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

While large language models (LLMs) have achieved impressive performance ingenerating fluent and realistic text, controlling the generated text so that itexhibits properties such as safety, factuality, and non-toxicity remainschallenging. % such as DExperts, GeDi, and rectification Existingdecoding-based methods are static in terms of the dimension of control; if thetarget subject is changed, they require new training. Moreover, it can quicklybecome prohibitive to concurrently control multiple subjects. In this work, weintroduce SF-GEN, which is grounded in two primary concepts: successor features(SFs) to decouple the LLM's dynamics from task-specific rewards, and languagemodel rectification to proportionally adjust the probability of selecting atoken based on the likelihood that the finished text becomes undesired. SF-GENseamlessly integrates the two to enable dynamic steering of text generationwith no need to alter the LLM's parameters. Thanks to the decoupling effectinduced by successor features, our method proves to be memory-wise andcomputationally efficient for training as well as decoding, especially whendealing with multiple target subjects. To the best of our knowledge, ourresearch represents the first application of successor features in textgeneration. In addition to its computational efficiency, the resultant languageproduced by our method is comparable to the SOTA (and outperforms baselines) inboth control measures as well as language quality, which we demonstrate througha series of experiments in various controllable text generation tasks.

Successor Features for Efficient Multisubject Controlled Text Generation