Repeated token replay attacks continue to be viable
"After the Scalable Extraction paper was published, OpenAI implemented filtering of prompt inputs containing repeated single tokens. As part of our regular application security review, Dropbox engineers discovered that OpenAI’s models were, under certain circumstances, still vulnerable to the repeated token attack. Dropbox used repeated multi-token (>1) sequences to induce divergence in ChatGPT models and demonstrated extraction of memorized training data from both GPT-3.5 and GPT-4"
https://t.co/MK2yLAznmZ
day 2 @CamlisOrg was a really great day of talks, with a couple favorites for me. starting out with a super great keynote from @tomgoldsteincs on the state of the art which was really energizing to hear about all the aspects of open research q's that exist for adversarial ML.
🛠️ llm-security
Scripts and related documentation that demonstrate attacks against large language models using repeated character sequences
By @Dropbox's @WHITEHACKSEC, @mlbr3it#cybersecurity#AI
https://t.co/wvicalC8t6
🔥 New type of prompt injection
Basically you can use control characters (like backspace) to circumvent system instructions
In extreme cases models will also hallucinate or respond with an answer to a completely different question
By @Dropbox#AI
https://t.co/DMfzxzN7fa
Thank you to @Verizon who will donate $1 for every RT, up to $1.5M, in support of our nations first responders through the @GarySiniseFound (1.18.19 – 2.8.19)! I’m proud to partner with Verizon and introduce The Team That Wouldn’t Be Here. Join me in recognizing first responders.
So, a shipment of crickets for the lizard arrived via FedEx today. It was my first time ordering bulk crickets off the internet, and I naively assumed that they would be in like, a bag or some other contraption to facilitate easy transfer to another container. They were not.
Hey guys, I know I usually just post shitty jokes on my Twitter but bear with me because I wanted to share something.
So in one of my Management classes I sit in the same seat in the front every day. Every single day I sit there.
Now, I also sit next to some foreign guy that