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

The recent wave of generative AI has sparked unprecedented global attention,with both excitement and concern over potentially superhuman levels ofartificial intelligence: models now take only seconds to produce outputs thatwould challenge or exceed the capabilities even of expert humans. At the sametime, models still show basic errors in understanding that would not beexpected even in non-expert humans. This presents us with an apparent paradox:how do we reconcile seemingly superhuman capabilities with the persistence oferrors that few humans would make? In this work, we posit that this tensionreflects a divergence in the configuration of intelligence in today'sgenerative models relative to intelligence in humans. Specifically, we proposeand test the Generative AI Paradox hypothesis: generative models, having beentrained directly to reproduce expert-like outputs, acquire generativecapabilities that are not contingent upon -- and can therefore exceed -- theirability to understand those same types of outputs. This contrasts with humans,for whom basic understanding almost always precedes the ability to generateexpert-level outputs. We test this hypothesis through controlled experimentsanalyzing generation vs. understanding in generative models, across bothlanguage and image modalities. Our results show that although models canoutperform humans in generation, they consistently fall short of humancapabilities in measures of understanding, as well as weaker correlationbetween generation and understanding performance, and more brittleness toadversarial inputs. Our findings support the hypothesis that models' generativecapability may not be contingent upon understanding capability, and call forcaution in interpreting artificial intelligence by analogy to humanintelligence.

The Generative AI Paradox: "What It Can Create, It May Not Understand"