Large language models (LLMs) have recently revolutionized automated text understanding and generation. The performance of these models relies on the high number of parameters of the underlying neural architectures, which allows LLMs to memorize part of the vast quantity of data seen during the training. This paper investigates whether and to what extent general-purpose pre-trained LLMs have memorized information from known ontologies. Our results show that LLMs partially know ontologies: they can, and do indeed, memorize concepts from ontologies mentioned in the text, but the level of memorization of their concepts seems to vary proportionally to their popularity on the Web, the primary source of their training material. We additionally propose new metrics to estimate the degree of memorization of ontological information in LLMs by measuring the consistency of the output produced across different prompt repetitions, query languages, and degrees of determinism.
相關內容
As Large Language Models (LLMs) are deployed with increasing real-world responsibilities, it is important to be able to specify and constrain the behavior of these systems in a reliable manner. Model developers may wish to set explicit rules for the model, such as "do not generate abusive content", but these may be circumvented by jailbreaking techniques. Existing evaluations of adversarial attacks and defenses on LLMs generally require either expensive manual review or unreliable heuristic checks. To address this issue, we propose Rule-following Language Evaluation Scenarios (RuLES), a programmatic framework for measuring rule-following ability in LLMs. RuLES consists of 14 simple text scenarios in which the model is instructed to obey various rules while interacting with the user. Each scenario has a programmatic evaluation function to determine whether the model has broken any rules in a conversation. Our evaluations of proprietary and open models show that almost all current models struggle to follow scenario rules, even on straightforward test cases. We also demonstrate that simple optimization attacks suffice to significantly increase failure rates on test cases. We conclude by exploring two potential avenues for improvement: test-time steering and supervised fine-tuning.
Vision-Language Models (VLMs) such as GPT-4V have recently demonstrated incredible strides on diverse vision language tasks. We dig into vision-based deductive reasoning, a more sophisticated but less explored realm, and find previously unexposed blindspots in the current SOTA VLMs. Specifically, we leverage Raven's Progressive Matrices (RPMs), to assess VLMs' abilities to perform multi-hop relational and deductive reasoning relying solely on visual clues. We perform comprehensive evaluations of several popular VLMs employing standard strategies such as in-context learning, self-consistency, and Chain-of-thoughts (CoT) on three diverse datasets, including the Mensa IQ test, IntelligenceTest, and RAVEN. The results reveal that despite the impressive capabilities of LLMs in text-based reasoning, we are still far from achieving comparable proficiency in visual deductive reasoning. We found that certain standard strategies that are effective when applied to LLMs do not seamlessly translate to the challenges presented by visual reasoning tasks. Moreover, a detailed analysis reveals that VLMs struggle to solve these tasks mainly because they are unable to perceive and comprehend multiple, confounding abstract patterns in RPM examples.
This study marks a significant advancement by harnessing Large Language Models (LLMs) for multi-intent spoken language understanding (SLU), proposing a unique methodology that capitalizes on the generative power of LLMs within an SLU context. Our innovative technique reconfigures entity slots specifically for LLM application in multi-intent SLU environments and introduces the concept of Sub-Intent Instruction (SII), enhancing the dissection and interpretation of intricate, multi-intent communication within varied domains. The resultant datasets, dubbed LM-MixATIS and LM-MixSNIPS, are crafted from pre-existing benchmarks. Our research illustrates that LLMs can match and potentially excel beyond the capabilities of current state-of-the-art multi-intent SLU models. It further explores LLM efficacy across various intent configurations and dataset proportions. Moreover, we introduce two pioneering metrics, Entity Slot Accuracy (ESA) and Combined Semantic Accuracy (CSA), to provide an in-depth analysis of LLM proficiency in this complex field.
Large language models (LLMs) open up new horizons for sequential recommendations, owing to their remarkable language comprehension and generation capabilities. However, there are still numerous challenges that should be addressed to successfully implement sequential recommendations empowered by LLMs. Firstly, user behavior patterns are often complex, and relying solely on one-step reasoning from LLMs may lead to incorrect or task-irrelevant responses. Secondly, the prohibitively resource requirements of LLM (e.g., ChatGPT-175B) are overwhelmingly high and impractical for real sequential recommender systems. In this paper, we propose a novel Step-by-step knowLedge dIstillation fraMework for recommendation (SLIM), paving a promising path for sequential recommenders to enjoy the exceptional reasoning capabilities of LLMs in a "slim" (i.e., resource-efficient) manner. We introduce CoT prompting based on user behavior sequences for the larger teacher model. The rationales generated by the teacher model are then utilized as labels to distill the downstream smaller student model (e.g., LLaMA2-7B). In this way, the student model acquires the step-by-step reasoning capabilities in recommendation tasks. We encode the generated rationales from the student model into a dense vector, which empowers recommendation in both ID-based and ID-agnostic scenarios. Extensive experiments demonstrate the effectiveness of SLIM over state-of-the-art baselines, and further analysis showcasing its ability to generate meaningful recommendation reasoning at affordable costs.
Large language models (LLMs) have led to a surge in collaborative writing with model assistance. As different users incorporate suggestions from the same model, there is a risk of decreased diversity in the produced content, potentially limiting diverse perspectives in public discourse. In this work, we measure the impact of co-writing on diversity via a controlled experiment, where users write argumentative essays in three setups -- using a base LLM (GPT3), a feedback-tuned LLM (InstructGPT), and writing without model help. We develop a set of diversity metrics and find that writing with InstructGPT (but not the GPT3) results in a statistically significant reduction in diversity. Specifically, it increases the similarity between the writings of different authors and reduces the overall lexical and content diversity. We additionally find that this effect is mainly attributable to InstructGPT contributing less diverse text to co-written essays. In contrast, the user-contributed text remains unaffected by model collaboration. This suggests that the recent improvement in generation quality from adapting models to human feedback might come at the cost of more homogeneous and less diverse content.
This paper explores the cutting-edge Large Language Model with analytical reasoning on sports. Our analytical reasoning embodies the tasks of letting large language models count how many points each team scores in a quarter in the NBA and NFL games. Our major discoveries are in two folds. Firstly, we find among all the models we employed, GPT-4 stands out in effectiveness, followed by Claude-2.1, with GPT-3.5, Gemini-Pro, and Llama-2-70b lagging behind. Specifically, we compare three different prompting techniques and a divide-and-conquer approach, we find that the latter was the most effective. Our divide-and-conquer approach breaks down play-by-play data into smaller, more manageable segments, solves each piece individually, and then aggregates them together. Besides the divide-and-conquer approach, we also explore the Chain of Thought (CoT) strategy, which markedly improves outcomes for certain models, notably GPT-4 and Claude-2.1, with their accuracy rates increasing significantly. However, the CoT strategy has negligible or even detrimental effects on the performance of other models like GPT-3.5 and Gemini-Pro. Secondly, to our surprise, we observe that most models, including GPT-4, struggle to accurately count the total scores for NBA quarters despite showing strong performance in counting NFL quarter scores. This leads us to further investigate the factors that impact the complexity of analytical reasoning tasks with extensive experiments, through which we conclude that task complexity depends on the length of context, the information density, and the presence of related information. Our research provides valuable insights into the complexity of analytical reasoning tasks and potential directions for developing future large language models.
Topic models are a popular tool for clustering and analyzing textual data. They allow texts to be classified on the basis of their affiliation to the previously calculated topics. Despite their widespread use in research and application, an in-depth analysis of topic models is still an open research topic. State-of-the-art methods for interpreting topic models are based on simple visualizations, such as similarity matrices, top-term lists or embeddings, which are limited to a maximum of three dimensions. In this paper, we propose an incidence-geometric method for deriving an ordinal structure from flat topic models, such as non-negative matrix factorization. These enable the analysis of the topic model in a higher (order) dimension and the possibility of extracting conceptual relationships between several topics at once. Due to the use of conceptual scaling, our approach does not introduce any artificial topical relationships, such as artifacts of feature compression. Based on our findings, we present a new visualization paradigm for concept hierarchies based on ordinal motifs. These allow for a top-down view on topic spaces. We introduce and demonstrate the applicability of our approach based on a topic model derived from a corpus of scientific papers taken from 32 top machine learning venues.
Large language models (LLMs) have demonstrated significant potential in the realm of natural language understanding and programming code processing tasks. Their capacity to comprehend and generate human-like code has spurred research into harnessing LLMs for code analysis purposes. However, the existing body of literature falls short in delivering a systematic evaluation and assessment of LLMs' effectiveness in code analysis, particularly in the context of obfuscated code. This paper seeks to bridge this gap by offering a comprehensive evaluation of LLMs' capabilities in performing code analysis tasks. Additionally, it presents real-world case studies that employ LLMs for code analysis. Our findings indicate that LLMs can indeed serve as valuable tools for automating code analysis, albeit with certain limitations. Through meticulous exploration, this research contributes to a deeper understanding of the potential and constraints associated with utilizing LLMs in code analysis, paving the way for enhanced applications in this critical domain.
Large text corpora are the backbone of language models. However, we have a limited understanding of the content of these corpora, including general statistics, quality, social factors, and inclusion of evaluation data (contamination). In this work, we propose What's In My Big Data? (WIMBD), a platform and a set of sixteen analyses that allow us to reveal and compare the contents of large text corpora. WIMBD builds on two basic capabilities -- count and search -- at scale, which allows us to analyze more than 35 terabytes on a standard compute node. We apply WIMBD to ten different corpora used to train popular language models, including C4, The Pile, and RedPajama. Our analysis uncovers several surprising and previously undocumented findings about these corpora, including the high prevalence of duplicate, synthetic, and low-quality content, personally identifiable information, toxic language, and benchmark contamination. For instance, we find that about 50% of the documents in RedPajama and LAION-2B-en are duplicates. In addition, several datasets used for benchmarking models trained on such corpora are contaminated with respect to important benchmarks, including the Winograd Schema Challenge and parts of GLUE and SuperGLUE. We open-source WIMBD's code and artifacts to provide a standard set of evaluations for new text-based corpora and to encourage more analyses and transparency around them.
Language model pre-training has proven to be useful in learning universal language representations. As a state-of-the-art language model pre-training model, BERT (Bidirectional Encoder Representations from Transformers) has achieved amazing results in many language understanding tasks. In this paper, we conduct exhaustive experiments to investigate different fine-tuning methods of BERT on text classification task and provide a general solution for BERT fine-tuning. Finally, the proposed solution obtains new state-of-the-art results on eight widely-studied text classification datasets.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191