Abhinav Rao

Abhinav Rao, Aditi Khandelwal, Kumar Tanmay et al. · microsoft-research

In this position paper, we argue that instead of morally aligning LLMs to specific set of ethical principles, we should infuse generic ethical reasoning capabilities into them so that they can handle value pluralism at a global scale. When provided with an ethical policy, an LLM should be capable of making decisions that are ethically consistent to the policy. We develop a framework that integrates moral dilemmas with moral principles pertaining to different foramlisms of normative ethics, and at different levels of abstractions. Initial experiments with GPT-x models shows that while GPT-4 is a nearly perfect ethical reasoner, the models still have bias towards the moral values of Western and English speaking societies.

Abhinav Rao, Ho Thi-Nga, Chng Eng-Siong

This paper presents the work of restoring punctuation for ASR transcripts generated by multilingual ASR systems. The focus languages are English, Mandarin, and Malay which are three of the most popular languages in Singapore. To the best of our knowledge, this is the first system that can tackle punctuation restoration for these three languages simultaneously. Traditional approaches usually treat the task as a sequential labeling task, however, this work adopts a slot-filling approach that predicts the presence and type of punctuation marks at each word boundary. The approach is similar to the Masked-Language Model approach employed during the pre-training stages of BERT, but instead of predicting the masked word, our model predicts masked punctuation. Additionally, we find that using Jieba1 instead of only using the built-in SentencePiece tokenizer of XLM-R can significantly improve the performance of punctuating Mandarin transcripts. Experimental results on English and Mandarin IWSLT2022 datasets and Malay News show that the proposed approach achieved state-of-the-art results for Mandarin with 73.8% F1-score while maintaining a reasonable F1-score for English and Malay, i.e. 74.7% and 78% respectively. Our source code that allows reproducing the results and building a simple web-based application for demonstration purposes is available on Github.

Abhinav Rao, Sachin Vashistha, Atharva Naik et al.

Recent explorations with commercial Large Language Models (LLMs) have shown that non-expert users can jailbreak LLMs by simply manipulating their prompts; resulting in degenerate output behavior, privacy and security breaches, offensive outputs, and violations of content regulator policies. Limited studies have been conducted to formalize and analyze these attacks and their mitigations. We bridge this gap by proposing a formalism and a taxonomy of known (and possible) jailbreaks. We survey existing jailbreak methods and their effectiveness on open-source and commercial LLMs (such as GPT-based models, OPT, BLOOM, and FLAN-T5-XXL). We further discuss the challenges of jailbreak detection in terms of their effectiveness against known attacks. For further analysis, we release a dataset of model outputs across 3700 jailbreak prompts over 4 tasks.

Abhinav Rao, Akhila Yerukola, Vishwa Shah et al. · allen-ai, cmu

To be effectively and safely deployed to global user populations, large language models (LLMs) may need to adapt outputs to user values and cultures, not just know about them. We introduce NormAd, an evaluation framework to assess LLMs' cultural adaptability, specifically measuring their ability to judge social acceptability across varying levels of cultural norm specificity, from abstract values to explicit social norms. As an instantiation of our framework, we create NormAd-Eti, a benchmark of 2.6k situational descriptions representing social-etiquette related cultural norms from 75 countries. Through comprehensive experiments on NormAd-Eti, we find that LLMs struggle to accurately judge social acceptability across these varying degrees of cultural contexts and show stronger adaptability to English-centric cultures over those from the Global South. Even in the simplest setting where the relevant social norms are provided, the best LLMs' performance (< 82\%) lags behind humans (> 95\%). In settings with abstract values and country information, model performance drops substantially (< 60\%), while human accuracy remains high (> 90\%). Furthermore, we find that models are better at recognizing socially acceptable versus unacceptable situations. Our findings showcase the current pitfalls in socio-cultural reasoning of LLMs which hinder their adaptability for global audiences.

Abhinav Rao, Monojit Choudhury, Somak Aditya

We introduce two paradoxes concerning jailbreak of foundation models: First, it is impossible to construct a perfect jailbreak classifier, and second, a weaker model cannot consistently detect whether a stronger (in a pareto-dominant sense) model is jailbroken or not. We provide formal proofs for these paradoxes and a short case study on Llama and GPT4-o to demonstrate this. We discuss broader theoretical and practical repercussions of these results.