Large language models (LLMs) that have been trained on a corpus that includes large amount of code exhibit a remarkable ability to understand HTML code. As web interfaces are primarily constructed using HTML, we design an in-depth study to see how LLMs can be used to retrieve and locate important elements for a user given query (i.e. task description) in a web interface. In contrast with prior works, which primarily focused on autonomous web navigation, we decompose the problem as an even atomic operation - Can LLMs identify the important information in the web page for a user given query? This decomposition enables us to scrutinize the current capabilities of LLMs and uncover the opportunities and challenges they present. Our empirical experiments show that while LLMs exhibit a reasonable level of performance in retrieving important UI elements, there is still a substantial room for improvement. We hope our investigation will inspire follow-up works in overcoming the current challenges in this domain.

In software engineering, the meticulous configuration of software tools is crucial in ensuring optimal performance within intricate systems. However, the complexity inherent in selecting optimal configurations is exacerbated by the high-dimensional search spaces presented in modern applications. Conventional trial-and-error or intuition-driven methods are both inefficient and error-prone, impeding scalability and reproducibility. In this study, we embark on an exploration of leveraging Large-Language Models (LLMs) to streamline the software configuration process. We identify that the task of hyperparameter configuration for machine learning components within intelligent applications is particularly challenging due to the extensive search space and performance-critical nature. Existing methods, including Bayesian optimization, have limitations regarding initial setup, computational cost, and convergence efficiency. Our work presents a novel approach that employs LLMs, such as Chat-GPT, to identify starting conditions and narrow down the search space, improving configuration efficiency. We conducted a series of experiments to investigate the variability of LLM-generated responses, uncovering intriguing findings such as potential response caching and consistent behavior based on domain-specific keywords. Furthermore, our results from hyperparameter optimization experiments reveal the potential of LLMs in expediting initialization processes and optimizing configurations. While our initial insights are promising, they also indicate the need for further in-depth investigations and experiments in this domain.

Large language models (LLMs) such as GPT-3, OPT, and LLaMA have demonstrated remarkable accuracy in a wide range of tasks. However, training these models can incur significant expenses, often requiring tens of thousands of GPUs for months of continuous operation. Typically, this training is carried out in specialized GPU clusters equipped with homogeneous high-speed Remote Direct Memory Access (RDMA) network interface cards (NICs). The acquisition and maintenance of such dedicated clusters is challenging. Current LLM training frameworks, like Megatron-LM and Megatron-DeepSpeed, focus primarily on optimizing training within homogeneous cluster settings. In this paper, we introduce Holmes, a training framework for LLMs that employs thoughtfully crafted data and model parallelism strategies over the heterogeneous NIC environment. Our primary technical contribution lies in a novel scheduling method that intelligently allocates distinct computational tasklets in LLM training to specific groups of GPU devices based on the characteristics of their connected NICs. Furthermore, our proposed framework, utilizing pipeline parallel techniques, demonstrates scalability to multiple GPU clusters, even in scenarios without high-speed interconnects between nodes in distinct clusters. We conducted comprehensive experiments that involved various scenarios in the heterogeneous NIC environment. In most cases, our framework achieves performance levels close to those achievable with homogeneous RDMA-capable networks (InfiniBand or RoCE), significantly exceeding training efficiency within the pure Ethernet environment. Additionally, we verified that our framework outperforms other mainstream LLM frameworks under heterogeneous NIC environment in terms of training efficiency and can be seamlessly integrated with them.

Potential harms of Large Language Models such as mass misinformation and plagiarism can be partially mitigated if there exists a reliable way to detect machine generated text. In this paper, we propose a new watermarking method to detect machine-generated texts. Our method embeds a unique pattern within the generated text, ensuring that while the content remains coherent and natural to human readers, it carries distinct markers that can be identified algorithmically. Specifically, we intervene with the token sampling process in a way which enables us to trace back our token choices during the detection phase. We show how watermarking affects textual quality and compare our proposed method with a state-of-the-art watermarking method in terms of robustness and detectability. Through extensive experiments, we demonstrate the effectiveness of our watermarking scheme in distinguishing between watermarked and non-watermarked text, achieving high detection rates while maintaining textual quality.

Transformers have demonstrated effectiveness in in-context solving data-fitting problems from various (latent) models, as reported by Garg et al. (2022). However, the absence of an inherent iterative structure in the transformer architecture presents a challenge in emulating the iterative algorithms, which are commonly employed in traditional machine learning methods. To address this, we propose the utilization of looped transformer architecture and its associated training methodology, with the aim of incorporating iterative characteristics into the transformer architectures. Experimental results suggest that the looped transformer achieves performance comparable to the standard transformer in solving various data-fitting problems, while utilizing less than 10% of the parameter count. Transformers (Vaswani et al., 2017; Brown et al., 2020; Devlin et al., 2019) have emerged as the preferred model in the field of natural language processing (NLP) and other domains requiring sequence-to-sequence modeling. Besides their state-of-art performance in natural language processing tasks, large language models (LLM) such as GPT-3 (Brown et al., 2020) and PaLM (Chowdhery et al., 2022) also exhibit the ability to learn in-context: they can adapt to various downstream tasks based on a brief prompt, thus bypassing the need for additional model fine-tuning. This intriguing ability of in-context learning has sparked interest in the research community, leading numerous studies (Min et al., 2022; Olsson et al., 2022; Li et al., 2023). However, the underlying mechanisms enabling these transformers to perform in-context learning remain unclear. In an effort to understand the in-context learning behavior of LLMs, Garg et al. (2022) investigated the performance of transformers, when trained from scratch, in solving specific function class learning problems in-context. Notably, transformers exhibited strong performance across all tasks, matching or even surpassing traditional solvers. Building on this, Akyürek et al. (2022) explored the transformerbased model's capability to address the linear regression learning problem, interpreting it as an implicit form of established learning algorithms. Their study included both theoretical and empirical perspectives to understand how transformers learn these functions. Subsequently, von Oswald et al. (2022) demonstrated empirically that, when trained to predict the linear function output, a linear self-attention-only transformer inherently learns to perform a single step of gradient descent to solve the linear regression task in-context. While the approach and foundational theory presented by von Oswald et al. (2022) are promising, there exists a significant gap between the simplified architecture they examined and the standard decoder transformer used in practice.

Transformer neural networks can exhibit a surprising capacity for in-context learning (ICL) despite not being explicitly trained for it. Prior work has provided a deeper understanding of how ICL emerges in transformers, e.g., through the lens of mechanistic interpretability, Bayesian inference, or by examining the distributional properties of training data. However, in each of these cases, ICL is treated largely as a persistent phenomenon; namely, once ICL emerges, it is assumed to persist asymptotically. Here, we show that the emergence of ICL during transformer training is, in fact, often transient. We train transformers on synthetic data designed so that both ICL and in-weights learning (IWL) strategies can lead to correct predictions. We find that ICL first emerges, then disappears and gives way to IWL, all while the training loss decreases, indicating an asymptotic preference for IWL. The transient nature of ICL is observed in transformers across a range of model sizes and datasets, raising the question of how much to "overtrain" transformers when seeking compact, cheaper-to-run models. We find that L2 regularization may offer a path to more persistent ICL that removes the need for early stopping based on ICL-style validation tasks. Finally, we present initial evidence that ICL transience may be caused by competition between ICL and IWL circuits.

Membership Inference Attacks (MIA) aim to infer whether a target data record has been utilized for model training or not. Prior attempts have quantified the privacy risks of language models (LMs) via MIAs, but there is still no consensus on whether existing MIA algorithms can cause remarkable privacy leakage on practical Large Language Models (LLMs). Existing MIAs designed for LMs can be classified into two categories: reference-free and reference-based attacks. They are both based on the hypothesis that training records consistently strike a higher probability of being sampled. Nevertheless, this hypothesis heavily relies on the overfitting of target models, which will be mitigated by multiple regularization methods and the generalization of LLMs. The reference-based attack seems to achieve promising effectiveness in LLMs, which measures a more reliable membership signal by comparing the probability discrepancy between the target model and the reference model. However, the performance of reference-based attack is highly dependent on a reference dataset that closely resembles the training dataset, which is usually inaccessible in the practical scenario. Overall, existing MIAs are unable to effectively unveil privacy leakage over practical fine-tuned LLMs that are overfitting-free and private. We propose a Membership Inference Attack based on Self-calibrated Probabilistic Variation (SPV-MIA). Specifically, since memorization in LLMs is inevitable during the training process and occurs before overfitting, we introduce a more reliable membership signal, probabilistic variation, which is based on memorization rather than overfitting. Furthermore, we introduce a self-prompt approach, which constructs the dataset to fine-tune the reference model by prompting the target LLM itself. In this manner, the adversary can collect a dataset with a similar distribution from public APIs.

In this work, we present Transformer-based Powered Descent Guidance (T-PDG), a scalable algorithm for reducing the computational complexity of the direct optimization formulation of the spacecraft powered descent guidance problem. T-PDG uses data from prior runs of trajectory optimization algorithms to train a transformer neural network, which accurately predicts the relationship between problem parameters and the globally optimal solution for the powered descent guidance problem. The solution is encoded as the set of tight constraints corresponding to the constrained minimum-cost trajectory and the optimal final time of landing. By leveraging the attention mechanism of transformer neural networks, large sequences of time series data can be accurately predicted when given only the spacecraft state and landing site parameters. When applied to the real problem of Mars powered descent guidance, T-PDG reduces the time for computing the 3 degree of freedom fuel-optimal trajectory, when compared to lossless convexification, from an order of 1-8 seconds to less than 500 milliseconds. A safe and optimal solution is guaranteed by including a feasibility check in T-PDG before returning the final trajectory.

Large language models (LLMs) show amazing proficiency and fluency in the use of language. Does this mean that they have also acquired insightful linguistic knowledge about the language, to an extent that they can serve as an "expert linguistic annotator"? In this paper, we examine the successes and limitations of the GPT-3, ChatGPT, and GPT-4 models in analysis of sentence meaning structure, focusing on the Abstract Meaning Representation (AMR; Banarescu et al. 2013) parsing formalism, which provides rich graphical representations of sentence meaning structure while abstracting away from surface forms. We compare models' analysis of this semantic structure across two settings: 1) direct production of AMR parses based on zero- and few-shot prompts, and 2) indirect partial reconstruction of AMR via metalinguistic natural language queries (e.g., "Identify the primary event of this sentence, and the predicate corresponding to that event."). Across these settings, we find that models can reliably reproduce the basic format of AMR, and can often capture core event, argument, and modifier structure -- however, model outputs are prone to frequent and major errors, and holistic analysis of parse acceptability shows that even with few-shot demonstrations, models have virtually 0% success in producing fully accurate parses. Eliciting natural language responses produces similar patterns of errors. Overall, our findings indicate that these models out-of-the-box can capture aspects of semantic structure, but there remain key limitations in their ability to support fully accurate semantic analyses or parses.

Large Language Models (LLMs) combined with instruction tuning have made significant progress when generalizing to unseen tasks. However, they have been less successful in Information Extraction (IE), lagging behind task-specific models. Typically, IE tasks are characterized by complex annotation guidelines which describe the task and give examples to humans. Previous attempts to leverage such information have failed, even with the largest models, as they are not able to follow the guidelines out-of-the-box. In this paper we propose GoLLIE (Guideline-following Large Language Model for IE), a model able to improve zero-shot results on unseen IE tasks by virtue of being fine-tuned to comply with annotation guidelines. Comprehensive evaluation empirically demonstrates that GoLLIE is able to generalize to and follow unseen guidelines, outperforming previous attempts at zero-shot information extraction. The ablation study shows that detailed guidelines is key for good results.