In addition to being a fun game, this website is part of a research project studying prompt injection vulnerabilities in AI systems. The aim is to use crowdsourced data (from you!) to better understand how large language models (like the neural network that powers ChatGPT or Bard) can be forced to behave in undesirable ways. This will help researchers to build more reliable AI systems in the future.
By creating an account, you are giving consent to have your data used for research purposes, as outlined below, and agreeing to the terms of use.
Please direct any questions or concerns to robust-llms@berkeley.edu.
At the conclusion of this project, we plan to publicly release all submissions to the website. This will include any text you submit, as well as submission timestamps and random identifiers that make it possible to group together submissions made by the same user. Please do not enter any information that you would not want to become public!
In addition to the text that you submit to the game, we also store metadata related to your browser and internet connection in order to fix bugs and prevent abuse. This includes things like your IP address and browser version. This information will not be included in any public dataset that we release.
Finally, our service providers may collect information in accordance with
their own privacy policies.
Details:
Cloudflare
OpenAI
Google
We ask that users abide by the following rules: