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

When provided with sufficient explanatory context, smaller Language Modelshave been shown to exhibit strong reasoning ability on challenging short-answerquestion-answering tasks where the questions are unseen in training. Weevaluate two methods for further improvement in this setting. Both methodsfocus on combining rationales generated by a larger Language Model with longercontexts created from a multi-hop dense retrieval system. The first method($\textit{RR}$) involves training a Rationale Ranking model to score bothgenerated rationales and retrieved contexts with respect to relevance andtruthfulness. We then use the scores to derive combined contexts from bothknowledge sources using a number of combinatory strategies. For the secondmethod ($\textit{RATD}$) we utilise retrieval-augmented training datasetsdeveloped by Hartill et al. 2023 to train a smaller Reasoning model such thatit becomes proficient at utilising relevant information from longer textsequences that may be only partially evidential and frequently contain manyirrelevant sentences. We find that both methods significantly improve results.Our single best Reasoning model materially improves upon strong comparableprior baselines for unseen evaluation datasets (StrategyQA 58.9 $\rightarrow$61.7 acc., CommonsenseQA 63.6 $\rightarrow$ 72.7 acc., ARC-DA 31.6$\rightarrow$ 52.1 F1, IIRC 25.5 $\rightarrow$ 27.3 F1) and a version utilisingour prior knowledge of each type of question in selecting a context combinationstrategy does even better. Our proposed models also generally outperform directprompts against much larger models (BLOOM 175B and StableVicuna 13B) in bothfew-shot chain-of-thought and standard few-shot settings.

Answering Unseen Questions With Smaller Language Models Using Rationale Generation and Dense Retrieval