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
click here for more information.
Flyphish is an Ansible playbook allowing cyber security consultants to deploy a phishing server in…
A crypto library to decrypt various encrypted D-Link firmware images. Confirmed to work on the…
Kali Linux 2024.4, the final release of 2024, brings a wide range of updates and…
This Go program applies a lifetime patch to PowerShell to disable ETW (Event Tracing for…
GPOHunter is a comprehensive tool designed to analyze and identify security misconfigurations in Active Directory…
Across small-to-medium enterprises (SMEs) and managed service providers (MSPs), the top priority for cybersecurity leaders…