Is ChatGPT a General-Purpose Natural Language Processing Task Solver?
Benchmarking the zero-shot performance of ChatGPT on 20 popular NLP datasets covering 7 representative task categories.
Provides insights on the effectiveness and limitations of ChatGPT in performing zero-shot learning on various NLP tasks. Recommends using ChatGPT for tasks requiring reasoning capabilities but caution is advised for specific tasks like sequence tagging.
Single Motion Diffusion
Presents a model called SinMDM that can learn the internal motifs of a single motion sequence and synthesize motions of arbitrary length that are faithful to them.
Provides a state-of-the-art approach for synthesizing realistic animations of humans, animals, and imaginary creatures. Recommends using SinMDM for various applications including spatial and temporal in-betweening, motion expansion, style transfer, and crowd animation.