89 stories tagged Language Models
A new arXiv study reveals that large language models produce homogenized opinions in tasks like synthetic surveys and public opinion prediction. Researchers identify which interventions actually increase diversity and which don't, offering a clearer path to more realistic AI-generated viewpoints.
Researchers at Moir introduced a knowledge-editing method that lets language models self-direct their own updates, preserving mathematical and programmatic reasoning while adding new encyclopedic facts. This addresses a key bottleneck in deploying editable AI assistants.
Researchers extracted DeepSeek's hidden criteria for judging literary quality by analyzing its reasoning traces. The model achieved 79.3% accuracy classifying texts from canonical literature to anonymous forum posts, revealing a consistent implicit theory of quality centered on coherence, originality, and emotional resonance.
A new study from ArXiv reveals that large language models can bypass safety filters when given incomplete harmful prompts, a vulnerability the researchers call 'incomplete prompt jailbreaks' (IPJ). The findings show that models systematically delay refusal until the sentence ends, allowing harmful continuations to slip through.
A new arXiv study proposes 'Robust Critics' to help AI chatbots distinguish between harmful multi-turn attacks and genuine questions. Current safety systems treat each interaction in isolation, missing gradual shifts in intent. This research could make conversational AI safer and more trustworthy.
Researchers from ArXiv cs.AI introduce Probabilistic Concept-Aware Steering, a new inference-time technique that uses concept-specific direction vectors to guide large language models (LLMs) toward more coherent, on-topic, and trustworthy responses — addressing a key limitation of existing steering vector methods.
A new study encoding EEG signals shows that advanced language models achieve next-word prediction accuracy closely aligned with human brain activity during reading. The findings bridge neuroscience and AI, with implications for more intuitive human-AI collaboration.
OpenLanguageModel (OLM) is an open-source PyTorch library that makes small language model pretraining readable and composable. Its code mirrors the architecture using components like Block, Residual, Repeat, and Parallel, enabling use in teaching notebooks, full pretraining runs, and research ablations without modification.
Researchers introduce Introspection Fine-Tuning (IFT), a method that trains small language models to detect and report perturbations in their own internal activations. This breakthrough could make AI systems more reliable and transparent by enabling self-monitoring in smaller, efficient models.
A new ArXiv paper proves that large language models (LLMs) have an inherent reliability ceiling that no amount of scaling can overcome. The study decomposes output uncertainty into a resolvable component (closable with more context) and a subjective component (inherent to task ambiguity), and shows that autoregressive generation further degrades this ceiling.
Researchers from ArXiv cs.CL have developed a system that automatically evolves prompt guidelines tailored to specific tasks, addressing the common problem of underspecified user queries. This could make large language models like ChatGPT and Claude more accurate and user-friendly without requiring prompt engineering expertise.
Researchers from ArXiv cs.CL discovered how language models distinguish between a character's belief and objective reality. The mechanism relies on a shared value slot that binds the attributed value and a router at the query position that selects which frame—belief or reality—to read from. This finding explains how AI handles nuanced, belief-based reasoning.
Bilibili has released a series of open small language models called Index-1.9B, trained on a massive dataset of Chinese and English tokens. These models are designed for various applications, from general use to specialized tasks like chat and character interaction.
Researchers introduced HALO, a method to enhance existing AI language models with minimal extra computation. It selectively refines responses, improving accuracy without wasting resources.
A new study reveals that AI language models often change African American English into Standard American English without users knowing. Researchers have developed a way to detect and reduce this bias in AI systems.
Researchers have introduced Wiola, a novel small language model architecture that breaks from existing designs. It incorporates five unique components aimed at improving efficiency and performance.
Researchers developed a method called CreativityNeuro to make AI models generate more diverse and creative responses. This could help AI avoid repetitive answers and offer more unique ideas.
Researchers created BayesBench to test how large language models update their beliefs with new evidence. The study reveals that AI models often struggle to adjust their reasoning as conversations progress.
Researchers created a dataset of 240 French doctor-patient conversations to train AI virtual patients. This could make medical training more accessible and effective for students worldwide.
A new paper highlights flaws in how we test AI models that handle text, images, and other inputs together. Current methods miss key aspects like understanding physical reality or combining different types of information. The authors propose better evaluation frameworks to address these gaps.
A new AI model called iLLaDA uses a different approach to understand language, potentially making it more efficient and accurate. This could lead to better AI assistants and tools that understand context more naturally.
Scientists developed a new method using Shapley values to measure how adjectives influence AI responses. Their findings reveal how small word choices can significantly impact AI performance and alignment.
A new paper finds that offline recommendations to mix AI models from different "families" for diversity may not hold up in real-time interactive settings—the very environments where multi-LLM systems are actually deployed. This discovery could change how multi-AI systems are designed.
Scientists discovered a gap in AI systems that combine logic solvers with language models. This gap can lead to incorrect answers despite using verified logic tools. Researchers propose solutions to maintain accuracy in these hybrid systems.