To address this, we consider Reverse Image Retrieval (RIR) augmented generation, a simple yet effective strategy to augment MLLMs with web-scale reverse image search results. RIR robustly improves knowledge-intensive visual question answering (VQA) of GPT-4V by 37-43%, GPT-4 Turbo by 25-27%, and GPT-4o by 18-20% in terms of open-ended VQA evaluation metrics. To our surprise, we discover that RIR helps the model to better access its own world knowledge. Concretely, our experiments suggest that RIR augmentation helps by providing further visual and textual cues without necessarily containing the direct answer to a query. In addition, we elucidate cases in which RIR can hurt performance and conduct a human evaluation. Finally, we find that the overall advantage of using RIR makes it difficult for an agent that can choose to use RIR to perform better than an approach where RIR is the default setting.

Large autoregressive models like Transformers can solve tasks through in-context learning (ICL) without learning new weights, suggesting avenues for efficiently solving new tasks. For many tasks, e.g., linear regression, the data factorizes: examples are independent given a task latent that generates the data, e.g., linear coefficients. While an optimal predictor leverages this factorization by inferring task latents, it is unclear if Transformers implicitly do so or if they instead exploit heuristics and statistical shortcuts enabled by attention layers. Both scenarios have inspired active ongoing work. In this paper, we systematically investigate the effect of explicitly inferring task latents. We minimally modify the Transformer architecture with a bottleneck designed to prevent shortcuts in favor of more structured solutions, and then compare performance against standard Transformers across various ICL tasks. Contrary to intuition and some recent works, we find little discernible difference between the two; biasing towards task-relevant latent variables does not lead to better out-of-distribution performance, in general. Curiously, we find that while the bottleneck effectively learns to extract latent task variables from context, downstream processing struggles to utilize them for robust prediction. Our study highlights the intrinsic limitations of Transformers in achieving structured ICL solutions that generalize, and shows that while inferring the right latents aids interpretability, it is not sufficient to alleviate this problem.

Transformer-based foundation models have become crucial for various domains, most notably natural language processing (NLP) or computer vision (CV). These models are predominantly deployed on high-performance GPUs or hardwired accelerators with highly customized, proprietary instruction sets. Until now, limited attention has been given to RISC-V-based general-purpose platforms. In our work, we present the first end-to-end inference results of transformer models on an open-source many-tiny-core RISC-V platform implementing distributed Softmax primitives and leveraging ISA extensions for SIMD floating-point operand streaming and instruction repetition, as well as specialized DMA engines to minimize costly main memory accesses and to tolerate their latency. We focus on two foundational transformer topologies, encoder-only and decoder-only models. For encoder-only models, we demonstrate a speedup of up to 12.8x between the most optimized implementation and the baseline version. We reach over 79% FPU utilization and 294 GFLOPS/W, outperforming State-of-the-Art (SoA) accelerators by more than 2x utilizing the HW platform while achieving comparable throughput per computational unit. For decoder-only topologies, we achieve 16.1x speedup in the Non-Autoregressive (NAR) mode and up to 35.6x speedup in the Autoregressive (AR) mode compared to the baseline implementation. Compared to the best SoA dedicated accelerator, we achieve 2.04x higher FPU utilization.

Aspect-based sentiment analysis (ABSA) identifies sentiment information related to specific aspects and provides deeper market insights to businesses and organizations. With the emergence of large language models (LMs), recent studies have proposed using fixed examples for instruction tuning to reformulate ABSA as a generation task. However, the performance is sensitive to the selection of in-context examples; several retrieval methods are based on surface similarity and are independent of the LM generative objective. This study proposes an instruction learning method with retrieval-based example ranking for ABSA tasks. For each target sample, an LM was applied as a scorer to estimate the likelihood of the output given the input and a candidate example as the prompt, and training examples were labeled as positive or negative by ranking the scores. An alternating training schema is proposed to train both the retriever and LM. Instructional prompts can be constructed using high-quality examples. The LM is used for both scoring and inference, improving the generation efficiency without incurring additional computational costs or training difficulties. Extensive experiments on three ABSA subtasks verified the effectiveness of the proposed method, demonstrating its superiority over various strong baseline models. Code and data are released at https://github.com/zgMin/IT-RER-ABSA.

The introduction of Encoder-Decoder architectures in natural language processing [26] (both recurrent [6] and Transformer-based [29]) has motivated researchers to apply them to software engineering. One important application is generating summaries of code [25, 2, 11]. A code summarization tool is useful for example to understand legacy code or to create documentation. Since the spread of Large Language Models (LLMs), the working programmer has many more opportunities to use deep learning-based tools. Closed models (such as GPT-4 [21] or Gemini [27]) and open models (such as CodeLlama [24] or WizardCoder [19]) demonstrate impressive capabilities of generating source code based on a task description, as well as generating natural-language summary of code. The main objective of this technical report is to investigate how well open-sourced LLMs handle source code in relation with natural language text. In particular, we discuss results of some of the most acknowledged open-source LLMs, focusing on their code summarization/explanation (code-to-text) capabilities. We also discuss code generation (text-to-code) capabilities of these LLMs, as this is often considered to be their most defining capability. That is, LLMs are often ranked simply based on results on a code generation benchmark.

Recently, the concept of artificial assistants has evolved from science fiction into real-world applications. GPT-4o, the newest multimodal large language model (MLLM) across audio, vision, and text, has further blurred the line between fiction and reality by enabling more natural human-computer interactions. However, the advent of GPT-4o's voice mode may also introduce a new attack surface. In this paper, we present the first systematic measurement of jailbreak attacks against the voice mode of GPT-4o. We show that GPT-4o demonstrates good resistance to forbidden questions and text jailbreak prompts when directly transferring them to voice mode. This resistance is primarily due to GPT-4o's internal safeguards and the difficulty of adapting text jailbreak prompts to voice mode. Inspired by GPT-4o's human-like behaviors, we propose VoiceJailbreak, a novel voice jailbreak attack that humanizes GPT-4o and attempts to persuade it through fictional storytelling (setting, character, and plot). VoiceJailbreak is capable of generating simple, audible, yet effective jailbreak prompts, which significantly increases the average attack success rate (ASR) from 0.033 to 0.778 in six forbidden scenarios. We also conduct extensive experiments to explore the impacts of interaction steps, key elements of fictional writing, and different languages on VoiceJailbreak's effectiveness and further enhance the attack performance with advanced fictional writing techniques. We hope our study can assist the research community in building more secure and well-regulated MLLMs.

Aligning human preference and value is an important requirement for contemporary foundation models. State-of-the-art techniques such as Reinforcement Learning from Human Feedback (RLHF) often consist of two stages: 1) supervised fine-tuning (SFT), where the model is fine-tuned by learning from human demonstration data; 2) Preference learning, where preference data is used to learn a reward model, which is in turn used by a reinforcement learning (RL) step to fine-tune the model. Such reward model serves as a proxy to human preference, and it is critical to guide the RL step towards improving the model quality. In this work, we argue that the SFT stage significantly benefits from learning a reward model as well. Instead of using the human demonstration data directly via supervised learning, we propose to leverage an Inverse Reinforcement Learning (IRL) technique to (explicitly or implicitly) build an reward model, while learning the policy model. This approach leads to new SFT algorithms that are not only efficient to implement, but also promote the ability to distinguish between the preferred and non-preferred continuations. Moreover, we identify a connection between the proposed IRL based approach, and certain self-play approach proposed recently, and showed that self-play is a special case of modeling a reward-learning agent. Theoretically, we show that the proposed algorithms converge to the stationary solutions of the IRL problem. Empirically, we align 1B and 7B models using proposed methods and evaluate them on a reward benchmark model and the HuggingFace Open LLM Leaderboard. The proposed methods show significant performance improvement over existing SFT approaches. Our results indicate that it is beneficial to explicitly or implicitly leverage reward learning throughout the entire alignment process.

Citation practices are crucial in shaping the structure of scientific knowledge, yet they are often influenced by contemporary norms and biases. The emergence of Large Language Models (LLMs) like GPT-4 introduces a new dynamic to these practices. Interestingly, the characteristics and potential biases of references recommended by LLMs that entirely rely on their parametric knowledge, and not on search or retrieval-augmented generation, remain unexplored. Here, we analyze these characteristics in an experiment using a dataset of 166 papers from AAAI, NeurIPS, ICML, and ICLR, published after GPT-4's knowledge cut-off date, encompassing 3,066 references in total. In our experiment, GPT-4 was tasked with suggesting scholarly references for the anonymized in-text citations within these papers. Our findings reveal a remarkable similarity between human and LLM citation patterns, but with a more pronounced high citation bias in GPT-4, which persists even after controlling for publication year, title length, number of authors, and venue. Additionally, we observe a large consistency between the characteristics of GPT-4's existing and non-existent generated references, indicating the model's internalization of citation patterns. By analyzing citation graphs, we show that the references recommended by GPT-4 are embedded in the relevant citation context, suggesting an even deeper conceptual internalization of the citation networks. While LLMs can aid in citation generation, they may also amplify existing biases and introduce new ones, potentially skewing scientific knowledge dissemination. Our results underscore the need for identifying the model's biases and for developing balanced methods to interact with LLMs in general.

Data selection has emerged as a core issue for large-scale visual-language model pretaining (e.g., CLIP), particularly with noisy web-curated datasets. Three main data selection approaches are: (1) leveraging external non-CLIP models to aid data selection, (2) training new CLIP-style embedding models that are more effective at selecting high-quality data than the original OpenAI CLIP model, and (3) designing better metrics or strategies universally applicable to any CLIP embedding without requiring specific model properties (e.g., CLIPScore is one popular metric). While the first two approaches have been extensively studied, the third remains under-explored. In this paper, we advance the third approach by proposing two new methods. Firstly, instead of classical CLIP scores that only consider the alignment between two modalities from a single sample, we introduce negCLIPLoss, a CLIP loss-inspired method that adds the alignment between one sample and its contrastive pairs as an extra normalization term for better quality measurement. Secondly, when downstream tasks are known, we propose a new norm-based metric, NormSim, to measure the similarity between pretraining data and target data. We test our methods on the data selection benchmark, DataComp~\cite{gadre2023datacomp}. Compared to the best baseline using only OpenAI's CLIP-L/14, our methods achieve a 5.3\% improvement on ImageNet-1k and a 2.8\% improvement on 38 downstream evaluation tasks. Moreover, both negCLIPLoss and NormSim are compatible with existing techniques. By combining our methods with the current best methods DFN~\cite{fang2023data} and HYPE~\cite{kim2024hype}, we can boost average performance on downstream tasks by 0.9\%, achieving a new state-of-the-art.

Since the inception of the Transformer architecture in 2017, Large Language Models (LLMs) such as GPT and BERT have evolved significantly, impacting various industries with their advanced capabilities in language understanding and generation. These models have shown potential to transform the medical field, highlighting the necessity for specialized evaluation frameworks to ensure their effective and ethical deployment. This comprehensive survey delineates the extensive application and requisite evaluation of LLMs within healthcare, emphasizing the critical need for empirical validation to fully exploit their capabilities in enhancing healthcare outcomes. Our survey is structured to provide an in-depth analysis of LLM applications across clinical settings, medical text data processing, research, education, and public health awareness. We begin by exploring the roles of LLMs in various medical applications, detailing their evaluation based on performance in tasks such as clinical diagnosis, medical text data processing, information retrieval, data analysis, and educational content generation. The subsequent sections offer a comprehensive discussion on the evaluation methods and metrics employed, including models, evaluators, and comparative experiments. We further examine the benchmarks and datasets utilized in these evaluations, providing a categorized description of benchmarks for tasks like question answering, summarization, information extraction, bioinformatics, information retrieval and general comprehensive benchmarks. This structure ensures a thorough understanding of how LLMs are assessed for their effectiveness, accuracy, usability, and ethical alignment in the medical domain. ...