💻 **Dark LLMs: The Growing Threat of Unaligned AI Models**
"_In our research, we uncovered a universal jailbreak attack that effectively compromises multiple state-of-the-art models, enabling them to answer almost any question and produce harmful outputs upon request._"
Fire, M. et al. (2025) Dark LLMs: The growing threat of unaligned AI models. https://arxiv.org/abs/2505.10066.
#AI #ArtificialIntelligence #LLMS #DarkLLMS #Technology #Tech #Preprint #Research #ComputerScience @ai
@bibliolater After reading the abstract I again find the term "democratization" - a buzzword I've learned to associate with "true believers" and bullshitters in a corporate environment.
The "safety" of LLM based chatbots is a thing that reflects that we hold actors accountable for information they share. Semiotics as a sociological phenomenon, as presented in Umberto Eco's novels, comes to my mind.
@tg9541 I think the difference between the two is that with a search engine query one is not given the impression that one is communicating with a 'sentient being' whereas LLMS for many that is the case.
@bibliolater The ethics and morality of accessible information... From the forbidden and restricted sections of libraries, or codified sociopathy in cults. We have all reasons to believe that information can be toxic, can harm society.
How is an LLM different from an Internet search? The accessibility or the fact that information is presented in a normalized form?