This study investigates the cognitive plausibility of a pretrained multimodal model, CLIP, in recognizing emotions evoked by abstract visual art. We employ a dataset comprising images with associated emotion labels and textual rationales of these labels provided by human annotators. We perform linguistic analyses of rationales, zero-shot emotion classification of images and rationales, apply similarity-based prediction of emotion, and investigate color-emotion associations. The relatively low, yet above baseline, accuracy in recognizing emotion for abstract images and rationales suggests that CLIP decodes emotional complexities in a manner not well aligned with human cognitive processes. Furthermore, we explore color-emotion interactions in images and rationales. Expected color-emotion associations, such as red relating to anger, are identified in images and texts annotated with emotion labels by both humans and CLIP, with the latter showing even stronger interactions. Our results highlight the disparity between human processing and machine processing when connecting image features and emotions.

Pretesting involves a small-scale trial of data collection procedures, aiming to assess them. It is a standard practice in both academic and applied research (Grimm 2010), and the output of the pretest is usually the feedback offered by interviewers on how to improve procedures and questions. The rapid advancements in generative artificial intelligence (GAI) have opened up new avenues for enhancing various aspects of research, including the design and evaluation of survey questionnaires. AI technologies like large language models (LLMs) have demonstrated remarkable potential in generating human-like text, offering a promising approach to pretesting survey instruments. This article explores the use of GPT models as a tool for pretesting survey questionnaires. Illustrated with two applications, it suggests incorporating GPT feedback as an additional stage before human pretesting, potentially reducing successive iterations. However, the article emphasizes the indispensable role of researchers' judgment in implementing AIgenerated feedback. GPT is an LLM that utilizes advanced algorithms to generate texts that mimic the syntax, semantics, and grammar of human writing, which are approximated by statistical patterns learned from training data (for a technical review, see OpenAI 2023). Like most of the LLMs, GPT models predict the next word in a sequence based on the preceding words.

Quantization can accelerate large language model (LLM) inference. Going beyond INT8 quantization, the research community is actively exploring even lower precision, such as INT4. Nonetheless, state-of-the-art INT4 quantization techniques only accelerate low-batch, edge LLM inference, failing to deliver performance gains in large-batch, cloud-based LLM serving. We uncover a critical issue: existing INT4 quantization methods suffer from significant runtime overhead (20-90%) when dequantizing either weights or partial sums on GPUs. To address this challenge, we introduce QoQ, a W4A8KV4 quantization algorithm with 4-bit weight, 8-bit activation, and 4-bit KV cache. QoQ stands for quattuor-octo-quattuor, which represents 4-8-4 in Latin. QoQ is implemented by the QServe inference library that achieves measured speedup. The key insight driving QServe is that the efficiency of LLM serving on GPUs is critically influenced by operations on low-throughput CUDA cores. Building upon this insight, in QoQ algorithm, we introduce progressive quantization that can allow low dequantization overhead in W4A8 GEMM. Additionally, we develop SmoothAttention to effectively mitigate the accuracy degradation incurred by 4-bit KV quantization. In the QServe system, we perform compute-aware weight reordering and take advantage of register-level parallelism to reduce dequantization latency. We also make fused attention memory-bound, harnessing the performance gain brought by KV4 quantization. As a result, QServe improves the maximum achievable serving throughput of Llama-3-8B by 1.2x on A100, 1.4x on L40S; and Qwen1.5-72B by 2.4x on A100, 3.5x on L40S, compared to TensorRT-LLM. Remarkably, QServe on L40S GPU can achieve even higher throughput than TensorRT-LLM on A100. Thus, QServe effectively reduces the dollar cost of LLM serving by 3x. Code is available at https://github.com/mit-han-lab/qserve.

Existing solutions (e.g., [11, 21, 25, Crafting effective prompts for code generation or editing with Large 30]), while useful, fall short in providing user feedback during the Language Models (LLMs) is not an easy task. Particularly, the absence prompt crafting process. The lack of such feedback can lead to of immediate, stable feedback during prompt crafting hinders uncertainty about the expected output and potential inefficiencies effective interaction, as users are left to mentally imagine possible in the development process, as users are left to mentally imagine outcomes until the code is generated. In response, we introduce possible outcomes until the code is generated [12, 29]. Language-Oriented Code Sketching, an interactive approach that provides However, providing user feedback during prompt crafting presents instant, incremental feedback in the form of code sketches (i.e., significant challenges. Ideally, feedback should not only be instant incomplete code outlines) during prompt crafting. This approach but also evolve incrementally as the user types their prompt, offering converts a prompt into a code sketch by leveraging the inherent clear and understandable intermediate results that progressively linguistic structures within the prompt and applying classic natural guide the LLM towards the final desired code.

Question answering (QA) and Machine Reading Comprehension (MRC) tasks have significantly advanced in recent years due to the rapid development of deep learning techniques and, more recently, large language models. At the same time, many benchmark datasets have become available for QA and MRC tasks. However, most existing large-scale benchmark datasets have been created predominantly using synchronous document collections like Wikipedia or the Web. Archival document collections, such as historical newspapers, contain valuable information from the past that is still not widely used to train large language models. To further contribute to advancing QA and MRC tasks and to overcome the limitation of previous datasets, we introduce ChroniclingAmericaQA, a large-scale temporal QA dataset with 487K question-answer pairs created based on the historical newspaper collection Chronicling America. Our dataset is constructed from a subset of the Chronicling America newspaper collection spanning 120 years. One of the significant challenges for utilizing digitized historical newspaper collections is the low quality of OCR text. Therefore, to enable realistic testing of QA models, our dataset can be used in three different ways: answering questions from raw and noisy content, answering questions from cleaner, corrected version of the content, as well as answering questions from scanned images of newspaper pages. This and the fact that ChroniclingAmericaQA spans the longest time period among available QA datasets make it quite a unique and useful resource.

By leveraging attention in recent years due to their impressive this numerical text quality score, we demonstrate performance in various natural language processing how it can be used to prune the original dataset, (NLP) tasks (Zhang et al., 2022; Penedo et al., enabling the training of LMs using only a fraction 2023; Touvron et al., 2023; Zhou et al., 2023; Liu of the data. Our approach aims to identify et al., 2019). However, their training process often and eliminate low-quality text instances, thereby relies on computationally intensive procedures that streamlining the training process and mitigating the involve massive datasets and compute requirements burden of handling large-scale datasets. We also remove which hinders training large scale LMs on noisy potentially harmful content from the data by real-world or domain specific datasets. What's ensuring that harmful content is rated poorly by our worse is that several of these datasets are uncurated text quality score which can then be pruned. We and may contain harmful content which the observe an absolute improvement of 0.9% averaged LM model can potentially pick up during the training over 14 downstream evaluation tasks for multiple process (Deshpande et al., 2023; Schramowski LM models while using 40% lesser data and training et al., 2022; Kuchnik et al., 2023).

Quantitatively explaining the strength of arguments under gradual semantics has recently received increasing attention. Specifically, several works in the literature provide quantitative explanations by computing the attribution scores of arguments. These works disregard the importance of attacks and supports, even though they play an essential role when explaining arguments' strength. In this paper, we propose a novel theory of Relation Attribution Explanations (RAEs), adapting Shapley values from game theory to offer fine-grained insights into the role of attacks and supports in quantitative bipolar argumentation towards obtaining the arguments' strength. We show that RAEs satisfy several desirable properties. We also propose a probabilistic algorithm to approximate RAEs efficiently. Finally, we show the application value of RAEs in fraud detection and large language models case studies.

By training on text in various languages, large language models (LLMs) typically possess multilingual support and demonstrate remarkable capabilities in solving tasks described in different languages. However, LLMs can exhibit linguistic discrimination due to the uneven distribution of training data across languages. That is, LLMs are hard to keep the consistency of responses when faced with the same task but depicted in different languages. In this study, we first explore the consistency in the LLMs' outputs responding to queries in various languages from two aspects: safety and quality. We conduct this analysis with two datasets (AdvBench and NQ) based on four LLMs (Llama2-13b, Gemma-7b, GPT-3.5-turbo and Gemini-pro). The results show that LLMs exhibit stronger human alignment capabilities with queries in English, French, Russian, and Spanish (only 1.04\% of harmful queries successfully jailbreak on average) compared to queries in Bengali, Georgian, Nepali and Maithili (27.7\% of harmful queries jailbreak successfully on average). Moreover, for queries in English, Danish, Czech and Slovenian, LLMs tend to produce responses with a higher quality (with 0.1494 $F_1$ score on average) compared to the other languages. Upon these findings, we propose LDFighter, a similarity-based voting, to mitigate the linguistic discrimination in LLMs. LDFighter ensures consistent service for different language speakers. We evaluate LDFighter with both benign queries and harmful queries. The results show that LDFighter not only significantly reduces the jailbreak success rate but also improve the response quality on average, demonstrating its effectiveness.

Does language help make sense of the visual world? How important is it to actually see the world rather than having it described with words? These basic questions about the nature of intelligence have been difficult to answer because we only had one example of an intelligent system -- humans -- and limited access to cases that isolated language or vision. However, the development of sophisticated Vision-Language Models (VLMs) by artificial intelligence researchers offers us new opportunities to explore the contributions that language and vision make to learning about the world. We ablate components from the cognitive architecture of these models to identify their contributions to learning new tasks from limited data. We find that a language model leveraging all components recovers a majority of a VLM's performance, despite its lack of visual input, and that language seems to allow this by providing access to prior knowledge and reasoning.

Recent advancements in Large Language Models (LLMs) have revolutionized the AI field but also pose potential safety and ethical risks. Deciphering LLMs' embedded values becomes crucial for assessing and mitigating their risks. Despite extensive investigation into LLMs' values, previous studies heavily rely on human-oriented value systems in social sciences. Then, a natural question arises: Do LLMs possess unique values beyond those of humans? Delving into it, this work proposes a novel framework, ValueLex, to reconstruct LLMs' unique value system from scratch, leveraging psychological methodologies from human personality/value research. Based on Lexical Hypothesis, ValueLex introduces a generative approach to elicit diverse values from 30+ LLMs, synthesizing a taxonomy that culminates in a comprehensive value framework via factor analysis and semantic clustering. We identify three core value dimensions, Competence, Character, and Integrity, each with specific subdimensions, revealing that LLMs possess a structured, albeit non-human, value system. Based on this system, we further develop tailored projective tests to evaluate and analyze the value inclinations of LLMs across different model sizes, training methods, and data sources.