arXiv:2502.15543v4 Announce Type: replace Abstract: Large language models (LLMs) integrated with retrieval-augmented generation (RAG) have improved factuality by grounding outputs in external evidence. However, they remain susceptible to unfaithful generation, where outputs contradict retrieved context despite its relevance and accuracy. Existing approaches aiming to improve faithfulness primarily focus on enhancing the utilization of external context, but often overlook the persistent influence of internal parametric knowledge during generation. In this work, we investigate the internal mecha
The increasing deployment of LLMs in critical applications is highlighting their inherent 'unfaithful generation' problem, driving a need for immediate solutions.
This research addresses a core limitation in LLM reliability, a prerequisite for broader, more trustworthy AI applications, particularly in agentic systems.
The ability to suppress internal parametric knowledge within RAG systems could significantly reduce instances of hallucination and improve factual consistency in LLM outputs.
- · AI developers
- · Enterprises deploying LLMs
- · Users of RAG systems
- · AI safety researchers
- · LLMs with high hallucination rates
- · Solutions solely focused on external context enhancement
LLMs integrated with RAG will become more reliable and trustworthy for factual tasks.
This improved reliability could accelerate the adoption of AI agents in domains requiring high factual accuracy.
Increased trust in AI systems may lead to their integration into even more sensitive and decision-making processes, blurring human-AI interfaces further.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL