The paper introduces Semantic Pyramid AutoEncoder (SPAE) to enable frozen language models (LLMs) to perform understanding and generation tasks with non-linguistic modalities. The approach converts between pixels and lexical tokens, allowing LLMs to comprehend and generate visual content. Experimental results show improved performance in image understanding tasks.

https://arxiv.org/abs//2306.17842

YouTube: https://www.youtube.com/@ArxivPapers

PODCASTS:

Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016

Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers