255 stories tagged LLMs · page 6 of 11
A new study explores whether fundamental reasoning patterns in LLMs can be decoupled from specific problem instances. The research highlights the challenges and implications for model controllability and reasoning capabilities.
Researchers propose a Bayesian statistical approach to confidently migrate LLMs in production. The method calibrates automated metrics with human judgments, demonstrated on a system handling 5.3M monthly interactions.
Researchers found that geometric relations between semantic features in LLMs' hidden states closely match human psychological associations. The study projects 360 words onto 32 semantic axes, showing high correlation with human ratings.
Gemma 4 31B demonstrated superior efficiency and speed in a game development contest, completing the task in a fraction of the time compared to Qwen 3.6 27B. The results highlight significant performance differences between the two models.
A new study analyzes how large language models (LLMs) transition between user prompts and responses in terms of safety. The research found that 61% of responses de-escalate harm, highlighting the dynamic nature of risk in AI interactions.
A new study explores how Large Language Models (LLMs) respond to legal arguments and the factors influencing their decisions. The research highlights the importance of understanding LLMs' persuadability in judicial and administrative contexts.
Researchers propose a novel approach to estimate the parameter count of black-box language models by analyzing their factual capacity. This method could revolutionize benchmarking and comparison of proprietary models.
Researchers have developed a new jailbreak technique called Incremental Completion Decomposition (ICD) that exploits LLMs by eliciting single-word continuations before extracting harmful responses. This method bypasses current safety mechanisms, raising concerns about the robustness of LLM safeguards.
A new study evaluated 72 large language models on safety for robotic health attendants, finding a 54.4% average violation rate. The results highlight significant risks in deploying LLMs for medical applications without rigorous safety protocols.
Researchers introduce BioGraphletQA, a new biomedical QA dataset with 119,856 pairs. The framework uses Knowledge Graph subgraphs to ensure factual grounding and control complexity.
Researchers discovered that providing one confirmed fact in a multi-step reasoning chain increases LLMs' likelihood of confidently producing wrong answers. This phenomenon, called anchored confabulation, challenges assumptions about how models handle partial evidence.
Researchers introduce AGEL-Comp, a neuro-symbolic AI architecture that enhances compositional generalization in interactive agents. The framework combines a dynamic Causal Program Graph and an Inductive Logic Programming engine to improve robustness in complex environments.
A new study analyzes intermediate reasoning steps of LLMs to reveal stigmatizing language and biases toward mental health conditions. The findings highlight limitations of traditional evaluation methods.
A GitHub project demonstrates how large language models can assist in reconstructing partially decompiled code. The project focuses on Minecraft 26.1.2, showcasing AI's potential in reverse engineering.
A new study identifies flaws in how LLMs handle clinical trial data, proposing a hybrid approach to improve reasoning. The research focuses on recovering implicit attributes from partially observed tables.
Researchers propose a roundtrip verification approach to ensure LLMs produce faithful formalizations of natural language. The method involves translating and re-formalizing statements to check for logical equivalence.
Researchers propose a structured approach to debugging LLMs, treating them as observable systems. The method offers model-agnostic techniques for issue detection and refinement, addressing the complexity of LLM errors.
Researchers propose a new AI-driven framework for aircraft fault diagnosis using digital twins and LLMs. The method addresses data scarcity and diverse fault types in general aviation. The system integrates high-fidelity simulations, fault injection, and interpretable report generation to improve diagnostic accuracy.
IBM has released Granite 4.1, a suite of open-source large language models trained using advanced techniques like mixture-of-experts and fine-tuning. The models are optimized for both efficiency and performance, setting a new standard in open-source AI.
Researchers introduce FormalScience, an AI pipeline that helps domain experts formalize scientific reasoning into verifiable code. This addresses a key challenge in applying LLMs to scientific domains with complex notations.
A new study shows that belief graphs significantly improve weaker LLMs in cooperative reasoning tasks, while stronger models see minimal benefits. The research highlights the importance of integration architecture in leveraging these graphs effectively.
Researchers found that the variability in LLM responses to different prompt styles stems from shared underlying task representations. This explains why the same question can yield different answers depending on phrasing.
A new study formalizes the concept of 'background temperature' to describe implementation-level nondeterminism in large language models. This helps explain why identical inputs can produce divergent outputs even at temperature T=0.
A study reveals that LLMs trained on Western data fail to detect health misinformation in non-Western contexts, such as cow urine remedies on Indian YouTube. This highlights a critical gap in AI's ability to handle culturally nuanced content.