We present SPHINX, a versatile multi-modal large language model (MLLM) with a joint mixing of model weights, tuning tasks, and visual embeddings. First, for stronger vision-language alignment, we unfreeze the large language model (LLM) during pre-training, and introduce a weight mix strategy between LLMs trained by real-world and synthetic data. By directly integrating the weights from two domains, the mixed LLM can efficiently incorporate diverse semantics with favorable robustness. Then, to enable multi-purpose capabilities, we mix a variety of tasks for joint visual instruction tuning, and design task-specific instructions to avoid inter-task conflict. In addition to the basic visual question answering, we include more challenging tasks such as region-level understanding, caption grounding, document layout detection, and human pose estimation, contributing to mutual enhancement over different scenarios. Additionally, we propose to extract comprehensive visual embeddings from various network architectures, pre-training paradigms, and information granularity, providing language models with more robust image representations. Based on our proposed joint mixing, SPHINX exhibits superior multi-modal understanding capabilities on a wide range of applications. On top of this, we further propose an efficient strategy aiming to better capture fine-grained appearances of high-resolution images. With a mixing of different scales and high-resolution sub-images, SPHINX attains exceptional visual parsing and reasoning performance on existing evaluation benchmarks. We hope our work may cast a light on the exploration of joint mixing in future MLLM research. Code is released at https://github.com/Alpha-VLLM/LLaMA2-Accessory.

We present MM-Navigator, a GPT-4V-based agent for the smartphone graphical user interface (GUI) navigation task. MM-Navigator can interact with a smartphone screen as human users, and determine subsequent actions to fulfill given instructions. Our findings demonstrate that large multimodal models (LMMs), specifically GPT-4V, excel in zero-shot GUI navigation through its advanced screen interpretation, action reasoning, and precise action localization capabilities. We first benchmark MM-Navigator on our collected iOS screen dataset. According to human assessments, the system exhibited a 91\% accuracy rate in generating reasonable action descriptions and a 75\% accuracy rate in executing the correct actions for single-step instructions on iOS. Additionally, we evaluate the model on a subset of an Android screen navigation dataset, where the model outperforms previous GUI navigators in a zero-shot fashion. Our benchmark and detailed analyses aim to lay a robust groundwork for future research into the GUI navigation task. The project page is at https://github.com/zzxslp/MM-Navigator.

Recent studies have demonstrated that large language models (LLMs) excel in diverse tasks through in-context learning (ICL) facilitated by task-specific prompts and examples. However, the existing literature shows that ICL encounters performance deterioration when exposed to adversarial inputs. Enhanced performance has been observed when ICL is augmented with natural language explanations (NLEs) (we refer to it as X-ICL). Thus, this work investigates whether X-ICL can improve the robustness of LLMs on a suite of seven adversarial and challenging natural language inference datasets. Moreover, we introduce a new approach to X-ICL by prompting an LLM (ChatGPT in our case) with few human-generated NLEs to produce further NLEs (we call it ChatGPT few-shot), which we show superior to both ChatGPT zero-shot and human-generated NLEs alone. We evaluate five popular LLMs (GPT3.5-turbo, LLaMa2, Vicuna, Zephyr, Mistral) and show that X-ICL with ChatGPT few-shot yields over 6% improvement over ICL. Furthermore, while prompt selection strategies were previously shown to significantly improve ICL on in-distribution test sets, we show that these strategies do not match the efficacy of the X-ICL paradigm in robustness-oriented evaluations.

Chain-of-thought (COT) prompting can help large language models (LLMs) reason toward correct answers, but its efficacy in reasoning toward incorrect answers is unexplored. This strategy of process of elimination (PoE), when used with COT, has the potential to enhance interpretability in tasks like medical diagnoses of exclusion. Thus, we propose PoE with COT, a new task where LLMs must reason toward incorrect options on multiple-choice questions. We evaluate the ability of GPT-3.5, LLaMA-2, and Falcon to perform PoE with COT on 2-choice commonsense and scientific reasoning datasets. We show that PoE consistently underperforms directly choosing the correct answer. The agreement of these strategies is also lower than the self-consistency of each strategy. To study these issues further, we conduct an error analysis and give suggestions for future work.

Enterprise applications of Large Language Models (LLMs) hold promise for question answering on enterprise SQL databases. However, the extent to which LLMs can accurately respond to enterprise questions in such databases remains unclear, given the absence of suitable Text-to-SQL benchmarks tailored to enterprise settings. Additionally, the potential of Knowledge Graphs (KGs) to enhance LLM-based question answering by providing business context is not well understood. This study aims to evaluate the accuracy of LLM-powered question answering systems in the context of enterprise questions and SQL databases, while also exploring the role of knowledge graphs in improving accuracy. To achieve this, we introduce a benchmark comprising an enterprise SQL schema in the insurance domain, a range of enterprise queries encompassing reporting to metrics, and a contextual layer incorporating an ontology and mappings that define a knowledge graph. Our primary finding reveals that question answering using GPT-4, with zero-shot prompts directly on SQL databases, achieves an accuracy of 16%. Notably, this accuracy increases to 54% when questions are posed over a Knowledge Graph representation of the enterprise SQL database. Therefore, investing in Knowledge Graph provides higher accuracy for LLM powered question answering systems.

While large language models exhibit remarkable performance in the Question Answering task, they are susceptible to hallucinations. Challenges arise when these models grapple with understanding multi-hop relations in complex questions or lack the necessary knowledge for a comprehensive response. To address this issue, we introduce the "Decompose-and-Query" framework (D&Q). This framework guides the model to think and utilize external knowledge similar to ReAct, while also restricting its thinking to reliable information, effectively mitigating the risk of hallucinations. Experiments confirm the effectiveness of D&Q: On our ChitChatQA dataset, D&Q does not lose to ChatGPT in 67% of cases; on the HotPotQA question-only setting, D&Q achieved an F1 score of 59.6%. Our code is available at https://github.com/alkaidpku/DQ-ToolQA.

Next-word probabilities from language models have been shown to successfully simulate human reading behavior. Building on this, we show that, interestingly, instruction-tuned large language models (LLMs) yield worse psychometric predictive power (PPP) for human reading behavior than base LLMs with equivalent perplexities. In other words, instruction tuning, which helps LLMs provide human-preferred responses, does not always make them human-like from the computational psycholinguistics perspective. In addition, we explore prompting methodologies in simulating human reading behavior with LLMs, showing that prompts reflecting a particular linguistic hypothesis lead LLMs to exhibit better PPP but are still worse than base LLMs. These highlight that recent instruction tuning and prompting do not offer better estimates than direct probability measurements from base LLMs in cognitive modeling.

Multi-modal large language models (LLM) have achieved powerful capabilities for visual semantic understanding in recent years. However, little is known about how LLMs comprehend visual information and interpret different modalities of features. In this paper, we propose a new method for identifying multi-modal neurons in transformer-based multi-modal LLMs. Through a series of experiments, We highlight three critical properties of multi-modal neurons by four well-designed quantitative evaluation metrics. Furthermore, we introduce a knowledge editing method based on the identified multi-modal neurons, for modifying a specific token to another designative token. We hope our findings can inspire further explanatory researches on understanding mechanisms of multi-modal LLMs.

Large language models (LLMs) can explain their own predictions, through post-hoc or Chain-of-Thought (CoT) explanations. However the LLM could make up reasonably sounding explanations that are unfaithful to its underlying reasoning. Recent work has designed tests that aim to judge the faithfulness of either post-hoc or CoT explanations. In this paper we argue that existing faithfulness tests are not actually measuring faithfulness in terms of the models' inner workings, but only evaluate their self-consistency on the output level. The aims of our work are two-fold. i) We aim to clarify the status of existing faithfulness tests in terms of model explainability, characterising them as self-consistency tests instead. This assessment we underline by constructing a Comparative Consistency Bank for self-consistency tests that for the first time compares existing tests on a common suite of 11 open-source LLMs and 5 datasets -- including ii) our own proposed self-consistency measure CC-SHAP. CC-SHAP is a new fine-grained measure (not test) of LLM self-consistency that compares a model's input contributions to answer prediction and generated explanation. With CC-SHAP, we aim to take a step further towards measuring faithfulness with a more interpretable and fine-grained method. Code available at \url{https://github.com/Heidelberg-NLP/CC-SHAP}

Recently, there has been a rapid advancement in research on Large Language Models (LLMs), resulting in significant progress in several Natural Language Processing (NLP) tasks. Consequently, there has been a surge in LLM evaluation research to comprehend the models' capabilities and limitations. However, much of this research has been confined to the English language, leaving LLM building and evaluation for non-English languages relatively unexplored. There has been an introduction of several new LLMs, necessitating their evaluation on non-English languages. This study aims to expand our MEGA benchmarking suite by including six new datasets to form the MEGAVERSE benchmark. The benchmark comprises 22 datasets covering 81 languages, including low-resource African languages. We evaluate several state-of-the-art LLMs like GPT-3.5-Turbo, GPT4, PaLM2, and Llama2 on the MEGAVERSE datasets. Additionally, we include two multimodal datasets in the benchmark and assess the performance of the LLaVa-v1.5 model. Our experiments suggest that GPT4 and PaLM2 outperform the Llama models on various tasks, notably on low-resource languages, with GPT4 outperforming PaLM2 on more datasets than vice versa. However, issues such as data contamination must be addressed to obtain an accurate assessment of LLM performance on non-English languages.