English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Warning: this paper includes model outputs showing offensive content. Recentlarge-scale Visual-Language Generative Models (VLGMs) have achievedunprecedented improvement in multimodal image/text generation. However, thesemodels might also generate toxic content, e.g., offensive text and pornographyimages, raising significant ethical risks. Despite exhaustive studies on toxicdegeneration of language models, this problem remains largely unexplored withinthe context of visual-language generation. This work delves into the propensityfor toxicity generation and susceptibility to toxic data across various VLGMs.For this purpose, we built ToViLaG, a dataset comprising 32Kco-toxic/mono-toxic text-image pairs and 1K innocuous but evocative text thattends to stimulate toxicity. Furthermore, we propose WInToRe, a novel toxicitymetric tailored to visual-language generation, which theoretically reflectsdifferent aspects of toxicity considering both input and output. On such abasis, we benchmarked the toxicity of a diverse spectrum of VLGMs anddiscovered that some models do more evil than expected while some are morevulnerable to infection, underscoring the necessity of VLGMs detoxification.Therefore, we develop an innovative bottleneck-based detoxification method. Ourmethod could reduce toxicity while maintaining comparable generation quality,providing a promising initial solution to this line of research.

ToViLaG: Your Visual-Language Generative Model is Also An Evildoer