[2209.02128] Evaluating the Susceptibility of Pre-Trained Language Models via Handcrafted Adversarial Examples

Abstract:Recent advances in the development of large language models have resulted in public access to state-of-the-art pre-trained language models (PLMs), including Generative Pre-trained Transformer 3 (GPT-3) and Bidirectional Encoder Representations from Transformers (BERT). However, evaluations of PLMs, in practice, have shown their susceptibility to adversarial attacks during the training and fine-tuning stages of development. Such attacks can result in erroneous outputs, model-generated hate speech, and the exposure of users' sensitive information. While existing research has focused on adversarial attacks during either the training or the fine-tuning of PLMs, there is a deficit of information on attacks made between these two development phases. In this work, we highlight a major security vulnerability in the public release of GPT-3 and further investigate this vulnerability in other state-of-the-art PLMs. We restrict our work to pre-trained models that have not undergone fine-tuning. Further, we underscore token distance-minimized perturbations as an effective adversarial approach, bypassing both supervised and unsupervised quality measures. Following this approach, we observe a significant decrease in text classification quality when evaluating for semantic similarity.

[2209.02128] Evaluating the Susceptibility of Pre-Trained Language Models via Handcrafted Adversarial Examples Skip to main content arXiv is now an independent nonprofit! Learn more × Search arXiv Press Enter to search · Advanced search --> Computer Science > Computation and Language arXiv:2209.02128 (cs) [Submitted on 5 Sep 2022] Title: Evaluating the Susceptibility of Pre-Trained Language Models via Handcrafted Adversarial Examples Authors: Hezekiah J. Branch , Jonathan Rodriguez Cefalu , Jeremy McHugh , Leyla Hujer , Aditya Bahl , Daniel del Castillo Iglesias , Ron Heichman , R

related reading

• gpt-4.pdfcdn.openai.com
• Adversarial Attacks on Aligned Language Models | Gray Swan Researchgrayswan.ai
• A small number of samples can poison LLMs of any size \ Anthropicanthropic.com
• Adversarial Attacks on LLMs | Lil'Loglilianweng.github.io
• Nicholas Carlininicholas.carlini.com
• Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailmentarxiv.org
• [2307.15043] Universal and Transferable Adversarial Attacks on Aligned Language Modelsarxiv.org
• Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilitiesarxiv.org
• Universal and Transferable Attacks on Aligned Language Modelsllm-attacks.org
• Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samplesarxiv.org
• Generalized Language Models | Lil'Loglilianweng.github.io
• GitHub - SoyGema/pulling_ace · GitHubgithub.com