When selecting data for training large-scale models, standard practice is to filter for examples that match human notions of data quality. Such filtering yields qualitatively clean datapoints that intuitively should improve model behavior. However, in practice the opposite can often happen: we find that selecting according to similarity with "high quality" data sources may not increase (and can even hurt) performance compared to randomly selecting data. To develop better methods for selecting data, we start by framing dataset selection as an optimization problem that we can directly solve for: given target tasks, a learning algorithm, and candidate data, select the subset that maximizes model performance. This framework thus avoids handpicked notions of data quality, and instead models explicitly how the learning process uses train datapoints to predict on the target tasks. Our resulting method greatly improves language model (LM) performance on both pre-specified tasks and previously unseen tasks. Specifically, choosing target tasks representative of standard LM problems and evaluating on diverse held-out benchmarks, our selected datasets provide a 2x compute multiplier over baseline methods.

VLMs (Vision-Language Models) extend the capabilities of LLMs (Large Language Models) to accept multimodal inputs. Since it has been verified that LLMs can be induced to generate harmful or inaccurate content through specific test cases (termed as Red Teaming), how VLMs perform in similar scenarios, especially with their combination of textual and visual inputs, remains a question. To explore this problem, we present a novel red teaming dataset RTVLM, which encompasses 10 subtasks (e.g., image misleading, multi-modal jail-breaking, face fairness, etc) under 4 primary aspects (faithfulness, privacy, safety, fairness). Our RTVLM is the first red-teaming dataset to benchmark current VLMs in terms of these 4 different aspects. Detailed analysis shows that 10 prominent open-sourced VLMs struggle with the red teaming in different degrees and have up to 31% performance gap with GPT-4V. Additionally, we simply apply red teaming alignment to LLaVA-v1.5 with Supervised Fine-tuning (SFT) using RTVLM, and this bolsters the models' performance with 10% in RTVLM test set, 13% in MM-Hal, and without noticeable decline in MM-Bench, overpassing other LLaVA-based models with regular alignment data. This reveals that current open-sourced VLMs still lack red teaming alignment. Our code and datasets will be open-source.

This survey paper delves into the burgeoning field of explainability for Large Language Models (LLMs), a critical yet challenging aspect of natural language processing. With LLMs playing a pivotal role in various applications, their "black-box" nature raises concerns about transparency and ethical use. This paper emphasizes the necessity for enhanced explainability in LLMs, addressing both the general public's trust and the technical community's need for a deeper understanding of these models. We concentrate on pre-trained Transformer-based LLMs, such as LLaMA, which present unique interpretability challenges due to their scale and complexity. Our review categorizes existing explainability methods and discusses their application in improving model transparency and reliability. We also discuss representative evaluation methods, highlighting their strengths and limitations. The goal of this survey is to bridge the gap between theoretical understanding and practical application, offering insights for future research and development in the field of LLM explainability.

Language models (LMs) can solve tasks such as answering questions about tables or images by writing programs. However, using primitive functions often leads to verbose and error-prone programs, and higher-level functions require expert design. To enable better solutions without human labor, we ask code LMs to curate reusable high-level functions, and use them to write solutions. We present TROVE, a training-free method of inducing a verifiable and efficient toolbox of functions, by generating via using, growing, and periodically trimming the toolbox. On 11 datasets from math, table question answering, and image reasoning tasks, TROVE consistently yields simpler solutions with higher accuracy than baselines using CODELLAMA and previous methods using GPT, while using 79-98% smaller toolboxes. TROVE further enables 31% faster and 13% more accurate human verification than baselines. With the same pipeline, it creates diverse functions for varied tasks and datasets, providing insights into their individual characteristics.

Large Language Models (LLMs) have demonstrated impressive performance in natural language processing tasks by leveraging chain of thought (CoT) that enables step-by-step thinking. Extending LLMs with multimodal capabilities is the recent interest, but incurs computational cost and requires substantial hardware resources. To address these challenges, we propose KAM-CoT a framework that integrates CoT reasoning, Knowledge Graphs (KGs), and multiple modalities for a comprehensive understanding of multimodal tasks. KAM-CoT adopts a two-stage training process with KG grounding to generate effective rationales and answers. By incorporating external knowledge from KGs during reasoning, the model gains a deeper contextual understanding reducing hallucinations and enhancing the quality of answers. This knowledge-augmented CoT reasoning empowers the model to handle questions requiring external context, providing more informed answers. Experimental findings show KAM-CoT outperforms the state-of-the-art methods. On the ScienceQA dataset, we achieve an average accuracy of 93.87%, surpassing GPT-3.5 (75.17%) by 18% and GPT-4 (83.99%) by 10%. Remarkably, KAM-CoT achieves these results with only 280M trainable parameters at a time, demonstrating its cost-efficiency and effectiveness.

The proliferation of open-source Large Language Models (LLMs) from various institutions has highlighted the urgent need for comprehensive evaluation methods. However, current evaluation platforms, such as the widely recognized HuggingFace open LLM leaderboard, neglect a crucial aspect -- uncertainty, which is vital for thoroughly assessing LLMs. To bridge this gap, we introduce a new benchmarking approach for LLMs that integrates uncertainty quantification. Our examination involves eight LLMs (LLM series) spanning five representative natural language processing tasks. Additionally, we introduce an uncertainty-aware evaluation metric, UAcc, which takes into account both prediction accuracy and prediction uncertainty. Our findings reveal that: I) LLMs with higher accuracy may exhibit lower certainty; II) Larger-scale LLMs may display greater uncertainty compared to their smaller counterparts; and III) Instruction-finetuning tends to increase the uncertainty of LLMs. By taking uncertainty into account, our new UAcc metric can either amplify or diminish the relative improvement of one LLM over another and may even change the relative ranking of two LLMs. These results underscore the significance of incorporating uncertainty in the evaluation of LLMs.

In the era of large models, the autoregressive nature of decoding often results in latency serving as a significant bottleneck. We propose a non-autoregressive LM-fused ASR system that effectively leverages the parallelization capabilities of accelerator hardware. Our approach combines the Universal Speech Model (USM) and the PaLM 2 language model in per-segment scoring mode, achieving an average relative WER improvement across all languages of 10.8% on FLEURS and 3.6% on YouTube captioning. Furthermore, our comprehensive ablation study analyzes key parameters such as LLM size, context length, vocabulary size, fusion methodology. For instance, we explore the impact of LLM size ranging from 128M to 340B parameters on ASR performance. This study provides valuable insights into the factors influencing the effectiveness of practical large-scale LM-fused speech recognition systems.

Increased availability of open-source software repositories and recent advances in code analysis using large language models (LLMs) has triggered a wave of new work to automate software engineering tasks that were previously very difficult to automate. In this paper, we investigate a recent line of work that hypothesises that comparing the probability of code generated by LLMs with the probability the current code would have had can indicate potential quality problems. We investigate the association between the cross-entropy of code generated by ten different models (based on GPT2 and Llama2) and the following quality aspects: readability, understandability, complexity, modularisation, and overall maintainability assessed by experts and available in an benchmark dataset. Our results show that, controlling for the number of logical lines of codes (LLOC), cross-entropy computed by LLMs is indeed a predictor of maintainability on a class level (the higher the cross-entropy the lower the maintainability). However, this relation is reversed when one does not control for LLOC (e.g., comparing small classes with longer ones). Furthermore, while the complexity of LLMs affects the range of cross-entropy (smaller models tend to have a wider range of cross-entropy), this plays a significant role in predicting maintainability aspects. Our study limits itself on ten different pretrained models (based on GPT2 and Llama2) and on maintainability aspects collected by Schnappinger et al. When controlling for logical lines of code (LLOC), cross-entropy is a predictor of maintainability. However, while related work has shown the potential usefulness of cross-entropy at the level of tokens or short sequences, at the class level this criterion alone may prove insufficient to predict maintainability and further research is needed to make best use of this information in practice.

The task of rumour verification in social media concerns assessing the veracity of a claim on the basis of conversation threads that result from it. While previous work has focused on predicting a veracity label, here we reformulate the task to generate model-centric, free-text explanations of a rumour's veracity. We follow an unsupervised approach by first utilising post-hoc explainability methods to score the most important posts within a thread and then we use these posts to generate informative explanatory summaries by employing template-guided summarisation. To evaluate the informativeness of the explanatory summaries, we exploit the few-shot learning capabilities of a large language model (LLM). Our experiments show that LLMs can have similar agreement to humans in evaluating summaries. Importantly, we show that explanatory abstractive summaries are more informative and better reflect the predicted rumour veracity than just using the highest ranking posts in the thread.

Graph analysis is fundamental in real-world applications. Traditional approaches rely on SPARQL-like languages or clicking-and-dragging interfaces to interact with graph data. However, these methods either require users to possess high programming skills or support only a limited range of graph analysis functionalities. To address the limitations, we propose a large language model (LLM)-based framework called ChatGraph. With ChatGraph, users can interact with graphs through natural language, making it easier to use and more flexible than traditional approaches. The core of ChatGraph lies in generating chains of graph analysis APIs based on the understanding of the texts and graphs inputted in the user prompts. To achieve this, ChatGraph consists of three main modules: an API retrieval module that searches for relevant APIs, a graph-aware LLM module that enables the LLM to comprehend graphs, and an API chain-oriented finetuning module that guides the LLM in generating API chains.