Zhen Li

Xinlong Wang, Xiaosong Zhang, Zhengxiong Luo et al. · tsinghua

While next-token prediction is considered a promising path towards artificial general intelligence, it has struggled to excel in multimodal tasks, which are still dominated by diffusion models (e.g., Stable Diffusion) and compositional approaches (e.g., CLIP combined with LLMs). In this paper, we introduce Emu3, a new suite of state-of-the-art multimodal models trained solely with next-token prediction. By tokenizing images, text, and videos into a discrete space, we train a single transformer from scratch on a mixture of multimodal sequences. Emu3 outperforms several well-established task-specific models in both generation and perception tasks, surpassing flagship models such as SDXL and LLaVA-1.6, while eliminating the need for diffusion or compositional architectures. Emu3 is also capable of generating high-fidelity video via predicting the next token in a video sequence. We simplify complex multimodal model designs by converging on a singular focus: tokens, unlocking great potential for scaling both during training and inference. Our results demonstrate that next-token prediction is a promising path towards building general multimodal intelligence beyond language. We open-source key techniques and models to support further research in this direction.

Julian Ramos, Zhen Li, Johana Rosas et al.

Pointing devices that reside on the keyboard can reduce the overall time needed to perform mixed pointing and typing tasks, since the hand of the user does not have to reach for the pointing device. However, previous implementations of this kind of device have a higher movement time compared to the mouse and trackpad due to large error rate, low speed and spatial resolution. In this paper we introduce Keyboard Surface Interaction (KSI), an interaction approach that turns the surface of a keyboard into an interaction surface and allows users to rest their hands on the keyboard at all times to minimize fatigue. We developed a proof-of-concept implementation, Fingers, which we optimized over a series of studies. Finally, we evaluated Fingers against the mouse and trackpad in a user study with 25 participants on a Fitts law test style, mixed typing and pointing task. Results showed that for users with more exposure to KSI, our KSI device had better performance (reduced movement and homing time) and reduced discomfort compared to the trackpad. When compared to the mouse, KSI had reduced homing time and reduced discomfort, but increased movement time. This interaction approach is not only a new way to capitalize on the space on top of the keyboard, but also a call to innovate and think beyond the touchscreen, touchpad, and mouse as our main pointing devices. The results of our studies serve as a specification for future KSI devices.