Brief

Researchers, including a team led by Kaiming He and composed primarily of undergraduate students, have introduced MiniT2I, a novel text-to-image generation model. This model achieves competitive results with significantly fewer parameters (258M) and lower training costs, comparable to standard ImageNet experiments. MiniT2I utilizes a new MM-JiT architecture that operates directly in pixel space, eliminating the need for VAEs and simplifying the diffusion process by removing mechanisms like AdaLN, which are common in other large-scale text-to-image models. AI

Researchers have developed a novel backdoor attack called Turn-based Structural Triggers (TST) that exploits the dialogue structure of Large Language Models (LLMs) rather than user-visible prompts. This attack uses the turn index within a conversation as the trigger, allowing a backdoored model to execute malicious behaviors at specific points in a dialogue without any discernible input trigger. TST demonstrated a high attack success rate across multiple LLM families while maintaining normal performance on non-triggered tasks, highlighting a new vulnerability in multi-turn conversational AI systems. AI