Chenwen Lin

PhD student, Fudan University

[Chenwen Lin @cw_lin928]

Suggestions:
1️⃣Start with assistive use — triage, report drafting, risk screening — always human-in-the-loop.
2️⃣Adopt progressively: deploy LVLMs first on perception-heavy tasks (Auto/Health) before complex reasoning.
3️⃣Invest in data + fine-tuning — build domain-specific models.

[Chenwen Lin @cw_lin928]

🎉Excited to present our paper at @ICCVConference, co-authored with @Bruce_Lyu17 , @JieboLuo and Xian Xu: “INS-MMBench: A Comprehensive Benchmark for Evaluating LVLMs' Performance in Insurance” !
Catch us at Poster Session 2 — Exhibit Hall 1, Booth #838, on Oct 21, 3–5 PM🌺

[Chenwen Lin @cw_lin928]

What’s more:
🔹Scenario tasks are hardest—models struggle with multi-step reasoning.
🔹Auto & Health perform better than Property & Agricultural (less reasoning required).
🔹Open-source LVLMs are catching up fast, showing strong potential for privacy-safe, domain-tuned model.