrate that non-sense Out-of-Distribution(OoD) prompts composed of random tokens can also elicit the LLMs to respond with hallucinations. This phenomenon forces us to revisit that hallucination may be another view of adversarial examples, and it shares similar features with conventional adversarial examples as the basic feature of LLMs. Therefore, we formalize an automatic hallucination triggering method called hallucination attack in an adversarial way. Following is a fake news example generating by hallucination attack.

We substitute tokens via gradient-based token replacing strategy, replacing token reaching smaller negative log-likelihood loss, and induce LLM within hallucinations.

You may config your own base models and their hyper-parameters within config.py. Then, you could attack the models or run our demo cases.

Clone this repo and run the code.

$ cd Hallucination-Attack

Install the requirements.

$ pip install -r requirements.txt

Run local demo of hallucination attacked prompt.

$ python demo.py

Start a new attack training to find a prompt trigger hallucination

$ python main.py

@article{yao2023llm,
  title={LLM Lies: Hallucinations are not Bugs, but Features as Adversarial Examples},
  author={Yao, Jia-Yu and Ning, Kun-Peng and Liu, Zhen-Hui and Ning, Mu-Nan and Yuan, Li},
  journal={arXiv preprint arXiv:2310.01469},
  year={2023}
}