20 stories tagged Hallucinations
A new arXiv study finds that small changes to how questions are phrased—called prompt perturbation—can reduce bias and hallucinations in large language models, challenging prior assumptions that perturbations worsen these issues.
Researchers have developed a method to detect when large language models (LLMs) generate false information by analyzing the topological structure of their attention graphs. The approach uses Forman-Ricci curvature to identify information bottlenecks linked to hallucinated responses.
Developer Damien Charlotin created a public database documenting AI hallucination incidents, aiming to improve transparency and accountability in AI systems by recording errors, context, and patterns.
A new arXiv study introduces Expert-Aware Contrast Decoding (EACD) for mixture-of-experts (MoE) models, showing it reduces AI hallucinations by leveraging expert-specific knowledge. The method outperforms standard contrastive decoding on transformer models, offering a path to more reliable AI assistants.
Researchers have developed a way to catch AI chatbots when they make up false information. This is especially important for systems that need to provide accurate details, like recommending restaurants or booking appointments.
Researchers developed CORTEX, a system that spots fake or made-up information in AI chatbot answers. It works by checking each word against the original sources. This could make AI responses more reliable for everyday users.
Researchers compare two approaches to world models for AI agents—agent-based and parameterized—showing that parameterized models reduce hallucination propagation by using measurable errors, leading to more reliable planning in step-by-step tasks.
A new arXiv paper proposes methods to detect when large language models hallucinate during knowledge graph reasoning. This could make AI-driven question answering, recommendations, and decision support more reliable.
KPMG has retracted a report on AI adoption after its AI model generated false statistics. This highlights how even major firms struggle with AI reliability.
Researchers developed TIGER, a new AI system that improves accuracy in multimodal generation by tracking and correcting factual errors. This could make AI-generated images and text more reliable for everyday users.
Scientists have discovered that AI hallucinations are easier to detect in intermediate layers of AI models, not just the final output. They've developed a method to automatically find the best layers for spotting these errors, making AI more reliable.
Researchers found that AI models can make worse predictions when given accurate context. This happens because the models sometimes ignore good information. The study highlights a hidden flaw in how AI systems process data.
Researchers have developed a framework to better detect and prevent AI-generated medical misinformation. This could make medical AI tools more reliable for everyday users.
OpenAI says its latest ChatGPT model, GPT-5.5 Instant, makes up facts 52.5% less often. This could make AI assistants more reliable for everyday use.
OpenAI has launched GPT-5.5 Instant, a new default model for ChatGPT that reduces hallucinations in sensitive areas like law, medicine, and finance. The model maintains the speed of its predecessor while improving accuracy.
OpenAI has updated ChatGPT’s default model to GPT-5.5 Instant, which offers more accurate, less confusing answers and better personalization. This makes it easier to get the information you need without unnecessary guesswork.
Researchers propose HalluSAE, a novel method to detect hallucinations in LLMs by modeling them as critical shifts in latent dynamics. This approach addresses the dynamic nature of hallucinations often overlooked by existing methods.
Fine-tuning large language models can increase hallucinations by degrading pre-trained knowledge. Researchers propose a self-distillation method to mitigate this issue. This could reshape how we approach model training.
Researchers propose a method to improve factuality in long-form LLM outputs by calibrating reasoning confidence. This approach reduces overconfident incorrect claims without post-hoc revision.
A new analysis highlights the critical need for fault tolerance in AI systems used by political campaigns. The piece emphasizes strategies to mitigate hallucinations and ensure reliable AI deployment.