LLMs Exhibit Omissions in Mental-Health Prompts
Researchers led by Congning Ni introduce UTCO, a four-part prompt framework, and use 2,075 generated prompts to stress-test Llama 3.3 for mental-health question answering on April 2, 2026. They report 6.5% hallucinations and 13.2% omissions, concentrated in crisis and suicidal-ideation prompts, and find failures tied to context and tone; they recommend prioritizing omission-based safety evaluations over static benchmarks.
Scoring Rationale
Fresh arXiv preprint with a novel UTCO framework and concrete failure rates boosts novelty and relevance; scope is domain-specific (mental health) and credibility is limited by being a single preprint rather than peer-reviewed, so the score reflects strong novelty and relevance but moderated credibility.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Health & Insurance problemsStep-by-step roadmaps from zero to job-ready — curated courses, salary data, and the exact learning order that gets you hired.
Sources
- Read Original[2604.00014] Disentangling Prompt Element Level Risk Factors for Hallucinations and Omissions in Mental Health LLM Responsesarxiv.org
