analiza aiqbench
This research demonstrates that truthfulness mechanisms in foundational large language models are robustly inherited by downstream variants, even after instruction tuning or multimodal adaptation. By identifying and amplifying specific attention heads responsible for factual grounding, the proposed TruthProbe method effectively reduces hallucinations across diverse model families without requiring retraining from scratch. This finding offers a practical pathway to enhance reliability in specialized AI systems while leveraging existing base-model capabilities.
The Truth Stays in the Family: Enhancing Contextual Grounding via Inherited Truthful Heads in Model Lineages
arXiv:2606.15821v1 Announce Type: new Abstract: Recent advances in large language models (LLMs) have produced many specialized multimodal LLMs (MLLMs) that share common foundational LLMs, forming distinct model lineages. It remains unclear whether a fundamental behavioral link exists between the foundational LLMs and downstream variants. We investigate this question by quantifying head-level context-truthfulness scores. Across diverse LLM and MLLM lineages, including Vicuna-, Qwen2.5-, LLaMA2-, and Mistral-based models, we find that Truth Scores are strongly preserved within model families, even after instruction tuning or multimodal adaptation. We further show that this inheritance is consistent with attention-head weight preservation, and that context-truthful heads attend to query-relevant evidence. Building on this finding, we propose TruthProbe, a soft-gating strategy that amplifies context-truthful heads while preserving other head contributions. TruthProbe improves contextual truthfulness on HaluEval and reduces multimodal hallucination on POPE and CHAIR, with base-LLM Truth Scores transferring effectively to their fine-tuned LLM and MLLM descendants. Code is available at https://github.com/miso-choi/TruthProbe.
Abstract:Recent advances in large language models (LLMs) have produced many specialized multimodal LLMs (MLLMs) that share common foundational LLMs, forming distinct model lineages. It remains unclear whether a fundamental behavioral link exists between the foundational LLMs and downstream variants. We investigate this question by quantifying head-level context-truthfulness scores. Across diverse LLM and MLLM lineages, including Vicuna-, Qwen2.5-, LLaMA2-, and Mistral-based models, we find that Truth Scores are strongly preserved within model families, even after instruction tuning or multimodal adaptation. We further show that this inheritance is consistent with attention-head weight preservation, and that context-truthful heads attend to query-relevant evidence. Building on this finding, we propose TruthProbe, a soft-gating strategy that amplifies context-truthful heads while preserving other head contributions. TruthProbe improves contextual truthfulness on HaluEval and reduces multimodal hallucination on POPE and CHAIR, with base-LLM Truth Scores transferring effectively to their fine-tuned LLM and MLLM descendants. Code is available at this https URL.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2606.15821 [cs.CL]
(or arXiv:2606.15821v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.15821 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 253 cuvinte
Subiecte conexe
Commonsense Reasoning in Computer Vision: Foundations, Recent Advancements, and Future Directions
arXiv · 7 sept. 2026
Anatomy Contextualized Adaption of CT Foundation Models
arXiv · 31 iul. 2026
Multi-Perspective Evidence Synthesis and Reasoning for Unsupervised Multimodal Entity Linking
arXiv · 21 oct. 2026
SFAD: Speculative Factuality-Aware Decoding
arXiv · 2 sept. 2026