Automatic prompt engineering aims to enhance the generation quality of large language models (LLMs). Recent works utilize feedbacks generated from erroneous cases to guide the prompt optimization. During inference, they may further retrieve several semantically-related exemplars and concatenate them to the optimized prompts to improve the performance. However, those works only utilize the feedback at the current step, ignoring historical and unseleccted feedbacks which are potentially beneficial. Moreover, the selection of exemplars only considers the general semantic relationship and may not be optimal in terms of task performance and matching with the optimized prompt. In this work, we propose an Exemplar-Guided Reflection with Memory mechanism (ERM) to realize more efficient and accurate prompt optimization. Specifically, we design an exemplar-guided reflection mechanism where the feedback generation is additionally guided by the generated exemplars. We further build two kinds of memory to fully utilize the historical feedback information and support more effective exemplar retrieval. Empirical evaluations show our method surpasses previous state-of-the-arts with less optimization steps, i.e., improving F1 score by 10.1 on LIAR dataset, and reducing half of the optimization steps on ProTeGi.

Vision Language Models (VLMs), pre-trained on large-scale image-text datasets, enable zero-shot predictions for unseen data but may underperform on specific unseen tasks. Continual learning (CL) can help VLMs effectively adapt to new data distributions without joint training, but faces challenges of catastrophic forgetting and generalization forgetting. Although significant progress has been achieved by distillation-based methods, they exhibit two severe limitations. One is the popularly adopted single-teacher paradigm fails to impart comprehensive knowledge, The other is the existing methods inadequately leverage the multimodal information in the original training dataset, instead they rely on additional data for distillation, which increases computational and storage overhead. To mitigate both limitations, by drawing on Knowledge Integration Theory (KIT), we propose a Multi-Stage Knowledge Integration network (MulKI) to emulate the human learning process in distillation methods. MulKI achieves this through four stages, including Eliciting Ideas, Adding New Ideas, Distinguishing Ideas, and Making Connections. During the four stages, we first leverage prototypes to align across modalities, eliciting cross-modal knowledge, then adding new knowledge by constructing fine-grained intra- and inter-modality relationships with prototypes. After that, knowledge from two teacher models is adaptively distinguished and re-weighted. Finally, we connect between models from intra- and inter-task, integrating preceding and new knowledge. Our method demonstrates significant improvements in maintaining zero-shot capabilities while supporting continual learning across diverse downstream tasks, showcasing its potential in adapting VLMs to evolving data distributions.

Set theory is foundational to mathematics and, when sets are finite, to reasoning about the world. An intelligent system should perform set operations consistently, regardless of superficial variations in the operands. Initially designed for semantically-oriented NLP tasks, large language models (LLMs) are now being evaluated on algorithmic tasks. Because sets are comprised of arbitrary symbols (e.g. numbers, words), they provide an opportunity to test, systematically, the invariance of LLMs' algorithmic abilities under simple lexical or semantic variations. To this end, we present the SetLexSem Challenge, a synthetic benchmark that evaluates the performance of LLMs on set operations. SetLexSem assesses the robustness of LLMs' instruction-following abilities under various conditions, focusing on the set operations and the nature and construction of the set members. Evaluating seven LLMs with SetLexSem, we find that they exhibit poor robustness to variation in both operation and operands. We show -- via the framework's systematic sampling of set members along lexical and semantic dimensions -- that LLMs are not only not robust to variation along these dimensions but demonstrate unique failure modes in particular, easy-to-create semantic groupings of "deceptive" sets. We find that rigorously measuring language model robustness to variation in frequency and length is challenging and present an analysis that measures them independently. The code for reproducing the results of this paper, and for generating the SetLexSem Challenge dataset, is available at \href{https://github.com/amazon-science/SetLexSem-Challenge}{https://github.com/amazon-science/SetLexSem-Challenge}.

Category-agnostic pose estimation (CAPE) has traditionally relied on support images with annotated keypoints, a process that is often cumbersome and may fail to fully capture the necessary correspondences across diverse object categories. Recent efforts have begun exploring the use of text-based queries, where the need for support keypoints is eliminated. However, the optimal use of textual descriptions for keypoints remains an underexplored area. In this work, we introduce CapeLLM, a novel approach that leverages a text-based multimodal large language model (MLLM) for CAPE. Our method only employs query image and detailed text descriptions as an input to estimate category-agnostic keypoints. We conduct extensive experiments to systematically explore the design space of LLM-based CAPE, investigating factors such as choosing the optimal description for keypoints, neural network architectures, and training strategies. Thanks to the advanced reasoning capabilities of the pre-trained MLLM, CapeLLM demonstrates superior generalization and robust performance. Our approach sets a new state-of-the-art on the MP-100 benchmark in the challenging 1-shot setting, marking a significant advancement in the field of category-agnostic pose estimation.

As modern web services increasingly rely on REST APIs, their thorough testing has become crucial. Furthermore, the advent of REST API specifications such as the OpenAPI Specification has led to the emergence of many black-box REST API testing tools. However, these tools often focus on individual test elements in isolation (e.g., APIs, parameters, values), resulting in lower coverage and less effectiveness in detecting faults (i.e., 500 response codes). To address these limitations, we present AutoRestTest, the first black-box framework to adopt a dependency-embedded multi-agent approach for REST API testing, integrating Multi-Agent Reinforcement Learning (MARL) with a Semantic Property Dependency Graph (SPDG) and Large Language Models (LLMs). Our approach treats REST API testing as a separable problem, where four agents -- API, dependency, parameter, and value -- collaborate to optimize API exploration. LLMs handle domain-specific value restrictions, the SPDG model simplifies the search space for dependencies using a similarity score between API operations, and MARL dynamically optimizes the agents' behavior. Evaluated on 12 real-world REST services, AutoRestTest outperforms the four leading black-box REST API testing tools, including those assisted by RESTGPT (which augments realistic test inputs using LLMs), in terms of code coverage, operation coverage, and fault detection. Notably, AutoRestTest is the only tool able to identify an internal server error in Spotify. Our ablation study underscores the significant contributions of the agent learning, SPDG, and LLM components.

When making predictions, a language model must trade off how much it relies on its context vs. its prior knowledge. Choosing how sensitive the model is to its context is a fundamental functionality, as it enables the model to excel at tasks like retrieval-augmented generation and question-answering. In this paper, we search for a knob which controls this sensitivity, determining whether language models answer from the context or their prior knowledge. To guide this search, we design a task for controllable context sensitivity. In this task, we first feed the model a context (Paris is in England) and a question (Where is Paris?); we then instruct the model to either use its prior or contextual knowledge and evaluate whether it generates the correct answer for both intents (either France or England). When fine-tuned on this task, instruction-tuned versions of Llama-3.1, Mistral-v0.3, and Gemma-2 can solve it with high accuracy (85-95%). Analyzing these high-performing models, we narrow down which layers may be important to context sensitivity using a novel linear time algorithm. Then, in each model, we identify a 1-D subspace in a single layer that encodes whether the model follows context or prior knowledge. Interestingly, while we identify this subspace in a fine-tuned model, we find that the exact same subspace serves as an effective knob in not only that model but also non-fine-tuned instruct and base models of that model family. Finally, we show a strong correlation between a model's performance and how distinctly it separates context-agreeing from context-ignoring answers in this subspace. These results suggest a single subspace facilitates how the model chooses between context and prior knowledge, hinting at a simple fundamental mechanism that controls this behavior.

Although text-to-image (T2I) models exhibit remarkable generation capabilities, they frequently fail to accurately bind semantically related objects or attributes in the input prompts; a challenge termed semantic binding. Previous approaches either involve intensive fine-tuning of the entire T2I model or require users or large language models to specify generation layouts, adding complexity. In this paper, we define semantic binding as the task of associating a given object with its attribute, termed attribute binding, or linking it to other related sub-objects, referred to as object binding. We introduce a novel method called Token Merging (ToMe), which enhances semantic binding by aggregating relevant tokens into a single composite token. This ensures that the object, its attributes and sub-objects all share the same cross-attention map. Additionally, to address potential confusion among main objects with complex textual prompts, we propose end token substitution as a complementary strategy. To further refine our approach in the initial stages of T2I generation, where layouts are determined, we incorporate two auxiliary losses, an entropy loss and a semantic binding loss, to iteratively update the composite token to improve the generation integrity. We conducted extensive experiments to validate the effectiveness of ToMe, comparing it against various existing methods on the T2I-CompBench and our proposed GPT-4o object binding benchmark. Our method is particularly effective in complex scenarios that involve multiple objects and attributes, which previous methods often fail to address. The code will be publicly available at \url{https://github.com/hutaihang/ToMe}.

There is great interest in fine-tuning frontier large language models (LLMs) to inject new information and update existing knowledge. While commercial LLM fine-tuning APIs from providers such as OpenAI and Google promise flexible adaptation for various applications, the efficacy of fine-tuning remains unclear. In this study, we introduce FineTuneBench, an evaluation framework and dataset for understanding how well commercial fine-tuning APIs can successfully learn new and updated knowledge. We analyze five frontier LLMs with commercially available fine-tuning APIs, including GPT-4o and Gemini 1.5 Pro, on their effectiveness in two settings: (1) ingesting novel information, such as recent news events and new people profiles, and (2) updating existing knowledge, such as updated medical guidelines and code frameworks. Our results reveal substantial shortcomings in all the models' abilities to effectively learn new information through fine-tuning, with an average generalization accuracy of 37% across all models. Overall, fine-tuning GPT-4o mini is the most effective for infusing new knowledge and updating knowledge, followed by GPT-3.5 Turbo and GPT-4o. The fine-tuning APIs for Gemini 1.5 Flesh and Gemini 1.5 Pro are unable to learn new knowledge or update existing knowledge. These findings underscore a major shortcoming in using current commercial fine-tuning services to achieve reliable knowledge infusion in common scenarios. As LLMs are increasingly used in diverse domains such as software development Kelly (2024) and medicine Gliadkovskaya (2024), it is important they contain up-to-date and relevant knowledge.

We evaluate the reliability of two chatbots, ChatGPT (4o and o1-preview versions), and Gemini Advanced, in providing references on financial literature and employing novel methodologies. Alongside the conventional binary approach commonly used in the literature, we developed a nonbinary approach and a recency measure to assess how hallucination rates vary with how recent a topic is. After analyzing 150 citations, ChatGPT-4o had a hallucination rate of 20.0% (95% CI, 13.6%-26.4%), while the o1-preview had a hallucination rate of 21.3% (95% CI, 14.8%-27.9%). In contrast, Gemini Advanced exhibited higher hallucination rates: 76.7% (95% CI, 69.9%-83.4%). While hallucination rates increased for more recent topics, this trend was not statistically significant for Gemini Advanced. These findings emphasize the importance of verifying chatbot-provided references, particularly in rapidly evolving fields.

This study investigates the linguistic understanding of Large Language Models (LLMs) regarding signifier (form) and signified (meaning) by distinguishing two LLM evaluation paradigms: psycholinguistic and neurolinguistic. Traditional psycholinguistic evaluations often reflect statistical biases that may misrepresent LLMs' true linguistic capabilities. We introduce a neurolinguistic approach, utilizing a novel method that combines minimal pair and diagnostic probing to analyze activation patterns across model layers. This method allows for a detailed examination of how LLMs represent form and meaning, and whether these representations are consistent across languages. Our contributions are three-fold: (1) We compare neurolinguistic and psycholinguistic methods, revealing distinct patterns in LLM assessment; (2) We demonstrate that LLMs exhibit higher competence in form compared to meaning, with the latter largely correlated to the former; (3) We present new conceptual minimal pair datasets for Chinese (COMPS-ZH) and German (COMPS-DE), complementing existing English datasets.