News for hackers. Stuff that matters.
Re:Mistral AI Launches New 8x22B MOE Model by Aissen on Wednesday April 10, 2024 @04:29AM (#39988324)
Exploit a memory safety issue in the tokenizer/or other parts of your LLM infra written in a native language.
Re:Mistral AI Launches New 8x22B MOE Model by moralestapia on Wednesday April 10, 2024 @07:57AM (#39989690)
??? With weights?
Re:Mistral AI Launches New 8x22B MOE Model by abound on Wednesday April 10, 2024 @09:10AM (#39990305)
There are plenty of exploits where the payload is just "data" read by some vulnerable program (PDF readers, image viewers, browsers, compression tools, messaging apps, etc)
Re:Mistral AI Launches New 8x22B MOE Model by sp332 on Wednesday April 10, 2024 @06:18PM (#39996293)
Yes, there's a reason weights are now distributed as "safetensors" files. Malicious weights files in the old formats are possible, and while I haven't seen evidence of the new format being exploitable, I wouldn't be surprised if someone figures out how to do it eventually.
Re:Mistral AI Launches New 8x22B MOE Model by fzzzy on Wednesday April 10, 2024 @07:55PM (#39996933)
There was a buffer overflow or some other exploit like that in llama.cpp and the gguf format. It has been fixed now, but it's definitely possible. Also weights distributed as python pickles can run arbitrary code.
Re:Mistral AI Launches New 8x22B MOE Model by bevekspldnw on Wednesday April 10, 2024 @10:33PM (#39997852)
Distributing anything as python pickles seems utterly batshit to me.
Re:Mistral AI Launches New 8x22B MOE Model by fzzzy on Thursday April 11, 2024 @06:46AM (#40000722)
Completely agree.