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

It has been shown that Large Language Model (LLM) alignments can becircumvented by appending specially crafted attack suffixes with harmfulqueries to elicit harmful responses. To conduct attacks against private targetmodels whose characterization is unknown, public models can be used as proxiesto fashion the attack, with successful attacks being transferred from publicproxies to private target models. The success rate of attack depends on howclosely the proxy model approximates the private model. We hypothesize that forattacks to be transferrable, it is sufficient if the proxy can approximate thetarget model in the neighborhood of the harmful query. Therefore, in thispaper, we propose \emph{Local Fine-Tuning (LoFT)}, \textit{i.e.}, fine-tuningproxy models on similar queries that lie in the lexico-semantic neighborhood ofharmful queries to decrease the divergence between the proxy and target models.First, we demonstrate three approaches to prompt private target models toobtain similar queries given harmful queries. Next, we obtain data for localfine-tuning by eliciting responses from target models for the generated similarqueries. Then, we optimize attack suffixes to generate attack prompts andevaluate the impact of our local fine-tuning on the attack's success rate.Experiments show that local fine-tuning of proxy models improves attacktransferability and increases attack success rate by $39\%$, $7\%$, and $0.5\%$(absolute) on target models ChatGPT, GPT-4, and Claude respectively.

LoFT: Local Proxy Fine-tuning For Improving Transferability Of Adversarial Attacks Against Large Language Model