Large Language Model (LLM) pretraining traditionally relies on autoregressive language modeling on randomly sampled data blocks from web-scale datasets. We take inspiration from human learning techniques like spaced repetition to hypothesize that random data sampling for LLMs leads to high training cost and low quality models which tend to forget data. In order to effectively commit web-scale information to long-term memory, we propose the LFR (Learn, Focus, and Review) pedagogy, a new dynamic training paradigm which focuses and repeatedly reviews complex data blocks at systematic intervals based on the model's learning pace and progress. LFR records the model perplexities for different data blocks and frequently revisits blocks with higher perplexity which are more likely to be forgotten. We pretrain the GPT-2 models (124M - 1.5B) from scratch on the OpenWebText dataset using LFR. We test on downstream tasks from the language modeling, question answering, translation, and problem solving domains to achieve consistently lower perplexity and higher accuracy than the baseline OpenAI models, while obtaining a 20x pretraining speed-up.

Knowledge representation has been a central aim of AI since its inception. Symbolic Knowledge Graphs (KGs) and neural Large Language Models (LLMs) can both represent knowledge. KGs provide highly accurate and explicit knowledge representation, but face scalability issue; while LLMs offer expansive coverage of knowledge, but incur significant training costs and struggle with precise and reliable knowledge manipulation. To this end, we introduce OneEdit, a neural-symbolic prototype system for collaborative knowledge editing using natural language, which facilitates easy-to-use knowledge management with KG and LLM. OneEdit consists of three modules: 1) The Interpreter serves for user interaction with natural language; 2) The Controller manages editing requests from various users, leveraging the KG with rollbacks to handle knowledge conflicts and prevent toxic knowledge attacks; 3) The Editor utilizes the knowledge from the Controller to edit KG and LLM. We conduct experiments on two new datasets with KGs which demonstrate that OneEdit can achieve superior performance.

This paper presents the solution of our team APEX in the Meta KDD CUP 2024: CRAG Comprehensive RAG Benchmark Challenge. The CRAG benchmark addresses the limitations of existing QA benchmarks in evaluating the diverse and dynamic challenges faced by Retrieval-Augmented Generation (RAG) systems. It provides a more comprehensive assessment of RAG performance and contributes to advancing research in this field. We propose a routing-based domain and dynamic adaptive RAG pipeline, which performs specific processing for the diverse and dynamic nature of the question in all three stages: retrieval, augmentation, and generation. Our method achieved superior performance on CRAG and ranked 2nd for Task 2&3 on the final competition leaderboard. Our implementation is available at this link: https://github.com/USTCAGI/CRAG-in-KDD-Cup2024.

Large language models (LLMs) can extract information from veterinary electronic health records (EHRs), but performance differences between models, the effect of temperature settings, and the influence of text ambiguity have not been previously evaluated. This study addresses these gaps by comparing the performance of GPT-4 omni (GPT-4o) and GPT-3.5 Turbo under different conditions and investigating the relationship between human interobserver agreement and LLM errors. The LLMs and five humans were tasked with identifying six clinical signs associated with Feline chronic enteropathy in 250 EHRs from a veterinary referral hospital. At temperature 0, the performance of GPT-4o compared to the majority opinion of human respondents, achieved 96.9% sensitivity (interquartile range [IQR] 92.9-99.3%), 97.6% specificity (IQR 96.5-98.5%), 80.7% positive predictive value (IQR 70.8-84.6%), 99.5% negative predictive value (IQR 99.0-99.9%), 84.4% F1 score (IQR 77.3-90.4%), and 96.3% balanced accuracy (IQR 95.0-97.9%). The performance of GPT-4o was significantly better than that of its predecessor, GPT-3.5 Turbo, particularly with respect to sensitivity where GPT-3.5 Turbo only achieved 81.7% (IQR 78.9-84.8%). Adjusting the temperature for GPT-4o did not significantly impact classification performance. GPT-4o demonstrated greater reproducibility than human pairs regardless of temperature, with an average Cohen's kappa of 0.98 (IQR 0.98-0.99) at temperature 0 compared to 0.8 (IQR 0.78-0.81) for humans. Most GPT-4o errors occurred in instances where humans disagreed (35/43 errors, 81.4%), suggesting that these errors were more likely caused by ambiguity of the EHR than explicit model faults. Using GPT-4o to automate information extraction from veterinary EHRs is a viable alternative to manual extraction.

Large language models (LLMs) are a promising venue for natural language understanding and generation. However, current LLMs are far from reliable: they are prone to generating non-factual information and, more crucially, to contradicting themselves when prompted to reason about relations between entities of the world. These problems are currently addressed with large scale fine-tuning or by delegating reasoning to external tools. In this work, we strive for a middle ground and introduce a loss based on neuro-symbolic reasoning that teaches an LLM to be logically consistent with an external set of facts and rules and improves self-consistency even when the LLM is fine-tuned on a limited set of facts. Our approach also allows to easily combine multiple logical constraints at once in a principled way, delivering LLMs that are more consistent w.r.t. all constraints and improve over several baselines w.r.t. a given constraint. Moreover, our method allows LLMs to extrapolate to unseen but semantically similar factual knowledge, represented in unseen datasets, more systematically.

The rapid advancement of Intelligent Transportation Systems (ITS) presents challenges, particularly with missing data in multi-modal transportation and the complexity of handling diverse sequential tasks within a centralized framework. To address these issues, we propose the Spatial-Temporal Large Language Model Diffusion (STLLM-DF), an innovative model that leverages Denoising Diffusion Probabilistic Models (DDPMs) and Large Language Models (LLMs) to improve multi-task transportation prediction. The DDPM's robust denoising capabilities enable it to recover underlying data patterns from noisy inputs, making it particularly effective in complex transportation systems. Meanwhile, the non-pretrained LLM dynamically adapts to spatial-temporal relationships within multi-modal networks, allowing the system to efficiently manage diverse transportation tasks in both long-term and short-term predictions. Extensive experiments demonstrate that STLLM-DF consistently outperforms existing models, achieving an average reduction of 2.40\% in MAE, 4.50\% in RMSE, and 1.51\% in MAPE. This model significantly advances centralized ITS by enhancing predictive accuracy, robustness, and overall system performance across multiple tasks, thus paving the way for more effective spatio-temporal traffic forecasting through the integration of frozen transformer language models and diffusion techniques.

Attention mechanisms and Transformer architectures have revolutionized Natural Language Processing (NLP) by enabling exceptional modeling of long-range dependencies and capturing intricate linguistic patterns. However, their inherent reliance on linear operations in the form of matrix multiplications limits their ability to fully capture inter-token relationships on their own. We propose MöbiusAttention, a novel approach that integrates Möbius transformations within the attention mechanism of Transformer-based models. Möbius transformations are non-linear operations in spaces over complex numbers with the ability to map between various geometries. By incorporating these properties, MöbiusAttention empowers models to learn more intricate geometric relationships between tokens and capture a wider range of information through complex-valued weight vectors. We build and pre-train a BERT and a RoFormer version enhanced with MöbiusAttention, which we then finetune on the GLUE benchmark. We evaluate empirically our approach against the baseline BERT and RoFormer models on a range of downstream tasks. Our approach compares favorably against the baseline models, even with smaller number of parameters suggesting the enhanced expressivity of MöbiusAttention. This research paves the way for exploring the potential of Möbius transformations in the complex projective space to enhance the expressivity and performance of foundation models. At the heart of their success lies the attention mechanism (Vaswani et al., 2017), a powerful tool that enables them to identify relationships between different parts of the data, be it words in a sentence or image patches in a scene. Despite their remarkable impact, current transformers face limitations. A key constraint is the inherent linearity of the attention mechanism, which primarily relies on weights learned through linear transformations, matrix multiplications, and the softmax function. While softmax is a non-linear operation, it is only used to produce a probability distribution over the elements signaling their relative importance in comparison to the others, and not to introduce non-linear interdependencies. Predominantly linear operations restrict the ability of models to capture complex linguistic dependencies, leading to potential information loss within each attention layer as shown by recent research (Zhang, 2023). Figure 1: Various Möbius transformations: Each sub-figure shows flows from a single point after successive transformations. Elliptic Möbius has two fixed points at the centers of two circular flows.

State-of-the-art Large Language Models (LLMs) are accredited with an increasing number of different capabilities, ranging from reading comprehension, over advanced mathematical and reasoning skills to possessing scientific knowledge. In this paper we focus on their multi-hop reasoning capability: the ability to identify and integrate information from multiple textual sources. Given the concerns with the presence of simplifying cues in existing multi-hop reasoning benchmarks, which allow models to circumvent the reasoning requirement, we set out to investigate, whether LLMs are prone to exploiting such simplifying cues. We find evidence that they indeed circumvent the requirement to perform multi-hop reasoning, but they do so in more subtle ways than what was reported about their fine-tuned pre-trained language model (PLM) predecessors. Motivated by this finding, we propose a challenging multi-hop reasoning benchmark, by generating seemingly plausible multi-hop reasoning chains, which ultimately lead to incorrect answers. We evaluate multiple open and proprietary state-of-the-art LLMs, and find that their performance to perform multi-hop reasoning is affected, as indicated by up to 45% relative decrease in F1 score when presented with such seemingly plausible alternatives. We conduct a deeper analysis and find evidence that while LLMs tend to ignore misleading lexical cues, misleading reasoning paths indeed present a significant challenge.

This paper presents insights from evaluating 16 frontier large language models (LLMs) on the WebApp1K benchmark, a test suite designed to assess the ability of LLMs to generate web application code. The results reveal that while all models possess similar underlying knowledge, their performance is differentiated by the frequency of mistakes they make. By analyzing lines of code (LOC) and failure distributions, we find that writing correct code is more complex than generating incorrect code. Furthermore, prompt engineering shows limited efficacy in reducing errors beyond specific cases. These findings suggest that further advancements in coding LLM should emphasize on model reliability and mistake minimization.

Large Language Models (LLMs) have rapidly increased in size and apparent capabilities in the last three years, but their training data is largely English text. There is growing interest in multilingual LLMs, and various efforts are striving for models to accommodate languages of communities outside of the Global North, which include many languages that have been historically underrepresented in digital realms. These languages have been coined as "low resource languages" or "long-tail languages", and LLMs performance on these languages is generally poor. While expanding the use of LLMs to more languages may bring many potential benefits, such as assisting cross-community communication and language preservation, great care must be taken to ensure that data collection on these languages is not extractive and that it does not reproduce exploitative practices of the past. Collecting data from languages spoken by previously colonized people, indigenous people, and non-Western languages raises many complex sociopolitical and ethical questions, e.g., around consent, cultural safety, and data sovereignty. Furthermore, linguistic complexity and cultural nuances are often lost in LLMs. This position paper builds on recent scholarship, and our own work, and outlines several relevant social, cultural, and ethical considerations and potential ways to mitigate them through qualitative research, community partnerships, and participatory design approaches. We provide twelve recommendations for consideration when collecting language data on underrepresented language communities outside of the Global North.