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

Transformer based large language models with emergent capabilities arebecoming increasingly ubiquitous in society. However, the task of understandingand interpreting their internal workings, in the context of adversarialattacks, remains largely unsolved. Gradient-based universal adversarial attackshave been shown to be highly effective on large language models and potentiallydangerous due to their input-agnostic nature. This work presents a novelgeometric perspective explaining universal adversarial attacks on largelanguage models. By attacking the 117M parameter GPT-2 model, we find evidenceindicating that universal adversarial triggers could be embedding vectors whichmerely approximate the semantic information in their adversarial trainingregion. This hypothesis is supported by white-box model analysis comprisingdimensionality reduction and similarity measurement of hidden representations.We believe this new geometric perspective on the underlying mechanism drivinguniversal attacks could help us gain deeper insight into the internal workingsand failure modes of LLMs, thus enabling their mitigation.

Why do universal adversarial attacks work on large language models?: Geometry might be the answer