← 返回发现
AI工具ai

jspace-hallucination-eval

dasjoms/jspace-hallucination-eval

Multi-dataset evaluation of Anthropic's J-Space on Qwen3-4B. Demonstrates that internal 'workspace noise' can reliably route high-confidence confabulations, while mapping the structural limitations of the metric across adversarial, reasoning, and multiple-choice tasks.

12
Stars
15
热度评分
+0
7日增长
4天
趋势

项目信息

分类AI工具
用途ai
发现日期2026-07-15