AI DA SIĘ OSZUKAĆ? 🤯 Uważaj na atak PROMPT INJECTION! Hakowanie AI jednym zdaniem?

The video delves into the mechanism of 'prompt injection,' a technique used to trick AI language models by providing counter-instructions that override existing safety protocols. The presenter illustrates this by showing how modern AI models operate on a multi-layered structure of instructions, with system instructions taking the highest priority and user prompts being evaluated later. By framing prompts as a simulation or test, users can theoretically trick the model into ignoring safety constraints to perform prohibited tasks. The presenter emphasizes that while this can demonstrate vulnerabilities, it is not a 'magical' way to break AI, and companies are constantly hardening their models against these types of attacks. The video provides a clear, educational overview of why these vulnerabilities exist and why robust safety alignment is critical for AI development.