Search papers, labs, and topics across Lattice.
This paper introduces LLUMI, a two-component (generation and improvement models) system for mental health support that leverages feedback from Reddit communities via upvotes/downvotes to train open-source LLMs using SFT and DPO. LLUMI is aligned using human evaluation across readability, empathy, connection, actionability, and safety dimensions. Results demonstrate that LLUMI achieves comparable performance to proprietary models in generating supportive responses, offering a privacy-preserving alternative for sensitive support contexts.
Open-source LLMs, trained with Reddit community feedback, can match the performance of proprietary models in generating supportive mental health responses, offering a privacy-preserving alternative.
Large language models (LLMs) show promise in generating supportive responses for mental health queries, but improving their usefulness, empathy, and safety often requires substantial compute, expert input, and labeled data. At the same time, deploying proprietary, cloud-based models for mental health-related interactions raises important privacy and data-governance concerns, given the sensitivities. To address this challenge, we introduce LLUMI setup that can be hosted in-house within protected environments. LLUMI consists of two complementary components: a generation model (GM), which drafts supportive responses to mental health queries, and an improvement model (IM), which revises an initial human-crafted response. We leverage feedback signals from Reddit mental health communities, using community endorsement patterns such as upvotes and downvotes to construct chosen-rejected response pairs for Supervised Fine Tuning (SFT) and Direct Preference Optimization (DPO). We further align LLUMI using human evaluation across five dimensions: readability, empathy, connection, actionability, and safety. Our results show that, despite relying on smaller open-source models rather than proprietary cloud-based GPT models, LLUMI achieves comparable performance across linguistic analyses and human evaluations. These findings suggest that open-source models, when trained with community-derived preference signals, can support high-quality mental health support assistance while offering a more privacy-preserving alternative for sensitive support contexts.
