Hao Xue

Aaron J. Snoswell, Lucinda Nelson, Hao Xue et al.

Generic `toxicity' classifiers continue to be used for evaluating the potential for harm in natural language generation, despite mounting evidence of their shortcomings. We consider the challenge of measuring misogyny in natural language generation, and argue that generic `toxicity' classifiers are inadequate for this task. We use data from two well-characterised `Incel' communities on Reddit that differ primarily in their degrees of misogyny to construct a pair of training corpora which we use to fine-tune two language models. We show that an open source `toxicity' classifier is unable to distinguish meaningfully between generations from these models. We contrast this with a misogyny-specific lexicon recently proposed by feminist subject-matter experts, demonstrating that, despite the limitations of simple lexicon-based approaches, this shows promise as a benchmark to evaluate language models for misogyny, and that it is sensitive enough to reveal the known differences in these Reddit communities. Our preliminary findings highlight the limitations of a generic approach to evaluating harms, and further emphasise the need for careful benchmark design and selection in natural language evaluation.

Yuncheng Hua, Lizhen Qu, Zhuang Li et al.

Alignment tuning is crucial for ensuring large language models (LLMs) behave ethically and helpfully. Current alignment approaches require high-quality annotations and significant training resources. This paper proposes a low-cost, tuning-free method using in-context learning (ICL) to enhance LLM alignment. Through an analysis of high-quality ICL demos, we identified style as a key factor influencing LLM alignment capabilities and explicitly restyled ICL exemplars based on this stylistic framework. Additionally, we combined the restyled demos to achieve a balance between the two conflicting aspects of LLM alignment--factuality and safety. We packaged the restyled examples as prompts to trigger few-shot learning, improving LLM alignment. Compared to the best baseline approach, with an average score of 5.00 as the maximum, our method achieves a maximum 0.10 increase on the Alpaca task (from 4.50 to 4.60), a 0.22 enhancement on the Just-eval benchmark (from 4.34 to 4.56), and a maximum improvement of 0.32 (from 3.53 to 3.85) on the MT-Bench dataset. We release the code and data at https://github.com/AnonymousCode-ComputerScience/RIDE.

Hanlin Dong, Arian Prabowo, Hao Xue et al.

Air quality prediction is a challenging forecasting task due to its spatio-temporal complexity and the inherent dynamics as well as uncertainty. Most of the current models handle these two challenges by applying Graph Neural Networks or known physics principles, and quantifying stochasticity through probabilistic networks like Diffusion models. Nevertheless, finding the right balancing point between the certainties and uncertainties remains an open question. Therefore, we propose Double-Diffusion, a novel diffusion probabilistic model that harnesses the power of known physics to guide air quality forecasting with stochasticity. To the best of our knowledge, while precedents have been made of using conditional diffusion models to predict air pollution, this is the first attempt to use physics as a conditional generative approach for air quality prediction. Along with a sampling strategy adopted from image restoration and a new denoiser architecture, Double-Diffusion ranks first in most evaluation scenarios across two real-life datasets compared with other probabilistic models, it also cuts inference time by 50% to 30% while enjoying an increase between 3-12% in Continuous Ranked Probabilistic Score (CRPS).