Recent studies reveal that Large Language Models (LLMs) face challenges in balancing safety with utility, particularly when processing long texts for NLP tasks like summarization and translation. Despite defenses against malicious short questions, the ability of LLMs to safely handle dangerous long content, such as manuals teaching illicit activities, remains unclear. Our work aims to develop robust defenses for LLMs in processing malicious documents alongside benign NLP task queries. We introduce a defense dataset comprised of safety-related examples and propose single-task and mixed-task losses for instruction tuning. Our empirical results demonstrate that LLMs can significantly enhance their capacity to safely manage dangerous content with appropriate instruction tuning. Additionally, strengthening the defenses of tasks most susceptible to misuse is effective in protecting LLMs against processing harmful information. We also observe that trade-offs between utility and safety exist in defense strategies, where Llama2, utilizing our proposed approach, displays a significantly better balance compared to Llama1.

Physical common sense plays an essential role in the cognition abilities of robots for human-robot interaction. Machine learning methods have shown promising results on physical commonsense learning in natural language processing but still suffer from model generalization. In this paper, we formulate physical commonsense learning as a knowledge graph completion problem to better use the latent relationships among training samples. Compared with completing general knowledge graphs, completing a physical commonsense knowledge graph has three unique characteristics: training data are scarce, not all facts can be mined from existing texts, and the number of relationships is small. To deal with these problems, we first use a pre-training language model BERT to augment training data, and then employ constrained tucker factorization to model complex relationships by constraining types and adding negative relationships. We compare our method with existing state-of-the-art knowledge graph embedding methods and show its superior performance.

We propose a novel tensor embedding method that can effectively extract lexical features for humor recognition. Specifically, we use word-word co-occurrence to encode the contextual content of documents, and then decompose the tensor to get corresponding vector representations. We show that this simple method can capture features of lexical humor effectively for continuous humor recognition. In particular, we achieve a distance of 0.887 on a global humor ranking task, comparable to the top performing systems from SemEval 2017 Task 6B (Potash et al., 2017) but without the need for any external training corpus. In addition, we further show that this approach is also beneficial for small sample humor recognition tasks through a semi-supervised label propagation procedure, which achieves about 0.7 accuracy on the 16000 One-Liners (Mihalcea and Strapparava, 2005) and Pun of the Day (Yang et al., 2015) humour classification datasets using only 10% of known labels.