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-AttackInstall the requirements.
$ pip install -r requirements.txtclick here for more information.
Introduction Bash scripting is a powerful way to automate Linux tasks, but writing a script…
Introduction A self-signed SSL certificate is a certificate that is created and signed by the…
Introduction Debugging is an important part of Bash scripting. When a script does not work…
Introduction Cron jobs are used in Linux to run commands or Bash scripts automatically at…
Introduction Pipes are an important feature in Linux and Bash scripting. A pipe allows you…
Introduction The grep, awk, and sed commands are powerful text-processing tools in Linux. They are…