Shimou Ling a, Liang Zhang a, Jiangwei Zhao a, Lili Pan a,, Hongliang Li a a University of Electronic Science and T echnology of China, Chengdu, ChinaAbstract LoRA-based continual learning represents a promising avenue for leveraging pre-trained models in downstream continual learning tasks. Recent studies have shown that orthogonal LoRA tuning e ffectively mitigates forgetting. However, this work unveils that under orthogonal LoRA tuning, the critical parameters for pre-tasks still change notably after learning post-tasks. To address this problem, we directly propose freezing the most critical parameter matrices in the Vision Transformer (ViT) for pre-tasks before learning post-tasks. In addition, building on orthogonal LoRA tuning, we propose orthogonal LoRA composition (LoRAC) based on QR decomposition, which may further enhance the plasticity of our method. Elaborate ablation studies and extensive comparisons demonstrate the e ffectiveness of our proposed method. Our results indicate that our method achieves state-of-the-art (SOT A) performance on several well-known continual learning benchmarks. For instance, on the Split CIFAR-100 dataset, our method shows a 6.35% improvement in accuracy and a 3.24% reduction in forgetting compared to previous methods. Introduction Continual learning (CL) is the process of sequentially training a model on multiple tasks while retaining knowledge acquired from previous tasks [1, 2]. Neural networks often forget knowledge learned from previous tasks after acquiring new knowledge, a phenomenon known as catastrophic forgetting [3]. Significant e fforts have been made to alleviate catastrophic forgetting in neural networks in recent years. These studies can be categorized into three main approaches: architecture-based [4, 5, 6, 7], regularization-based [8, 9, 10, 11], and replay-based [12, 13, 14, 15]. Despite their proven eff ectiveness, these approaches still fall short of practical requirements. Over the past year, visual continual learning combined with pre-trained models (PTMs) has demonstrated significant superiority in alleviating forgetting. Prompt tuning has become the most common method to integrate PTMs with continual learning.

The widespread adoption of AI systems in the economy hinges on their ability to generate economic value that outweighs their inference costs. Evaluating this tradeoff requires metrics that account for both performance and costs. We propose a framework grounded in production theory for evaluating language models by combining accuracy and inference cost. We introduce "cost-of-pass", the expected monetary cost of generating a correct solution. We then define the "frontier cost-of-pass" as the minimum cost-of-pass achievable across available models or the "human-expert, using the approximate cost of hiring an expert. Our analysis reveals distinct economic insights. First, lightweight models are most cost-effective for basic quantitative tasks, large models for knowledge-intensive ones, and reasoning models for complex quantitative problems, despite higher per-token costs. Second, tracking this frontier cost-of-pass over the past year reveals significant progress, particularly for complex quantitative tasks where the cost has roughly halved every few months. Third, to trace key innovations driving this progress, we examine counterfactual frontiers: estimates of cost-efficiency without specific model classes. We find that innovations in lightweight, large, and reasoning models have been essential for pushing the frontier in basic quantitative, knowledge-intensive, and complex quantitative tasks, respectively. Finally, we assess the cost-reductions afforded by common inference-time techniques like majority voting and self-refinement, finding that their marginal accuracy gains rarely justify their costs. Our findings underscore that complementary model-level innovations are the primary drivers of cost-efficiency, and our economic framework provides a principled tool for measuring this progress and guiding deployment.

This paper proposes the "Academy of Athens" multi-agent seven-layer framework, aimed at systematically addressing challenges in multi-agent systems (MAS) within artificial intelligence (AI) art creation, such as collaboration efficiency, role allocation, environmental adaptation, and task parallelism. The framework divides MAS into seven layers: multi-agent collaboration, single-agent multi-role playing, single-agent multi-scene traversal, single-agent multi-capability incarnation, different single agents using the same large model to achieve the same target agent, single-agent using different large models to achieve the same target agent, and multi-agent synthesis of the same target agent. Through experimental validation in art creation, the framework demonstrates its unique advantages in task collaboration, cross-scene adaptation, and model fusion. This paper further discusses current challenges such as collaboration mechanism optimization, model stability, and system security, proposing future exploration through technologies like meta-learning and federated learning. The framework provides a structured methodology for multi-agent collaboration in AI art creation and promotes innovative applications in the art field.

From 2000 to 2015, the UN's Millennium Development Goals guided global priorities. The subsequent Sustainable Development Goals (SDGs) adopted a more dynamic approach, with annual indicator updates. As 2030 nears and progress lags, innovative acceleration strategies are critical. This study develops an AI-powered knowledge graph system to analyze SDG interconnections, discover potential new goals, and visualize them online. Using official SDG texts, Elsevier's keyword dataset, and 1,127 TED Talk transcripts (2020.01-2024.04), a pilot on 269 talks from 2023 applies AI-speculative design, large language models, and retrieval-augmented generation. Key findings include: (1) Heatmap analysis reveals strong associations between Goal 10 and Goal 16, and minimal coverage of Goal 6. (2) In the knowledge graph, simulated dialogue over time reveals new central nodes, showing how richer data supports divergent thinking and goal clarity. (3) Six potential new goals are proposed, centered on equity, resilience, and technology-driven inclusion. This speculative-AI framework offers fresh insights for policymakers and lays groundwork for future multimodal and cross-system SDG applications.

Large Reasoning Models (LRMs) like DeepSeek-R1 and OpenAI-o1 have demonstrated remarkable reasoning capabilities, raising important questions about their biases in LLM-as-a-judge settings. We present a comprehensive benchmark comparing judging biases between LLMs and LRMs across both subjective preference-alignment datasets and objective fact-based datasets. Through investigation of bandwagon, authority, position, and distraction biases, we uncover four key findings: (1) despite their advanced reasoning capabilities, LRMs remain susceptible to the above biases; (2) LRMs demonstrate better robustness than LLMs specifically on fact-related datasets; (3) LRMs exhibit notable position bias, preferring options in later positions; and (4) we identify a novel "superficial reflection bias" where phrases mimicking reasoning (e.g., "wait, let me think...") significantly influence model judgments. To address these biases, we design and evaluate three mitigation strategies: specialized system prompts that reduce judging biases by up to 19\% in preference alignment datasets and 14\% in fact-related datasets, in-context learning that provides up to 27\% improvement on preference tasks but shows inconsistent results on factual tasks, and a self-reflection mechanism that reduces biases by up to 10\% in preference datasets and 16\% in fact-related datasets, with self-reflection proving particularly effective for LRMs. Our work provides crucial insights for developing more reliable LLM-as-a-Judge frameworks, especially as LRMs become increasingly deployed as automated judges.

Purpose: The rapid emergence of large language models (LLMs) such as ChatGPT has significantly impacted foreign language education, yet their pedagogical grammar competence remains under-assessed. This paper introduces CPG-EVAL, the first dedicated benchmark specifically designed to evaluate LLMs' knowledge of pedagogical grammar within the context of foreign language instruction. Methodology: The benchmark comprises five tasks designed to assess grammar recognition, fine-grained grammatical distinction, categorical discrimination, and resistance to linguistic interference. Findings: Smaller-scale models can succeed in single language instance tasks, but struggle with multiple instance tasks and interference from confusing instances. Larger-scale models show better resistance to interference but still have significant room for accuracy improvement. The evaluation indicates the need for better instructional alignment and more rigorous benchmarks, to effectively guide the deployment of LLMs in educational contexts. Value: This study offers the first specialized, theory-driven, multi-tiered benchmark framework for systematically evaluating LLMs' pedagogical grammar competence in Chinese language teaching contexts. CPG-EVAL not only provides empirical insights for educators, policymakers, and model developers to better gauge AI's current abilities in educational settings, but also lays the groundwork for future research on improving model alignment, enhancing educational suitability, and ensuring informed decision-making concerning LLM integration in foreign language instruction.

For large-scale multi-gateway LoRaWAN networks, this study proposes a cloud-edge collaborative resource allocation and decision-making method based on edge intelligence, HEAT-LDL (HEAT-Local Distill Lyapunov), which realizes collaborative decision-making between gateways and terminal nodes. HEAT-LDL combines the Actor-Critic architecture and the Lyapunov optimization method to achieve intelligent downlink control and gateway load balancing. When the signal quality is good, the network server uses the HEAT algorithm to schedule the terminal nodes. To improve the efficiency of autonomous decision-making of terminal nodes, HEAT-LDL performs cloud-edge knowledge distillation on the HEAT teacher model on the terminal node side. When the downlink decision instruction is lost, the terminal node uses the student model and the edge decider based on prior knowledge and local history to make collaborative autonomous decisions. Simulation experiments show that compared with the optimal results of all compared algorithms, HEAT-LDL improves the packet success rate and energy efficiency by 20.5% and 88.1%, respectively.

Although the LoRaW AN network can support a larger node scale than the LoRa private network, as the number of devices increases, the performance of the LoRaW AN network in terms of network congestion and energy consumption faces significant challenges. The limited spectrum resources and channel congestion will lead to a decrease in the communication efficiency of the netwo rk, which in turn affects the reliability of data transmission. How to achieve efficient and energy - saving resource allocation while ensuring network performance remains a key issue. In order to improve the overall performance of the LoRaW AN network, optim izing the transmission strategy parameters such as the spreading factor, transmit power, and receive window of the uplink and downlink is considered to be an effective means. By reasonably configuring these parameters, network conflicts can be effectively reduced, signal attenuation can be reduced, and signal coverage can be increased, thereby improving network reliability and communication quality. However, most of the existing optimization methods focus on the adjustment of the spreading factor and transm it power of the uplink, and rarely consider the impact of the downlink on network performance. To address this problem, this chapter proposes a History - E nhanced t wo - phase Actor - Critic a lgorithm with a s hared Transformer (HEA T), which aims to improve the resource allocation strategy of the LoRaW AN network and improve the overall performance of the network. This chapter conducts multiple sets of comparative experiments between HEA T and various popular methods under different device densities and traffic int ensities to verify the effectiveness of HEA T. 2 System Model and Problem Representation In order to efficiently verify the effectiveness of various LoRaW AN resource allocation strategies, this section describes and models the LoRa link behavior and the LoRaW AN standard in detail. Subsequently, this section proposes the target problem of LoRaW AN resource allocation and expresses the target problem as a Markov decision process.

This study presents a comprehensive evaluation of five leading large language models (LLMs) - Chat GPT 4o, Copilot Pro, Gemini Advanced, Claude Pro, and Meta AI - on their performance in solving calculus differentiation problems. The investigation assessed these models across 13 fundamental problem types, employing a systematic cross-evaluation framework where each model solved problems generated by all models. Results revealed significant performance disparities, with Chat GPT 4o achieving the highest success rate (94.71%), followed by Claude Pro (85.74%), Gemini Advanced (84.42%), Copilot Pro (76.30%), and Meta AI (56.75%). All models excelled at procedural differentiation tasks but showed varying limitations with conceptual understanding and algebraic manipulation. Notably, problems involving increasing/decreasing intervals and optimization word problems proved most challenging across all models. The cross-evaluation matrix revealed that Claude Pro generated the most difficult problems, suggesting distinct capabilities between problem generation and problem-solving. These findings have significant implications for educational applications, highlighting both the potential and limitations of LLMs as calculus learning tools. While they demonstrate impressive procedural capabilities, their conceptual understanding remains limited compared to human mathematical reasoning, emphasizing the continued importance of human instruction for developing deeper mathematical comprehension.

To walk into the lion's den once might be considered foolhardy. To do so again after being mauled by the lion? Six years ago I gave a talk at Ted, the world's leading technology and ideas conference. It led to a gruelling lawsuit and a series of consequences that reverberate through my life to this day. And last week I returned. To give another talk that would incorporate some of my experience: a Ted Talk about being sued for giving a Ted Talk, and how the lessons I'd learned from surviving all that were a model for surviving "broligarchy" – a concept I first wrote about in the Observer in July last year: the alignment of Silicon Valley and autocracy, and a kind of power the world has never seen before.