Hallucination is a Consequence of Space-Optimality: A Rate-Distortion Theorem for Membership Testingcs.AI updates on arXiv.org arXiv:2602.00906v4 Announce Type: replace-cross
Abstract: Large language models often hallucinate with high confidence on “random facts” that lack inferable patterns. We formalize the memorization of such facts as a membership testing problem, unifying the discrete error metrics of Bloom filters with the continuous log-loss of LLMs. By analyzing this problem in the regime where facts are sparse in the universe of plausible claims, we establish a rate-distortion theorem: the optimal memory efficiency is characterized by the minimum KL divergence between score distributions on facts and non-facts. This theoretical framework provides a distinctive explanation for hallucination: even with optimal training, perfect data, and a simplified “closed world” setting, the information-theoretically optimal strategy under limited capacity is not to abstain or forget, but to assign high confidence to some non-facts, resulting in hallucination. We validate this theory empirically on synthetic data, showing that hallucinations persist as a natural consequence of lossy compression.
arXiv:2602.00906v4 Announce Type: replace-cross
Abstract: Large language models often hallucinate with high confidence on “random facts” that lack inferable patterns. We formalize the memorization of such facts as a membership testing problem, unifying the discrete error metrics of Bloom filters with the continuous log-loss of LLMs. By analyzing this problem in the regime where facts are sparse in the universe of plausible claims, we establish a rate-distortion theorem: the optimal memory efficiency is characterized by the minimum KL divergence between score distributions on facts and non-facts. This theoretical framework provides a distinctive explanation for hallucination: even with optimal training, perfect data, and a simplified “closed world” setting, the information-theoretically optimal strategy under limited capacity is not to abstain or forget, but to assign high confidence to some non-facts, resulting in hallucination. We validate this theory empirically on synthetic data, showing that hallucinations persist as a natural consequence of lossy compression. Read More
Beyond touch-based human-machine interface: Control your machines in natural language by utilizing large language models and OPC UAcs.AI updates on arXiv.org arXiv:2510.11300v2 Announce Type: replace-cross
Abstract: This paper proposes an agent-based approach toward a more natural interface between humans and machines. Large language models equipped with tools and the communication standard OPC UA are utilized to control machines in natural language. Instead of touch interaction, which is currently the state-of-the-art medium for interaction in operations, the proposed approach enables operators to talk or text with machines. This allows commands such as ‘Please decrease the temperature by 20 % in machine 1 and start the cleaning operation in machine 2.’ The large language model receives the user input and selects one of three predefined tools that connect to an OPC UA server and either change or read the value of a node. Afterwards, the result of the tool execution is passed back to the language model, which then provides a final response to the user. The approach is universally designed and can therefore be applied to any machine that supports the OPC UA standard. The large language model is neither fine-tuned nor requires training data, only the relevant machine credentials and a parameter dictionary are included within the system prompt. The tool-calling ability and their design is evaluated on a demonstrator setup with a Siemens S7-1500 programmable logic controller with four machine parameters. Fifty synthetically generated commands on five different models were tested and the results demonstrate high success rate, with proprietary GPT-5 models achieving accuracies between 96.0 % and 98.0 %, and open-weight models reaching up to 90.0 %. Afterwards the approach was transferred to a deployed spay-coating machine. The proposed concept is supposed to contribute in advancing natural interaction in industrial human-machine interfaces.
arXiv:2510.11300v2 Announce Type: replace-cross
Abstract: This paper proposes an agent-based approach toward a more natural interface between humans and machines. Large language models equipped with tools and the communication standard OPC UA are utilized to control machines in natural language. Instead of touch interaction, which is currently the state-of-the-art medium for interaction in operations, the proposed approach enables operators to talk or text with machines. This allows commands such as ‘Please decrease the temperature by 20 % in machine 1 and start the cleaning operation in machine 2.’ The large language model receives the user input and selects one of three predefined tools that connect to an OPC UA server and either change or read the value of a node. Afterwards, the result of the tool execution is passed back to the language model, which then provides a final response to the user. The approach is universally designed and can therefore be applied to any machine that supports the OPC UA standard. The large language model is neither fine-tuned nor requires training data, only the relevant machine credentials and a parameter dictionary are included within the system prompt. The tool-calling ability and their design is evaluated on a demonstrator setup with a Siemens S7-1500 programmable logic controller with four machine parameters. Fifty synthetically generated commands on five different models were tested and the results demonstrate high success rate, with proprietary GPT-5 models achieving accuracies between 96.0 % and 98.0 %, and open-weight models reaching up to 90.0 %. Afterwards the approach was transferred to a deployed spay-coating machine. The proposed concept is supposed to contribute in advancing natural interaction in industrial human-machine interfaces. Read More
DeepAgent: A General Reasoning Agent with Scalable Toolsetscs.AI updates on arXiv.org arXiv:2510.21618v3 Announce Type: replace
Abstract: Large reasoning models have demonstrated strong problem-solving abilities, yet real-world tasks often require external tools and long-horizon interactions. Existing agent frameworks typically follow predefined workflows, which limit autonomous and global task completion. In this paper, we introduce DeepAgent, an end-to-end deep reasoning agent that performs autonomous thinking, tool discovery, and action execution within a single, coherent reasoning process. To manage long-horizon interactions, we introduce an autonomous memory folding mechanism that compresses past interactions into structured episodic, working, and tool memories, reducing error accumulation while preserving critical information. To teach general-purpose tool use efficiently and stably, we develop an end-to-end reinforcement learning strategy, namely ToolPO, that leverages LLM-simulated APIs and applies tool-call advantage attribution to assign fine-grained credit to the tool invocation tokens. Extensive experiments on eight benchmarks, including general tool-use tasks (ToolBench, API-Bank, TMDB, Spotify, ToolHop) and downstream applications (ALFWorld, WebShop, GAIA, HLE), demonstrate that DeepAgent consistently outperforms baselines across both labeled-tool and open-set tool retrieval scenarios. The code and demo are available at https://github.com/RUC-NLPIR/DeepAgent.
arXiv:2510.21618v3 Announce Type: replace
Abstract: Large reasoning models have demonstrated strong problem-solving abilities, yet real-world tasks often require external tools and long-horizon interactions. Existing agent frameworks typically follow predefined workflows, which limit autonomous and global task completion. In this paper, we introduce DeepAgent, an end-to-end deep reasoning agent that performs autonomous thinking, tool discovery, and action execution within a single, coherent reasoning process. To manage long-horizon interactions, we introduce an autonomous memory folding mechanism that compresses past interactions into structured episodic, working, and tool memories, reducing error accumulation while preserving critical information. To teach general-purpose tool use efficiently and stably, we develop an end-to-end reinforcement learning strategy, namely ToolPO, that leverages LLM-simulated APIs and applies tool-call advantage attribution to assign fine-grained credit to the tool invocation tokens. Extensive experiments on eight benchmarks, including general tool-use tasks (ToolBench, API-Bank, TMDB, Spotify, ToolHop) and downstream applications (ALFWorld, WebShop, GAIA, HLE), demonstrate that DeepAgent consistently outperforms baselines across both labeled-tool and open-set tool retrieval scenarios. The code and demo are available at https://github.com/RUC-NLPIR/DeepAgent. Read More
Towards Green AI: Decoding the Energy of LLM Inference in Software Developmentcs.AI updates on arXiv.org arXiv:2602.05712v1 Announce Type: cross
Abstract: Context: AI-assisted tools are increasingly integrated into software development workflows, but their reliance on large language models (LLMs) introduces substantial computational and energy costs. Understanding and reducing the energy footprint of LLM inference is therefore essential for sustainable software development. Objective: In this study, we conduct a phase-level analysis of LLM inference energy consumption, distinguishing between the (1) prefill, where the model processes the input and builds internal representations, and (2) decoding, where output tokens are generated using the stored state. Method: We investigate six 6B-7B and four 3B-4B transformer-based models, evaluating them on code-centric benchmarks HumanEval for code generation and LongBench for code understanding. Results: Our findings show that, within both parameter groups, models exhibit distinct energy patterns across phases. Furthermore, we observed that increases in prefill cost amplify the energy cost per token during decoding, with amplifications ranging from 1.3% to 51.8% depending on the model. Lastly, three out of ten models demonstrate babbling behavior, adding excessive content to the output that unnecessarily inflates energy consumption. We implemented babbling suppression for code generation, achieving energy savings ranging from 44% to 89% without affecting generation accuracy. Conclusion: These findings show that prefill costs influence decoding, which dominates energy consumption, and that babbling suppression can yield up to 89% energy savings. Reducing inference energy therefore requires both mitigating babbling behavior and limiting impact of prefill on decoding.
arXiv:2602.05712v1 Announce Type: cross
Abstract: Context: AI-assisted tools are increasingly integrated into software development workflows, but their reliance on large language models (LLMs) introduces substantial computational and energy costs. Understanding and reducing the energy footprint of LLM inference is therefore essential for sustainable software development. Objective: In this study, we conduct a phase-level analysis of LLM inference energy consumption, distinguishing between the (1) prefill, where the model processes the input and builds internal representations, and (2) decoding, where output tokens are generated using the stored state. Method: We investigate six 6B-7B and four 3B-4B transformer-based models, evaluating them on code-centric benchmarks HumanEval for code generation and LongBench for code understanding. Results: Our findings show that, within both parameter groups, models exhibit distinct energy patterns across phases. Furthermore, we observed that increases in prefill cost amplify the energy cost per token during decoding, with amplifications ranging from 1.3% to 51.8% depending on the model. Lastly, three out of ten models demonstrate babbling behavior, adding excessive content to the output that unnecessarily inflates energy consumption. We implemented babbling suppression for code generation, achieving energy savings ranging from 44% to 89% without affecting generation accuracy. Conclusion: These findings show that prefill costs influence decoding, which dominates energy consumption, and that babbling suppression can yield up to 89% energy savings. Reducing inference energy therefore requires both mitigating babbling behavior and limiting impact of prefill on decoding. Read More
Google AI Introduces PaperBanana: An Agentic Framework that Automates Publication Ready Methodology Diagrams and Statistical PlotsMarkTechPost Generating publication-ready illustrations is a labor-intensive bottleneck in the research workflow. While AI scientists can now handle literature reviews and code, they struggle to visually communicate complex discoveries. A research team from Google and Peking University introduce new framework called ‘PaperBanana‘ which is changing that by using a multi-agent system to automate high-quality academic diagrams
The post Google AI Introduces PaperBanana: An Agentic Framework that Automates Publication Ready Methodology Diagrams and Statistical Plots appeared first on MarkTechPost.
Generating publication-ready illustrations is a labor-intensive bottleneck in the research workflow. While AI scientists can now handle literature reviews and code, they struggle to visually communicate complex discoveries. A research team from Google and Peking University introduce new framework called ‘PaperBanana‘ which is changing that by using a multi-agent system to automate high-quality academic diagrams
The post Google AI Introduces PaperBanana: An Agentic Framework that Automates Publication Ready Methodology Diagrams and Statistical Plots appeared first on MarkTechPost. Read More
What I Am Doing to Stay Relevant as a Senior Analytics Consultant in 2026Towards Data Science Learn how to work with AI, while strengthening your unique human skills that technology cannot replace
The post What I Am Doing to Stay Relevant as a Senior Analytics Consultant in 2026 appeared first on Towards Data Science.
Learn how to work with AI, while strengthening your unique human skills that technology cannot replace
The post What I Am Doing to Stay Relevant as a Senior Analytics Consultant in 2026 appeared first on Towards Data Science. Read More
A major U.S. payment gateway and solutions provider says a ransomware attack has knocked key systems offline, triggering a widespread outage affecting multiple services. The incident began on Friday and quickly escalated into a nationwide disruption across BridgePay’s platform. […] Read More
Automated Customization of LLMs for Enterprise Code Repositories Using Semantic Scopescs.AI updates on arXiv.org arXiv:2602.05780v1 Announce Type: cross
Abstract: Code completion (CC) is a task frequently used by developers when working in collaboration with LLM-based programming assistants. Despite the increased performance of LLMs on public benchmarks, out of the box LLMs still have a hard time generating code that aligns with a private code repository not previously seen by the model’s training data. Customizing code LLMs to a private repository provides a way to improve the model performance. In this paper we present our approach for automated LLM customization based on semantic scopes in the code. We evaluate LLMs on real industry cases with two private enterprise code repositories with two customization strategies: Retrieval-Augmented Generation (RAG) and supervised Fine-Tuning (FT). Our mechanism for ingesting the repository’s data and formulating the training data pairs with semantic scopes helps models to learn the underlying patterns specific to the repository, providing more precise code to developers and helping to boost their productivity. The code completions of moderately sized customized models can be significantly better than those of uncustomized models of much larger capacity. We also include an analysis of customization on two public benchmarks and present opportunities for future work.
arXiv:2602.05780v1 Announce Type: cross
Abstract: Code completion (CC) is a task frequently used by developers when working in collaboration with LLM-based programming assistants. Despite the increased performance of LLMs on public benchmarks, out of the box LLMs still have a hard time generating code that aligns with a private code repository not previously seen by the model’s training data. Customizing code LLMs to a private repository provides a way to improve the model performance. In this paper we present our approach for automated LLM customization based on semantic scopes in the code. We evaluate LLMs on real industry cases with two private enterprise code repositories with two customization strategies: Retrieval-Augmented Generation (RAG) and supervised Fine-Tuning (FT). Our mechanism for ingesting the repository’s data and formulating the training data pairs with semantic scopes helps models to learn the underlying patterns specific to the repository, providing more precise code to developers and helping to boost their productivity. The code completions of moderately sized customized models can be significantly better than those of uncustomized models of much larger capacity. We also include an analysis of customization on two public benchmarks and present opportunities for future work. Read More
Collaborating with AI Agents: Field Experiments on Teamwork, Productivity, and Performancecs.AI updates on arXiv.org arXiv:2503.18238v3 Announce Type: replace-cross
Abstract: We examined the mechanisms underlying productivity and performance gains from AI agents using a large-scale experiment on Pairit, a platform we developed to study human-AI collaboration. We randomly assigned 2,234 participants to human-human and human-AI teams that produced 11,024 ads for a think tank. We evaluated the ads using independent human ratings and a field experiment on X which garnered ~5M impressions. We found human-AI teams produced 50% more ads per worker and higher text quality, while human-human teams produced higher image quality, suggesting a jagged frontier of AI agent capability. Human-AI teams also produced more homogeneous, or self-similar, outputs. The field experiment revealed higher text quality improved click-through rates and view-through duration, while higher image quality improved cost-per-click rates. We found three mechanisms explained these effects. First, human-AI collaboration was more task-oriented, with 25% more task-oriented messages and 18% fewer interpersonal messages. Second, human-AI collaboration displayed more delegation, as participants delegated 17% more work to AI agents than to human partners and performed 62% fewer direct text edits when working with AI. Third, recognition that the collaborator was an AI moderated these effects as participants who correctly identified they were working with AI were more task-oriented and more likely to delegate work. These mechanisms then explained performance as task-oriented communication improved ad quality, specifically when working with AI, while interpersonal communication reduced ad quality; delegation improved text quality but had no effect on image quality and was positively associated with diversity collapse, creating homogeneous outputs of higher average quality. The results suggest AI agents drive changes in productivity, performance, and output diversity by reshaping teamwork.
arXiv:2503.18238v3 Announce Type: replace-cross
Abstract: We examined the mechanisms underlying productivity and performance gains from AI agents using a large-scale experiment on Pairit, a platform we developed to study human-AI collaboration. We randomly assigned 2,234 participants to human-human and human-AI teams that produced 11,024 ads for a think tank. We evaluated the ads using independent human ratings and a field experiment on X which garnered ~5M impressions. We found human-AI teams produced 50% more ads per worker and higher text quality, while human-human teams produced higher image quality, suggesting a jagged frontier of AI agent capability. Human-AI teams also produced more homogeneous, or self-similar, outputs. The field experiment revealed higher text quality improved click-through rates and view-through duration, while higher image quality improved cost-per-click rates. We found three mechanisms explained these effects. First, human-AI collaboration was more task-oriented, with 25% more task-oriented messages and 18% fewer interpersonal messages. Second, human-AI collaboration displayed more delegation, as participants delegated 17% more work to AI agents than to human partners and performed 62% fewer direct text edits when working with AI. Third, recognition that the collaborator was an AI moderated these effects as participants who correctly identified they were working with AI were more task-oriented and more likely to delegate work. These mechanisms then explained performance as task-oriented communication improved ad quality, specifically when working with AI, while interpersonal communication reduced ad quality; delegation improved text quality but had no effect on image quality and was positively associated with diversity collapse, creating homogeneous outputs of higher average quality. The results suggest AI agents drive changes in productivity, performance, and output diversity by reshaping teamwork. Read More
Enhancing Personality Recognition by Comparing the Predictive Power of Traits, Facets, and Nuancescs.AI updates on arXiv.org arXiv:2602.05650v1 Announce Type: cross
Abstract: Personality is a complex, hierarchical construct typically assessed through item-level questionnaires aggregated into broad trait scores. Personality recognition models aim to infer personality traits from different sources of behavioral data. However, reliance on broad trait scores as ground truth, combined with limited training data, poses challenges for generalization, as similar trait scores can manifest through diverse, context dependent behaviors. In this work, we explore the predictive impact of the more granular hierarchical levels of the Big-Five Personality Model, facets and nuances, to enhance personality recognition from audiovisual interaction data. Using the UDIVA v0.5 dataset, we trained a transformer-based model including cross-modal (audiovisual) and cross-subject (dyad-aware) attention mechanisms. Results show that nuance-level models consistently outperform facet and trait-level models, reducing mean squared error by up to 74% across interaction scenarios.
arXiv:2602.05650v1 Announce Type: cross
Abstract: Personality is a complex, hierarchical construct typically assessed through item-level questionnaires aggregated into broad trait scores. Personality recognition models aim to infer personality traits from different sources of behavioral data. However, reliance on broad trait scores as ground truth, combined with limited training data, poses challenges for generalization, as similar trait scores can manifest through diverse, context dependent behaviors. In this work, we explore the predictive impact of the more granular hierarchical levels of the Big-Five Personality Model, facets and nuances, to enhance personality recognition from audiovisual interaction data. Using the UDIVA v0.5 dataset, we trained a transformer-based model including cross-modal (audiovisual) and cross-subject (dyad-aware) attention mechanisms. Results show that nuance-level models consistently outperform facet and trait-level models, reducing mean squared error by up to 74% across interaction scenarios. Read More
