Open source initiative with the goal of documenting, tracking and discussing language model vulnerabilities and exposures (LVEs).
Joined December 2023
- Tweets29
- Following7
- Followers109
- Likes13
LVE Project retweeted
We are delighted to announce Invariant Labs (invariantlabs.ai)! Our mission is to make AI agents secure and reliable! It is founded by @mvechev, @mbalunovic, @lbeurerkellner, @marc_r_fischer, @florian_tramer and builds on years of experience in industry and academia.
LVE Project retweeted
I'm excited to represent @projectlve at the webinar. Hope to see you there!
Projects have been gathering a lot of data about LLM vulnerabilities via gamification. Join @LakeraAI, @TensorTrust, @learnprompting, and @projectlve for our Lessons Learned from Crowdsourced LLM Threat Intelligence webinar on 2024-01-25 at 18:00 UTC.
lakera.ai/event/crowdsourcin…
We are really excited to present thoughts and insights from our project at the Lessons Learned from Crowdsourced LLM Threat Intelligence!
Projects have been gathering a lot of data about LLM vulnerabilities via gamification. Join @LakeraAI, @TensorTrust, @learnprompting, and @projectlve for our Lessons Learned from Crowdsourced LLM Threat Intelligence webinar on 2024-01-25 at 18:00 UTC.
lakera.ai/event/crowdsourcin…
LVE Project retweeted
🗒️ LVE Repository
Like CVEs but for LLMs
A project documenting and tracking vulnerabilities and exposures of large language models (LVEs)
lve-project.org/index.html
LVE Project retweeted
Projects have been gathering a lot of data about LLM vulnerabilities via gamification. Join @LakeraAI, @TensorTrust, @learnprompting, and @projectlve for our Lessons Learned from Crowdsourced LLM Threat Intelligence webinar on 2024-01-25 at 18:00 UTC.
lakera.ai/event/crowdsourcin…
Check out our latest copyright LVE based on the @nytimes vs @OpenAI lawsuit!
lve-project.org/responsibili…
What do you think about the lawsuit outcome and its implications on the future of AI?
If you found more examples, we encourage you to contribute them to the open-source LVE repository: github.com/lve-org/lve
You can find more copyright issues in the lawsuit document created by @nytimes: nytco-assets.nytimes.com/202…
🎄For the festive season, we have added a number of bonus levels to the current batch of LVE community challenges.
Have a look, and help us red team LLMs in the process!
Happy holidays everyone!
LVE Challenges: lve-project.org/challenges/
LLM safety filters like Purple Llama promise responsible and safe deployment of AI, but how effective are they?
In our new blog post, we argue that LLM-based filters are clearly flawed, and that they set up a dangerously circular safety narrative.
Blog: lve-project.org/blog/how-eff…
Kudos to @andyzou_jiaming for running the suffix attack with Llama Guard. You can find their code here: github.com/andyzoujm/breakin….
Did you know that you can abuse LLMs to infer a person's location from just a simple online comment?
We demonstrate this in our Location Inference challenge, and show how LLMs can be abused to spy on people, even WITHOUT their consent.
Challenge: lve-project.org/challenges/l…
LLMs can also infer other private attributes and do so with similar accuracy as humans.
But LLMs are also much cheaper and faster. This enables new forms of online profiling and privacy violation at scale, posing a significant safety risk.
Paper: arxiv.org/abs/2310.07298v1
Want to chat about the LVE project at @NeurIPSConf #NeurIPS2023?
We made it easy to spot @marc_r_fischer.
LVE Project retweeted
As you can tell from my previous Tweet I'm making my way to NOLA for @NeurIPSConf #NeurIPS2023.
Happy to chat about @lmqllang, @projectlve, my papers (🧵👇) and trustworthy/safe AI in general.
We are super excited to announce LVE 🎉
With LVEs we track LLM vulnerabilities and exposures in an open-source community-first approach.
Announcement: lve-project.org/blog/launchi…
🧵 A thread on the LVE project and why it matters:
Through challenges, the community can easily participate and helps us to discover new LLM safety issues and failure modes.
All of the collected submissions are released open source after a challenge concludes, so the entire community can benefit from them.
LVE is a project by @the_sri_lab @ETH_en @ETH and is lead by @mbalunovic, @lbeurerkellner, @marc_r_fischer, @a_yukh, @florian_tramer, @mvechev.