A new research paper explores the 'inertia of confidence' in legal AI models, where systems like ChatGPT, Meta AI, and Perplexity AI provide incorrect legal verdicts with high certainty. The study found Meta AI had the highest High-Confidence Error Rate (HCER) at 31.7% when tested on Indian Contract Act cases, with ChatGPT showing a much lower HCER of 6.7%. The research also surveyed Indian law students, revealing that while most were aware of the risks of submitting hallucinated cases, a significant majority had not received formal training on ethical AI use. AI
IMPACT Highlights risks of overconfidence in legal AI, potentially impacting judicial systems and legal education.
RANK_REASON Academic paper detailing a new phenomenon and evaluation metric for AI models. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- ChatGPT
- Dunning–Kruger effect
- GPT-5.2
- Hugging Face
- Indian Contract Act, 1872
- Jerrin Thomas Panachakel
- Meta AI
- Perplexity AI
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
