用 Jev 分类模型逐步选择合法 token 编写 C++
作者用 TypeSafe 的 Jev 决策/分类模型做实验:每步只提供语法合法的候选 token,让模型选择,从而生成 C++ 代码。
dotpals集成Jev,在AI代理声称所有测试通过时,对实际输出进行验证,并以92%的置信度标记失败,为每次测试运行增加无法伪造的独立检查。
My AI agent said: "All tests pass. Done!" ✅I believed it. I never really knew what was going on behind the scenes. Then dotpals + Jev checked the actual output: "Failed. 92% sure." 🫠New in dotpals: a second opinion on every test run your agent can't prove. Jev in the cloud,