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

Big models, exemplified by Large Language Models (LLMs), are models typicallypre-trained on massive data and comprised of enormous parameters, which notonly obtain significantly improved performance across diverse tasks but alsopresent emergent capabilities absent in smaller models. However, the growingintertwining of big models with everyday human lives poses potential risks andmight cause serious social harm. Therefore, many efforts have been made toalign LLMs with humans to make them better follow user instructions and satisfyhuman preferences. Nevertheless, `what to align with' has not been fullydiscussed, and inappropriate alignment goals might even backfire. In thispaper, we conduct a comprehensive survey of different alignment goals inexisting work and trace their evolution paths to help identify the mostessential goal. Particularly, we investigate related works from twoperspectives: the definition of alignment goals and alignment evaluation. Ouranalysis encompasses three distinct levels of alignment goals and reveals agoal transformation from fundamental abilities to value orientation, indicatingthe potential of intrinsic human values as the alignment goal for enhancedLLMs. Based on such results, we further discuss the challenges of achievingsuch intrinsic value alignment and provide a collection of available resourcesfor future research on the alignment of big models.

From Instructions to Intrinsic Human Values -- A Survey of Alignment Goals for Big Models