The Truth Stays in the Family: Enhancing Contextual Grounding via Inherited Truthful Heads in Model Lineages
Recent advances in large language models (LLMs) have produced many specialized multimodal LLMs (MLLMs) that share com...
Source Evidence
What Changed
Recent advances in large language models (LLMs) have produced many specialized multimodal LLMs (MLLMs) that share com...
Why It Matters
The study demonstrates that a model’s foundational “truthfulness” is inherited through fine‑tuning and multimodal adaptation, meaning that lineage‑aware tuning can reliably transfer veracity guarantees to downstream products. By soft‑gating those inherited heads, developers can boost factual consistency without sacrificing general performance—critical for trustworthy deployment of multimodal LLMs in regulated or high‑stakes domains.
Confirmed Facts
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.
Who Is Affected
- Mistral
- Meta AI
- Hugging Face
- Qwen
- AI product teams
What To Watch Next
- Watch for benchmark validation, API availability, pricing, limits, and early customer adoption.
- Watch whether additional sources confirm the same claim.
You will be redirected to arxiv.org.