Current LLM alignment methods are readily broken through specifically crafted
adversarial prompts. While crafting adversarial prompts using discrete
optimization is highly effective, such attacks typically use more than 100,000
LLM calls. This high computational cost makes them unsuitable for, e.g.,
quantitative analyses and adversarial training. To remedy this, we revisit
Projected Gradient Descent (PGD) on the continuously relaxed input prompt.
Although previous attempts with ordinary gradient-based attacks largely failed,
we show that carefully controlling the error introduced by the continuous
relaxation tremendously boosts their efficacy. Our PGD for LLMs is up to one
order of magnitude faster than state-of-the-art discrete optimization to
achieve the same devastating attack results.

通过控制连续放松引入的误差，我们改进了投影梯度下降（PGD）对连续放松输入提示的攻击方法，实现了与现有离散优化相同的毁灭性攻击结果，PGD 对 LLMs 的速度比最新的离散优化方法快了一个数量级。

使用投影梯度下降攻击大规模语言模型

Attacking Large Language Models with Projected Gradient Descent

We present preliminary results in quantitative analyses of color usage in
selected authors' works from LitBank. Using Glasgow Norms, human ratings on
5000+ words, we measure attributes of nouns dependent on color terms. Early
results demonstrate a significant increase in noun concreteness over time. We
also propose future research directions for computational literary color
analytics.

通过 Glasgow 规范和人类对 5000 个单词的评分，我们在来自 LitBank 的作家作品中进行了颜色使用的定量分析，发现名词的具体性随时间显著增加，并提出了计算文学色彩分析的未来研究方向。