-
Notifications
You must be signed in to change notification settings - Fork 3
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
8 changed files
with
25 additions
and
16 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Large diffs are not rendered by default.
Oops, something went wrong.
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
|
@@ -255,7 +255,7 @@ | |
<link rel="alternate" href="/index.xml" type="application/rss+xml" title="PrivacyLens: Evaluating Privacy Norm Awareness of Language Models in Action"> | ||
|
||
|
||
<link rel="manifest" href="index.webmanifest"> | ||
<link rel="manifest" href="/index.webmanifest"> | ||
<link rel="icon" type="image/png" href="images/icon_hua0cadf66651d69d307ba4e93758c0005_10827_32x32_fill_lanczos_center_3.png"> | ||
<link rel="apple-touch-icon" type="image/png" href="images/icon_hua0cadf66651d69d307ba4e93758c0005_10827_192x192_fill_lanczos_center_3.png"> | ||
|
||
|
@@ -460,24 +460,24 @@ <h1 class="text-4xl font-bold"> | |
<a href="https://tianshili.me/" style="color: inherit;">Tianshi Li</span></a><sup>2</sup>, | ||
</span> | ||
<span class="block"> | ||
<a href="https://wyshi.github.io/" style="color: inherit;">Weiyan Shi</a><sup>1</sup>, | ||
<a href="https://wyshi.github.io/" style="color: inherit;">Weiyan Shi</a><sup>2</sup>, | ||
</span> | ||
<span class="block"> | ||
<a href="https://liuyanchen1015.github.io/" style="color: inherit;">Yanchen Liu</a><sup>1</sup>, | ||
<a href="https://liuyanchen1015.github.io/" style="color: inherit;">Yanchen Liu</a><sup>3</sup>, | ||
</span> | ||
<span class="block"> | ||
<a href="https://cs.stanford.edu/~diyiy/" style="color: inherit;">Diyi Yang</a><sup>1</sup> | ||
</span> | ||
</div> | ||
|
||
<div class="text-lg mt-1"> | ||
<span class="block"><sup>1</sup>Stanford University <sup>2</sup>Northeastern University</span> | ||
<span class="block"><sup>1</sup>Stanford University <sup>2</sup>Northeastern University <sup>3</sup>Harvard University</span> | ||
</div> | ||
|
||
<div class="text-center mt-6"> | ||
<div class="flex flex-wrap justify-center gap-4"> | ||
<span class="ml-4 mr-4"> | ||
<a href=""> | ||
<a href="https://drive.google.com/file/d/1u3ERxLCqAOhJHhgXEFcH5abE7pmc24YD/view?usp=sharing"> | ||
<i class="ai ai-arxiv mr-2"></i> | ||
Paper | ||
</a> | ||
|
@@ -508,9 +508,18 @@ <h1 class="text-4xl font-bold"> | |
</div> | ||
|
||
<div class="gif-container"> | ||
<img src="files/example1.gif" alt="GPT-4 Agent Privacy Leakage Example"> | ||
<img src="files/example2.gif" alt="Claude-3 Agent Privacy Leakage Example"> | ||
<img src="files/example3.gif" alt="Llama-3-70B Agent Privacy Leakage Example"> | ||
<figure> | ||
<img src="files/example1.gif" alt="GPT-4 Agent Privacy Leakage Example"> | ||
<figcaption>GPT-4 agent leaks employee resignation and termination details.</figcaption> | ||
</figure> | ||
<figure> | ||
<img src="files/example2.gif" alt="Claude-3 Agent Privacy Leakage Example"> | ||
<figcaption>Claude-3 agent leaks a child's online username and password.</figcaption> | ||
</figure> | ||
<figure> | ||
<img src="files/example3.gif" alt="Llama-3-70B Agent Privacy Leakage Example"> | ||
<figcaption>Llama-3-70B agent leaks the sex of the baby discovered during an ultrasound.</figcaption> | ||
</figure> | ||
</div> | ||
|
||
</div> | ||
|
@@ -782,7 +791,7 @@ <h1>Results Highlight</h1> | |
<img src="files/prompting.jpg" style="max-width: 100%; height: auto; margin-top: 20px; margin-bottom: 20px;"> | ||
<ol start="3"> | ||
<li>We experiment with adding privacy-enhancing content to the agent system prompt, but <strong>such prompt engineering does not significantly boost performance in action-based evaluation</strong>.</li> | ||
<li>Our paper includes further analysis of the safety-helpfulness trade-off.</li> | ||
<li><a href="https://drive.google.com/file/d/1u3ERxLCqAOhJHhgXEFcH5abE7pmc24YD/view?usp=sharing" target="_blank" rel="noopener">Our paper</a> includes further analysis of the safety-helpfulness trade-off.</li> | ||
</ol> | ||
<div style="border: 2px solid #000; border-radius: 10px; padding: 10px; width: 100%; margin: 0 auto; background-color: lightyellow;" class="mt-2"> | ||
<strong>⭐ Takeaways:</strong><br> | ||
|
@@ -848,7 +857,7 @@ <h1>Limitations & Reflection</h1> | |
<div class="col-12 col-lg-8"> | ||
<p>We consider our work to be <span style="background-color: lightyellow;">a first step in exploring privacy norm awareness of LMs in action</span> and recognize two main limitations that can serve as interesting directions for future work:</p> | ||
<ol> | ||
<li>We collect general privacy norms in the U.S to construct our dataset, but inherently, <strong>privacy concerns can differ across individuals and cultural groups</strong> (check out additional experiments by leveraging <a href="https://arxiv.org/abs/2404.15238" target="_blank" rel="noopener">CultureBank</a> in Appendix D).</li> | ||
<li>We collect general privacy norms in the U.S to construct our dataset, but inherently, <strong>privacy concerns can differ across individuals and cultural groups</strong> (check out additional experiments by leveraging <a href="https://arxiv.org/abs/2404.15238" target="_blank" rel="noopener">CultureBank</a> in <a href="https://drive.google.com/file/d/1u3ERxLCqAOhJHhgXEFcH5abE7pmc24YD/view?usp=sharing" target="_blank" rel="noopener">Appendix D</a>).</li> | ||
<li>Our setup focuses on LM-mediated communication (e.g., using LM agent to send emails/messages, etc.). <strong>Unintentional LM privacy leakage can occur in other scenarios</strong> (e.g., web agents interacting with websites).</li> | ||
</ol> | ||
<p>If you have any thoughts or are interested in this line of research, feel free to <a href="mailto:[email protected]">reach out to us</a>.</p> | ||
|
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters