Low-rank adaptation (LoRA) has emerged as a new paradigm for cost-efficient finetuning of large language models (LLMs). However, fine-tuned LLMs often become overconfident especially when fine-tuned on small datasets. Bayesian methods, with their inherent ability to estimate uncertainty, serve as potent tools to mitigate overconfidence and enhance calibration. In this work, we introduce Laplace-LoRA, which applies a Bayesian approach to the LoRA parameters. Specifically, Laplace-LoRA applies a Laplace approximation to the posterior over the LoRA parameters, considerably improving the calibration of fine-tuned LLMs. In recent years, fine-tuning large language models (LLMs) have become increasingly important (Houlsby et al., 2019; Hu et al., 2021; Liu et al., 2022; Ding et al., 2022; 2023). Fine-tuning is used both to adapt LLMs for specific tasks and to create general instruction-following models (e.g. using Reinforcement Learning from Human Feedback; RLHF Wei et al., 2021; Ouyang et al., 2022; Chung et al., 2022; Wang et al., 2022). However, fine-tuned LLMs have a notable limitation: they often exhibit overconfidence (Jiang et al., 2021; Xiao et al., 2022; He et al., 2023; Tian et al., 2023; OpenAI, 2023). This is particularly problematic in safety-critical applications or when making decisions in areas where limited data is available, such as medical diagnosis, finance and experimental design (Singhal et al., 2022; Wu et al., 2023; Lampinen et al., 2023; Li et al., 2022). Consequently, there is an urgent need for strategies that enhance the calibration of fine-tuned LLMs, ensuring that their predictions are as trustworthy as they are powerful. Bayesian deep learning is commonly proposed as a solution to overconfidence in deep networks (e.g. Historically, the field of Bayesian deep learning has frequently considered ResNets for image classification (Shridhar et al., 2019; Dusenberry et al., 2020; Izmailov et al., 2021).

Prompt engineering has emerged as an indispensable technique for extending the capabilities of large language models (LLMs) and vision-language models (VLMs). This approach leverages task-specific instructions, known as prompts, to enhance model efficacy without modifying the core model parameters. Rather than updating the model parameters, prompts allow seamless integration of pre-trained models into downstream tasks by eliciting desired model behaviors solely based on the given prompt. Prompts can be natural language instructions that provide context to guide the model or learned vector representations that activate relevant knowledge. This burgeoning field has enabled success across various applications, from question-answering to commonsense reasoning. However, there remains a lack of systematic organization and understanding of the diverse prompt engineering methods and techniques. This survey paper addresses the gap by providing a structured overview of recent advancements in prompt engineering, categorized by application area. For each prompting approach, we provide a summary detailing the prompting methodology, its applications, the models involved, and the datasets utilized. We also delve into the strengths and limitations of each approach and include a taxonomy diagram and table summarizing datasets, models, and critical points of each prompting technique. This systematic analysis enables a better understanding of this rapidly developing field and facilitates future research by illuminating open challenges and opportunities for prompt engineering.

Machine learning has enabled the development of powerful systems capable of editing images from natural language instructions. However, in many common scenarios it is difficult for users to specify precise image transformations with text alone. For example, in an image with several dogs, it is difficult to select a particular dog and move it to a precise location. Doing this with text alone would require a complex prompt that disambiguates the target dog and describes the destination. However, direct manipulation is well suited to visual tasks like selecting objects and specifying locations. We introduce Point and Instruct, a system for seamlessly combining familiar direct manipulation and textual instructions to enable precise image manipulation. With our system, a user can visually mark objects and locations, and reference them in textual instructions. This allows users to benefit from both the visual descriptiveness of natural language and the spatial precision of direct manipulation.

Vision-Language Models (VLMs) excel in generating textual responses from visual inputs, yet their versatility raises significant security concerns. This study takes the first step in exposing VLMs' susceptibility to data poisoning attacks that can manipulate responses to innocuous, everyday prompts. We introduce Shadowcast, a stealthy data poisoning attack method where poison samples are visually indistinguishable from benign images with matching texts. Shadowcast demonstrates effectiveness in two attack types. The first is Label Attack, tricking VLMs into misidentifying class labels, such as confusing Donald Trump for Joe Biden. The second is Persuasion Attack, which leverages VLMs' text generation capabilities to craft narratives, such as portraying junk food as health food, through persuasive and seemingly rational descriptions. We show that Shadowcast are highly effective in achieving attacker's intentions using as few as 50 poison samples. Moreover, these poison samples remain effective across various prompts and are transferable across different VLM architectures in the black-box setting. This work reveals how poisoned VLMs can generate convincing yet deceptive misinformation and underscores the importance of data quality for responsible deployments of VLMs. Our code is available at: https://github.com/umd-huang-lab/VLM-Poisoning.

Large Language Models (LLMs) are nowadays expected to generate content aligned with human preferences. Current work focuses on alignment at model training time, through techniques such as Reinforcement Learning with Human Feedback (RLHF). However, it is unclear if such methods are an effective choice to teach alignment objectives to the model. First, the inability to incorporate multiple, custom rewards and reliance on a model developer's view of universal and static principles are key limitations. Second, the residual gaps in model training and the reliability of such approaches are also questionable (e.g. susceptibility to jail-breaking even after safety training). To address these, we propose DeAL, a framework that allows the user to customize reward functions and enables Decoding-time Alignment of LLMs (DeAL). At its core, we view decoding as a heuristic-guided search process and facilitate the use of a wide variety of alignment objectives. Our experiments with programmatic constraints such as keyword and length constraints (studied widely in the pre-LLM era) and abstract objectives such as harmlessness and helpfulness (proposed in the post-LLM era) show that we can DeAL with fine-grained trade-offs, improve adherence to alignment objectives, and address residual gaps in LLMs. Lastly, while DeAL can be effectively paired with RLHF and prompting techniques, its generality makes decoding slower, an optimization we leave for future work.

Large Language Models (LLMs) have shown remarkable promise in their ability to interact proficiently with humans. Subsequently, their potential use as artificial confederates and surrogates in sociological experiments involving conversation is an exciting prospect. But how viable is this idea? This paper endeavors to test the limits of current-day LLMs with a pre-registered study integrating real people with LLM agents acting as people. The study focuses on debate-based opinion consensus formation in three environments: humans only, agents and humans, and agents only. Our goal is to understand how LLM agents influence humans, and how capable they are in debating like humans. We find that LLMs can blend in and facilitate human productivity but are less convincing in debate, with their behavior ultimately deviating from human's. We elucidate these primary failings and anticipate that LLMs must evolve further before being viable debaters.

With the exponential growth in data volume and the emergence of data-intensive applications, particularly in the field of machine learning, concerns related to resource utilization, privacy, and fairness have become paramount. This paper focuses on the textual domain of data and addresses challenges regarding encoding sentences to their optimized representations through the lens of information-theory. In particular, we use empirical estimates of mutual information, using the Donsker-Varadhan definition of Kullback-Leibler divergence. Our approach leverages this estimation to train an information-theoretic sentence embedding, called TexShape, for (task-based) data compression or for filtering out sensitive information, enhancing privacy and fairness. In this study, we employ a benchmark language model for initial text representation, complemented by neural networks for information-theoretic compression and mutual information estimations. Our experiments demonstrate significant advancements in preserving maximal targeted information and minimal sensitive information over adverse compression ratios, in terms of predictive accuracy of downstream models that are trained using the compressed data.

Large Language Models (LLMs) have ushered in a new era of text annotation, as their ease-of-use, high accuracy, and relatively low costs have meant that their use has exploded in recent months. However, the rapid growth of the field has meant that LLM-based annotation has become something of an academic Wild West: the lack of established practices and standards has led to concerns about the quality and validity of research. Researchers have warned that the ostensible simplicity of LLMs can be misleading, as they are prone to bias, misunderstandings, and unreliable results. Recognizing the transformative potential of LLMs, this paper proposes a comprehensive set of standards and best practices for their reliable, reproducible, and ethical use. These guidelines span critical areas such as model selection, prompt engineering, structured prompting, prompt stability analysis, rigorous model validation, and the consideration of ethical and legal implications. The paper emphasizes the need for a structured, directed, and formalized approach to using LLMs, aiming to ensure the integrity and robustness of text annotation practices, and advocates for a nuanced and critical engagement with LLMs in social scientific research.

Textbook question answering (TQA) is a challenging task in artificial intelligence due to the complex nature of context and multimodal data. Although previous research has significantly improved the task, there are still some limitations including the models' weak reasoning and inability to capture contextual information in the lengthy context. The introduction of large language models (LLMs) has revolutionized the field of AI, however, directly applying LLMs often leads to inaccurate answers. This paper proposes a methodology that handle the out-of-domain scenario in TQA where concepts are spread across different lessons by incorporating the retrieval augmented generation (RAG) technique and utilize transfer learning to handle the long context and enhance reasoning abilities. Through supervised fine-tuning of the LLM model Llama-2 and the incorporation of RAG, our architecture outperforms the baseline, achieving a 4.12% accuracy improvement on validation set and 9.84% on test set for non-diagram multiple-choice questions.

This paper introduces a novel paradigm for depression detection and treatment using advanced Large Language Models (LLMs): Generative Pre-trained Transformer 4 (GPT-4), Llama 2 chat, and Gemini. These LLMs are fine-tuned with specialized prompts to diagnose, explain, and suggest therapeutic interventions for depression. A unique few-shot prompting method enhances the models' ability to analyze and explain depressive symptoms based on the DSM-5 criteria. In the interaction phase, the models engage in empathetic dialogue management, drawing from resources like PsychDB and a Cognitive Behavioral Therapy (CBT) Guide, fostering supportive interactions with individuals experiencing major depressive disorders. Additionally, the research introduces the Illuminate Database, enriched with various CBT modules, aiding in personalized therapy recommendations. The study evaluates LLM performance using metrics such as F1 scores, Precision, Recall, Cosine similarity, and Recall-Oriented Understudy for Gisting Evaluation (ROUGE) across different test sets, demonstrating their effectiveness. This comprehensive approach blends cutting-edge AI with established psychological methods, offering new possibilities in mental health care and showcasing the potential of LLMs in revolutionizing depression diagnosis and treatment strategies.