In this study, we explored the progression trajectories of artificial intelligence (AI) systems through the lens of complexity theory. We challenged the conventional linear and exponential projections of AI advancement toward Artificial General Intelligence (AGI) underpinned by transformer-based architectures, and posited the existence of critical points, akin to phase transitions in complex systems, where AI performance might plateau or regress into instability upon exceeding a critical complexity threshold. We employed agent-based modelling (ABM) to simulate hypothetical scenarios of AI systems' evolution under specific assumptions, using benchmark performance as a proxy for capability and complexity. Our simulations demonstrated how increasing the complexity of the AI system could exceed an upper criticality threshold, leading to unpredictable performance behaviours. Additionally, we developed a practical methodology for detecting these critical thresholds using simulation data and stochastic gradient descent to fine-tune detection thresholds. This research offers a novel perspective on AI advancement that has a particular relevance to Large Language Models (LLMs), emphasising the need for a tempered approach to extrapolating AI's growth potential and underscoring the importance of developing more robust and comprehensive AI performance benchmarks.

Evaluating the cognitive capacities of large language models (LLMs) requires overcoming not only anthropomorphic but also anthropocentric biases. This article identifies two types of anthropocentric bias that have been neglected: overlooking how auxiliary factors can impede LLM performance despite competence (Type-I), and dismissing LLM mechanistic strategies that differ from those of humans as not genuinely competent (Type-II). Mitigating these biases necessitates an empirically-driven, iterative approach to mapping cognitive tasks to LLM-specific capacities and mechanisms, which can be done by supplementing carefully designed behavioral experiments with mechanistic studies.

Multi-hop question answering (MHQA) requires a model to retrieve and integrate information from multiple passages to answer a complex question. Recent systems leverage the power of large language models and integrate evidence retrieval with reasoning prompts (e.g., chain-of-thought reasoning) for the MHQA task. However, the complexities in the question types (bridge v.s. comparison questions) and the reasoning types (sequential v.s. parallel reasonings) require more novel and fine-grained prompting methods to enhance the performance of MHQA under the zero-shot setting. In this paper, we propose STOC-TOT, a stochastic tree-of-thought reasoning prompting method with constrained decoding for MHQA and conduct a detailed comparison with other reasoning prompts on different question types and reasoning types. Specifically, we construct a tree-like reasoning structure by prompting the model to break down the original question into smaller sub-questions to form different reasoning paths. In addition, we prompt the model to provide a probability estimation for each reasoning path at each reasoning step. At answer time, we conduct constrained decoding on the model to generate more grounded answers and reduce hallucination. Experiments comparing STOC-TOT with two MHQA datasets and five large language models showed that our framework outperforms other reasoning prompts by a significant margin.

In this paper we examine the limitations of Large Language Models (LLMs) for complex reasoning tasks. Although recent works have started to employ formal languages as an intermediate representation for reasoning tasks, they often face challenges in accurately generating and refining these formal specifications to ensure correctness. To address these issues, this paper proposes Logic-LM++, an improvement on Logic-LM . It uses the ability of LLMs to do pairwise comparisons, allowing the evaluation of the refinements suggested by the LLM. The paper demonstrates that Logic-LM++ outperforms Logic-LM and other contemporary techniques across natural language reasoning tasks on three datasets, FOLIO, ProofWriter and AR-LSAT, with an average improvement of 18.5% on standard prompting, 12.3% on chain of thought prompting and 5% on Logic-LM.

Large Language Models (LLMs) have exhibited an impressive ability to perform In-Context Learning (ICL) from only a few examples. Recent works have indicated that the functions learned by ICL can be represented through compressed vectors derived from the transformer. However, the working mechanisms and optimization of these vectors are yet to be thoroughly explored. In this paper, we address this gap by presenting a comprehensive analysis of these compressed vectors, drawing parallels to the parameters trained with gradient descent, and introducing the concept of state vector. Inspired by the works on model soup and momentum-based gradient descent, we propose inner and momentum optimization methods that are applied to refine the state vector progressively as test-time adaptation. Moreover, we simulate state vector aggregation in the multiple example setting, where demonstrations comprising numerous examples are usually too lengthy for regular ICL, and further propose a divide-and-conquer aggregation method to address this challenge. We conduct extensive experiments using Llama-2 and GPT-J in both zero-shot setting and few-shot setting. The experimental results show that our optimization method effectively enhances the state vector and achieves the state-of-the-art performance on diverse tasks.

Evaluating the generalisation capabilities of multimodal models based solely on their performance on out-of-distribution data fails to capture their true robustness. This work introduces a comprehensive evaluation framework that systematically examines the role of instructions and inputs in the generalisation abilities of such models, considering architectural design, input perturbations across language and vision modalities, and increased task complexity. The proposed framework uncovers the resilience of multimodal models to extreme instruction perturbations and their vulnerability to observational changes, raising concerns about overfitting to spurious correlations. By employing this evaluation framework on current Transformer-based multimodal models for robotic manipulation tasks, we uncover limitations and suggest future advancements should focus on architectural and training innovations that better integrate multimodal inputs, enhancing a model's generalisation prowess by prioritising sensitivity to input content over incidental correlations.

This paper explores using GPT-3.5 and GPT-4 to automate the data annotation process with automatic prompting techniques. The main aim of this paper is to reuse human annotation guidelines along with some annotated data to design automatic prompts for LLMs, focusing on the semantic proximity annotation task. Automatic prompts are compared to customized prompts. We further implement the prompting strategies into an open-source text annotation tool, enabling easy online use via the OpenAI API. Our study reveals the crucial role of accurate prompt design and suggests that prompting GPT-4 with human-like instructions is not straightforwardly possible for the semantic proximity task. We show that small modifications to the human guidelines already improve the performance, suggesting possible ways for future research.

Time series models, typically trained on numerical data, are designed to forecast future values. These models often rely on weighted averaging techniques over time intervals. However, real-world time series data is seldom isolated and is frequently influenced by non-numeric factors. For instance, stock price fluctuations are impacted by daily random events in the broader world, with each event exerting a unique influence on price signals. Previously, forecasts in financial markets have been approached in two main ways: either as time-series problems over price sequence or sentiment analysis tasks. The sentiment analysis tasks aim to determine whether news events will have a positive or negative impact on stock prices, often categorizing them into discrete labels. Recognizing the need for a more comprehensive approach to accurately model time series prediction, we propose a collaborative modeling framework that incorporates textual information about relevant events for predictions. Specifically, we leverage the intuition of large language models about future changes to update real number time series predictions. We evaluated the effectiveness of our approach on financial market data.

Natural Language Counterfactual generation aims to minimally modify a given text such that the modified text will be classified into a different class. The generated counterfactuals provide insight into the reasoning behind a model's predictions by highlighting which words significantly influence the outcomes. Additionally, they can be used to detect model fairness issues or augment the training data to enhance the model's robustness. A substantial amount of research has been conducted to generate counterfactuals for various NLP tasks, employing different models and methodologies. With the rapid growth of studies in this field, a systematic review is crucial to guide future researchers and developers. To bridge this gap, this survey comprehensively overview textual counterfactual generation methods, particularly including those based on Large Language Models. We propose a new taxonomy that categorizes the generation methods into four groups and systematically summarize the metrics for evaluating the generation quality. Finally, we discuss ongoing research challenges and outline promising directions for future work.

Large language models (LLMs) have brought significant advancements to code generation and code repair, benefiting both novice and experienced developers. However, their training using unsanitized data from open-source repositories, like GitHub, raises the risk of inadvertently propagating security vulnerabilities. Despite numerous studies investigating the safety of code LLMs, there remains a gap in comprehensively addressing their security features. In this work, we aim to present a comprehensive study aimed at precisely evaluating and enhancing the security aspects of code LLMs. To support our research, we introduce CodeSecEval, a meticulously curated dataset designed to address 44 critical vulnerability types with 180 distinct samples. CodeSecEval serves as the foundation for the automatic evaluation of code models in two crucial tasks: code generation and code repair, with a strong emphasis on security. Our experimental results reveal that current models frequently overlook security issues during both code generation and repair processes, resulting in the creation of vulnerable code. In response, we propose different strategies that leverage vulnerability-aware information and insecure code explanations to mitigate these security vulnerabilities. Furthermore, our findings highlight that certain vulnerability types particularly challenge model performance, influencing their effectiveness in real-world applications. Based on these findings, we believe our study will have a positive impact on the software engineering community, inspiring the development of improved methods for training and utilizing LLMs, thereby leading to safer and more trustworthy model deployment.