LeftoverLocals: Listening to LLM responses through leaked GPU local memory
Date : 2024-01-16
Description
This summary was drafted with mixtral-8x7b-instruct-v0.1.Q5_K_M.gguf
In the Trail of Bits blog, Tyler Sorensen and Heidy Khlaaf discuss LeftoverLocals, a vulnerability that allows an attacker to recover data from GPU local memory created by another process on Apple, Qualcomm, AMD, and Imagination GPUs. This vulnerability is particularly significant for LLMs and ML models run on impacted GPU platforms, as it can leak local memory optimized for GPU use, potentially reconstructing LLM responses with high precision.
Read article here
Recently on :
Information Processing | Computing
Security | Surveillance | Privacy
WEB - 2024-03-04
Nvidia bans using translation layers for CUDA software | Tom's Hardware
Tom's Hardware - Nvidia has banned running CUDA-based software on other hardware platforms using translation layers in its lice...
WEB - 2024-02-21
Groq Inference Tokenomics: Speed, But At What Cost? | Semianalysis
Semianalysis - Groq, an AI hardware startup, has been making waves with their impressive demos showcasing Mistral Mixtral 8x7b ...
WEB - 2024-02-20
Let's build the GPT Tokenizer
VIDEO | Andrej Karpathy builds from scratch the Tokenizer used in the GPT series from OpenAI, showing that a lot of weird behav...
WEB - 2024-02-17
Nexus - 3D semantic graph of hacker interests
Freeman Jiang introduces Nexus, a 3D data visualization of hacker interests
WEB - 2024-02-16
SPAR: Personalized Content-Based Recommendation via Long Engagement Attention
Chiyu Zhang, Yifei Sun, Jun Chen, Jie Lei, Muhammad Abdul-Mageed, Sinong Wang, Rong Jin, Sem Park and Ning Yao and Bo Long use ...