We primarily focus on the field of large language models (LLMs) for recommendation, which has been actively explored recently and poses a significant challenge in effectively enhancing recommender systems with logical reasoning abilities and open-world knowledge. Current mainstream efforts mainly center around injecting personalized information from recommendation models into LLMs by customizing input templates or aligning representations between semantic and recommendation spaces at the prediction layer. However, they face three significant limitations: (1) LoRA is mostly used as a core component in existing works, but personalization is not well established in LoRA parameters as the LoRA matrix shared by every user may not cater to different users' characteristics, leading to suboptimal performance. (2) Although lifelong personalized behavior sequences are ideal for personalization, their use raises effectiveness and efficiency issues since LLMs require escalating training and inference time to extend text lengths. (3) Existing approaches aren't scalable for large datasets due to training efficiency constraints. Thus, LLMs only see a small fraction of the datasets (e.g., less than 10%) instead of the whole datasets, limiting their exposure to the full training space. To address these problems, we propose RecLoRA. This model incorporates a Personalized LoRA module that maintains independent LoRAs for different users and a Long-Short Modality Retriever that retrieves different history lengths for different modalities, significantly improving performance while adding minimal time cost. Furthermore, we design a Few2Many Learning Strategy, using a conventional recommendation model as a lens to magnify small training spaces to full spaces. Extensive experiments on public datasets demonstrate the efficacy of our RecLoRA compared to existing baseline models.

This paper extends recent investigations on the emotional reasoning abilities of Large Language Models (LLMs). Current research on LLMs has not directly evaluated the distinction between how LLMs predict the self-attribution of emotions and the perception of others' emotions. We first look at carefully crafted emotion-evoking stimuli, originally designed to find patterns of brain neural activity representing fine-grained inferred emotional attributions of others. We show that GPT-4 is especially accurate in reasoning about such stimuli. This suggests LLMs agree with humans' attributions of others' emotions in stereotypical scenarios remarkably more than self-attributions of emotions in idiosyncratic situations. To further explore this, our second study utilizes a dataset containing annotations from both the author and a third-person perspective. We find that GPT-4's interpretations align more closely with human judgments about the emotions of others than with self-assessments. Notably, conventional computational models of emotion primarily rely on self-reported ground truth as the gold standard. However, an average observer's standpoint, which LLMs appear to have adopted, might be more relevant for many downstream applications, at least in the absence of individual information and adequate safety considerations.

Despite their recent successes, Transformer-based large language models show surprising failure modes. A well-known example of such failure modes is their inability to length-generalize: solving problem instances at inference time that are longer than those seen during training. In this work, we further explore the root cause of this failure by performing a detailed analysis of model behaviors on the simple parity task. Our analysis suggests that length generalization failures are intricately related to a model's inability to perform random memory accesses within its context window. We present supporting evidence for this hypothesis by demonstrating the effectiveness of methodologies that circumvent the need for indexing or that enable random token access indirectly, through content-based addressing. We further show where and how the failure to perform random memory access manifests through attention map visualizations.

Recently, there has been an extensive research effort in building efficient large language model (LLM) inference serving systems. These efforts not only include innovations in the algorithm and software domains but also constitute developments of various hardware acceleration techniques. Nevertheless, there is a lack of simulation infrastructure capable of accurately modeling versatile hardware-software behaviors in LLM serving systems without extensively extending the simulation time. This paper aims to develop an effective simulation tool, called LLMServingSim, to support future research in LLM serving systems. In designing LLMServingSim, we focus on two limitations of existing simulators: (1) they lack consideration of the dynamic workload variations of LLM inference serving due to its autoregressive nature, and (2) they incur repetitive simulations without leveraging algorithmic redundancies in LLMs. To address these limitations, LLMServingSim simulates the LLM serving in the granularity of iterations, leveraging the computation redundancies across decoder blocks and reusing the simulation results from previous iterations. Additionally, LLMServingSim provides a flexible framework that allows users to plug in any accelerator compiler-and-simulation stacks for exploring various system designs with heterogeneous processors. Our experiments demonstrate that LLMServingSim produces simulation results closely following the performance behaviors of real GPU-based LLM serving system with less than 14.7% error rate, while offering 91.5x faster simulation speed compared to existing accelerator simulators.

Assembly code search is vital for reducing the burden on reverse engineers, allowing them to quickly identify specific functions using natural language within vast binary programs. Despite its significance, this critical task is impeded by the complexities involved in building high-quality datasets. This paper explores training a Large Language Model (LLM) to emulate a general compiler. By leveraging Ubuntu packages to compile a dataset of 20 billion tokens, we further continue pre-train CodeLlama as a Virtual Compiler (ViC), capable of compiling any source code of any language to assembly code. This approach allows for virtual compilation across a wide range of programming languages without the need for a real compiler, preserving semantic equivalency and expanding the possibilities for assembly code dataset construction. Furthermore, we use ViC to construct a sufficiently large dataset for assembly code search. Employing this extensive dataset, we achieve a substantial improvement in assembly code search performance, with our model surpassing the leading baseline by 26%.

Conditional question answering (CQA) is an important task that aims to find probable answers and identify conditions that need to be satisfied to support the answer. Existing approaches struggle with CQA due to two main challenges: (1) precisely identifying conditions and their logical relationship, and (2) verifying and solving the conditions. To address these challenges, we propose Chain of Condition, a novel prompting approach by firstly identifying all conditions and constructing their logical relationships explicitly according to the document, then verifying whether these conditions are satisfied, finally solving the logical expression by tools to indicate any missing conditions and generating the answer based on the resolved conditions. The experiments on two benchmark conditional question answering datasets shows chain of condition outperforms existing prompting baselines, establishing a new state-of-the-art. Furthermore, with backbone models like GPT-3.5-Turbo or GPT-4, it surpasses all supervised baselines with only few-shot settings.

Recommender systems aim to estimate the dynamically changing user preferences and sequential dependencies between historical user behaviour and metadata. Although transformer-based models have proven to be effective in sequential recommendations, their state growth is proportional to the length of the sequence that is being processed, which makes them expensive in terms of memory and inference costs. Our research focused on three promising directions in sequential recommendations: enhancing speed through the use of State Space Models (SSM), as they can achieve SOTA results in the sequential recommendations domain with lower latency, memory, and inference costs, as proposed by Liu et al. (2024); improving the quality of recommendations with Large Language Models (LLMs) via Monolithic Preference Optimization without Reference Model (ORPO); and implementing adaptive batch-and step-size algorithms to reduce costs and accelerate training processes. Recently, transformer models have been shown to be effective in sequential recommendation tasks as the backbone of larger models Kang & McAuley (2018) and as individual LLMs Li et al. (2023), Yue et al. (2023). Despite their success, attention-based methods face inference inefficiencies due to the quadratic computational complexity inherent in attention operators and their rapid state growth, which is proportional to the sequence length.

Large Language Models (LLMs) exhibit potentially harmful biases that reinforce culturally inherent stereotypes, cloud moral judgments, or amplify positive evaluations of majority groups. Previous explanations mainly attributed bias in LLMs to human annotators and the selection of training data. Consequently, they have typically been addressed with bottom-up approaches such as reinforcement learning or debiasing corpora. However, these methods only treat the effects of LLM biases by indirectly influencing the model architecture, but do not address the underlying causes in the computational process. Here, we propose metacognitive myopia as a cognitive-ecological framework that can account for a conglomerate of established and emerging LLM biases and provide a lever to address problems in powerful but vulnerable tools. Our theoretical framework posits that a lack of the two components of metacognition, monitoring and control, causes five symptoms of metacognitive myopia in LLMs: integration of invalid tokens and embeddings, susceptibility to redundant information, neglect of base rates in conditional computation, decision rules based on frequency, and inappropriate higher-order statistical inference for nested data structures. As a result, LLMs produce erroneous output that reaches into the daily high-stakes decisions of humans. By introducing metacognitive regulatory processes into LLMs, engineers and scientists can develop precise remedies for the underlying causes of these biases. Our theory sheds new light on flawed human-machine interactions and raises ethical concerns regarding the increasing, imprudent implementation of LLMs in organizational structures.

Large language models (LLMs) have been applied in many fields and have developed rapidly in recent years. As a classic machine learning task, time series forecasting has recently been boosted by LLMs. Recent works treat large language models as \emph{zero-shot} time series reasoners without further fine-tuning, which achieves remarkable performance. However, there are some unexplored research problems when applying LLMs for time series forecasting under the zero-shot setting. For instance, the LLMs' preferences for the input time series are less understood. In this paper, by comparing LLMs with traditional time series forecasting models, we observe many interesting properties of LLMs in the context of time series forecasting. First, our study shows that LLMs perform well in predicting time series with clear patterns and trends, but face challenges with datasets lacking periodicity. This observation can be explained by the ability of LLMs to recognize the underlying period within datasets, which is supported by our experiments. In addition, the input strategy is investigated, and it is found that incorporating external knowledge and adopting natural language paraphrases substantially improve the predictive performance of LLMs for time series. Overall, our study contributes insight into LLMs' advantages and limitations in time series forecasting under different conditions.

This paper critically examines the ethical implications of integrating AI tools like GitHub Copilot and ChatGPT into the software development process. It explores issues such as code ownership, bias, accountability, privacy, and the potential impact on the job market. While these AI tools offer significant benefits in terms of productivity and efficiency, they also introduce complex ethical challenges. The paper argues that addressing these challenges is essential to ensuring that AI's integration into software development is both responsible and beneficial to society.