Breaking Lab srl
A new research paper argues that large language models cannot be made fully secure against prompt injection and related attacks due to a fundamental architectural flaw. The findings carry major implications for every product built on top of AI models.
A new study demonstrates that an AI model named Evo can design functional viruses that replicate and destroy bacteria. Trained on bacteriophage genomes, the AI created viruses that were in some cases more infectious than their natural counterparts, showcasing its ability to innovate beyond human intuition. While this technology holds immense potential for therapeutic applications in medicine, particularly in fighting antibiotic-resistant bacteria, it also raises significant ethical and security concerns about the potential for creating new, dangerous pathogens.
This article explores a novel method for interacting with AI models like ChatGPT, moving beyond simple queries to create a personalized expert consultant. By using a structured prompt that assigns a specific role, requests clarifying questions, and demands a step-by-step solution, users can unlock the AI's full potential for a wide range of tasks, from creative problem-solving to learning new concepts. The article highlights the importance of this method for obtaining more relevant and actionable advice while also cautioning that the AI should be used as a brainstorming partner rather than a replacement for human expertise.
Google has expanded its AI Mode to over 180 countries and introduced new features to enhance its functionality. These include a powerful "agentic capability" that allows the AI to perform complex, multi-step tasks like booking a restaurant. The update also adds a new personalised response feature that uses a user's history for more relevant answers, as well as a "Share" button for easy communication. This move signifies Google's focus on making AI an active assistant rather than a passive information source.
