Abid et al. (2021) showed a tendency in GPT-3 to generate violent completions
when prompted about Muslims, compared with other religions. Two pre-registered
replication attempts found few violent completions and only the weakest
anti-Muslim bias in the Instruct version, fine-tuned to eliminate biased and
toxic outputs. However, more pre-registered experiments showed that using
common names associated with the religions in prompts increases several-fold
the rate of violent completions, revealing a highly significant second-order
bias against Muslims. Our content analysis revealed religion-specific violent
themes containing highly offensive ideas regardless of prompt format.
Replications with ChatGPT suggest that any effects of GPT-3's de-biasing have
disappeared with continued model development, as this newer model showed both a
strong Muslim-violence bias and rates of violent completions closer to Abid et
al. (2021). Our results show the need for continual de-biasing of models in
ways that address higher-order associations.

GPT-3 存在针对穆斯林的暴力生成倾向和反穆斯林偏见，复制实验表明去偏置措施在新模型中不再有效，加强高级关联的去偏置需求。

GPT 模型中存在对穆斯林暴力的偏见

Muslim-Violence Bias Persists in Debiased GPT Models

The prevalence of violence in daily life poses significant threats to
individuals' physical and mental well-being. Using surveillance cameras in
public spaces has proven effective in proactively deterring and preventing such
incidents. However, concerns regarding privacy invasion have emerged due to
their widespread deployment. To address the problem, we leverage Dynamic Vision
Sensors (DVS) cameras to detect violent incidents and preserve privacy since it
captures pixel brightness variations instead of static imagery. We introduce
the Bullying10K dataset, encompassing various actions, complex movements, and
occlusions from real-life scenarios. It provides three benchmarks for
evaluating different tasks: action recognition, temporal action localization,
and pose estimation. With 10,000 event segments, totaling 12 billion events and
255 GB of data, Bullying10K contributes significantly by balancing violence
detection and personal privacy persevering. And it also poses a challenge to
the neuromorphic dataset. It will serve as a valuable resource for training and
developing privacy-protecting video systems. The Bullying10K opens new
possibilities for innovative approaches in these domains.

使用动态视觉传感器（DVS）相机可以检测到暴力事件并保护隐私，并介绍了一个新的数据集 Bullying10K，可用于培训和开发保护隐私的视频系统，为此领域开启了创新方法的新可能性。

Bullying10K: 面向隐私保护的神经形态数据集用于欺凌识别

Bullying10K: A Neuromorphic Dataset towards Privacy-Preserving Bullying  Recognition

We present the first dataset and evaluation results on a newly defined
computational task of trigger warning assignment. Labeled corpus data has been
compiled from narrative works hosted on Archive of Our Own (AO3), a well-known
fanfiction site. In this paper, we focus on the most frequently assigned
trigger type--violence--and define a document-level binary classification task
of whether or not to assign a violence trigger warning to a fanfiction,
exploiting warning labels provided by AO3 authors. SVM and BERT models trained
in four evaluation setups on the corpora we compiled yield $F_1$ results
ranging from 0.585 to 0.798, proving the violence trigger warning assignment to
be a doable, however, non-trivial task.

本研究是第一次针对新定义的触发警告分配计算任务提供数据集和评估结果。该任务的数据来自 Archive of Our Own (AO3) 上托管的叙事作品的编译语料库，作者依靠 AO3 提供的警告标签为 Fanfiction 分配触发警告。本文主要讨论最常分配的触发类型（暴力），并定义了一个文档级的二进制分类任务，即是否为一个 Fanfiction 分配暴力触发警告。在我们编译的语料库上，训练的 SVM 和 BERT 模型在四个评估设置上得出的 $ F_1 $ 结果在 0.585 到 0.798 之间，证明了这项暴力触发警告分配任务是可行的，但并不是容易的任务。

Trigger Warnings: FanFiction 暴力检测器的引导启动

Trigger Warnings: Bootstrapping a Violence Detector for FanFiction

In every connected smart city around the world, CCTVs have played a pivotal
role in enforcing the safety and security of the citizens by recording unlawful
activities for the authorities to take action. To ensure the efficiency and
effectiveness of CCTVs in this domain, different DNN architectures were created
and used by researchers and developers to either detect violence or detect
weapons using bounding boxes or masks. These weapons are limited to guns,
knives, and other obvious handheld weapons. To remove these limits and detect
weapons more efficiently, non-weaponized violence footage from CCTV must be
differentiable from weaponized ones. Since there are no current datasets that
are tailored to this purpose of generalizability in weaponized violence
detection, we introduced a new dataset that contains videos depicting
weaponized violence, non-weaponized violence, and non-violent events. We also
propose a novel data-centric method that arranges video frames into salient
images while minimizing information loss for comfortable inference by SOTA
image classifiers. This was done to simplify video classification tasks and
optimize inference latency to improve sustainability in smart cities. Our
experiments show that Image Classifiers can efficiently detect and distinguish
violence with weapons from violence without weapons with performances as high
as 99\% on our dataset, which are comparable with current SOTA 3D networks for
action recognition and video classification.

为了提高智能城市安全的效率性，本研究提出了一个聚焦于武装暴力、非武装暴力和非暴力事件的新数据集，同时提出了一种基于数据的新方法用于图像分类，以提高智能城市中武器检测的性能。经过实验，该方法可高效地检测和区分带武器和不带武器的暴力事件，达到 99% 的性能。