The interactive nature of Large Language Models (LLMs) theoretically allows models to refine and improve their answers, yet systematic analysis of the multi-turn behavior of LLMs remains limited. In this paper, we propose the FlipFlop experiment: in the first round of the conversation, an LLM responds to a prompt containing a classification task. In a second round, the LLM is challenged with a follow-up phrase like "Are you sure?", offering an opportunity for the model to reflect on its initial answer, and decide whether to confirm or flip its answer. A systematic study of nine LLMs on seven classification tasks reveals that models flip their answers on average 46% of the time and that all models see a deterioration of accuracy between their first and final prediction, with an average drop of 17%. The FlipFlop experiment illustrates the universality of sycophantic behavior in LLMs and provides a robust framework to analyze model behavior and evaluate potential solutions.

Over the past ten years, modern natural language processing models have revolutionised the field of artificial intelligence, transforming how computers understand and generate language. ChatGPT is one key example of how AI technology is becoming increasingly important in generating more and more precise human-like responses to a wide variety of problems, including business ones. These advancements have significant implications for the financial industry, where vast amounts of regulatory data and policies must be analysed and understood in order to comply with laws and regulations.

Large Language Models (LLMs) are central to a multitude of applications but struggle with significant risks, notably in generating harmful content and biases. Drawing an analogy to the human psyche's conflict between evolutionary survival instincts and societal norm adherence elucidated in Freud's psychoanalysis theory, we argue that LLMs suffer a similar fundamental conflict, arising between their inherent desire for syntactic and semantic continuity, established during the pre-training phase, and the post-training alignment with human values. This conflict renders LLMs vulnerable to adversarial attacks, wherein intensifying the models' desire for continuity can circumvent alignment efforts, resulting in the generation of harmful information. Through a series of experiments, we first validated the existence of the desire for continuity in LLMs, and further devised a straightforward yet powerful technique, such as incomplete sentences, negative priming, and cognitive dissonance scenarios, to demonstrate that even advanced LLMs struggle to prevent the generation of harmful information. In summary, our study uncovers the root of LLMs' vulnerabilities to adversarial attacks, hereby questioning the efficacy of solely relying on sophisticated alignment methods, and further advocates for a new training idea that integrates modal concepts alongside traditional amodal concepts, aiming to endow LLMs with a more nuanced understanding of real-world contexts and ethical considerations.

As the demand for sophisticated Natural Language Processing (NLP) models continues to grow, so does the need for efficient pre-training techniques. Current NLP models undergo resource-intensive pre-training. In response, we introduce $FastDoc$ (Fast Pre-training Technique using Document-Level Metadata and Taxonomy), a novel approach designed to significantly reduce computational demands. $FastDoc$ leverages document metadata and domain-specific taxonomy as supervision signals. It involves continual pre-training of an open-domain transformer encoder using sentence-level embeddings, followed by fine-tuning using token-level embeddings. We evaluate $FastDoc$ on six tasks across nine datasets spanning three distinct domains. Remarkably, $FastDoc$ achieves remarkable compute reductions of approximately 1,000x, 4,500x, 500x compared to competitive approaches in Customer Support, Scientific, and Legal domains, respectively. Importantly, these efficiency gains do not compromise performance relative to competitive baselines. Furthermore, reduced pre-training data mitigates catastrophic forgetting, ensuring consistent performance in open-domain scenarios. $FastDoc$ offers a promising solution for resource-efficient pre-training, with potential applications spanning various domains.

Recent work on extending coreference resolution across domains and languages relies on annotated data in both the target domain and language. At the same time, pre-trained large language models (LMs) have been reported to exhibit strong zero- and few-shot learning abilities across a wide range of NLP tasks. However, prior work mostly studied this ability using artificial sentence-level datasets such as the Winograd Schema Challenge. In this paper, we assess the feasibility of prompt-based coreference resolution by evaluating instruction-tuned language models on difficult, linguistically-complex coreference benchmarks (e.g., CoNLL-2012). We show that prompting for coreference can outperform current unsupervised coreference systems, although this approach appears to be reliant on high-quality mention detectors. Further investigations reveal that instruction-tuned LMs generalize surprisingly well across domains, languages, and time periods; yet continued fine-tuning of neural models should still be preferred if small amounts of annotated examples are available.

Text-to-SQL parsing, which aims at converting natural language instructions into executable SQLs, has gained increasing attention in recent years. In particular, Codex and ChatGPT have shown impressive results in this task. However, most of the prevalent benchmarks, i.e., Spider, and WikiSQL, focus on database schema with few rows of database contents leaving the gap between academic study and real-world applications. To mitigate this gap, we present Bird, a big benchmark for large-scale database grounded in text-to-SQL tasks, containing 12,751 pairs of text-to-SQL data and 95 databases with a total size of 33.4 GB, spanning 37 professional domains. Our emphasis on database values highlights the new challenges of dirty database contents, external knowledge between NL questions and database contents, and SQL efficiency, particularly in the context of massive databases. To solve these problems, text-to-SQL models must feature database value comprehension in addition to semantic parsing. The experimental results demonstrate the significance of database values in generating accurate text-to-SQLs for big databases. Furthermore, even the most effective text-to-SQL models, i.e. ChatGPT, only achieves 40.08% in execution accuracy, which is still far from the human result of 92.96%, proving that challenges still stand. Besides, we also provide an efficiency analysis to offer insights into generating text-to-efficient-SQLs that are beneficial to industries. We believe that BIRD will contribute to advancing real-world applications of text-to-SQL research. The leaderboard and source code are available: https://bird-bench.github.io/.

Recent breakthroughs in generative modeling have sparked interest in practical single-model attribution. Such methods predict whether a sample was generated by a specific generator or not, for instance, to prove intellectual property theft. However, previous works are either limited to the closed-world setting or require undesirable changes to the generative model. We address these shortcomings by, first, viewing single-model attribution through the lens of anomaly detection. Arising from this change of perspective, we propose FLIPAD, a new approach for single-model attribution in the open-world setting based on final-layer inversion and anomaly detection. We show that the utilized final-layer inversion can be reduced to a convex lasso optimization problem, making our approach theoretically sound and computationally efficient. The theoretical findings are accompanied by an experimental study demonstrating the effectiveness of our approach and its flexibility to various domains.

The rise of large language models (LLMs) has brought a critical need for high-quality human-labeled data, particularly for processes like human feedback and evaluation. A common practice is to label data via consensus annotation over crowdworker judgments. However, annotators' judgments for subjective tasks can differ in many ways: they may have different qualitative judgments about an example, and they may map those to a labeling scheme in different ways. We show that these nuances can be captured by natural language explanations, and propose a method to rescale ordinal annotations and explanations using LLMs. Specifically, we feed annotators' Likert ratings and corresponding explanations into an LLM and prompt it to produce a numeric score anchored in a scoring rubric. These scores should reflect the annotators' underlying assessments of the example. The rubric can be designed or modified after annotation, and include distinctions that may not have been known when the original error taxonomy was devised. We explore our technique in the context of rating system outputs for a document-grounded question answering task, where LLMs achieve near-human performance. Our method rescales the raw judgments without impacting agreement and brings the scores closer to human judgments grounded in the same scoring rubric.

Generative AI systems (ChatGPT, DALL-E, etc) are expanding into multiple areas of our lives, from art Rombach et al. [2021] to mental health Rob Morris and Kareem Kouddous [2022]; their rapidly growing societal impact opens new opportunities, but also raises ethical concerns. The emerging field of AI alignment aims to make AI systems reflect human values. This paper focuses on evaluating the ethics of multimodal AI systems involving both text and images - a relatively under-explored area, as most alignment work is currently focused on language models. We first create a multimodal ethical database from human feedback on ethicality. Then, using this database, we develop algorithms, including a RoBERTa-large classifier and a multilayer perceptron, to automatically assess the ethicality of system responses.

Feedforward neural networks (FNNs) are typically viewed as pure prediction algorithms, and their strong predictive performance has led to their use in many machine-learning applications. However, their flexibility comes with an interpretability trade-off; thus, FNNs have been historically less popular among statisticians. Nevertheless, classical statistical theory, such as significance testing and uncertainty quantification, is still relevant. Supplementing FNNs with methods of statistical inference, and covariate-effect visualisations, can shift the focus away from black-box prediction and make FNNs more akin to traditional statistical models. This can allow for more inferential analysis, and, hence, make FNNs more accessible within the statistical-modelling context.