Jev 与文本生成模型的取舍
Jev 不生成自由文本。Diogo 用一段对比视频说明并行决策模型与逐字生成模型在任务形态上的差别。

用户分享使用体验:将 Jev 的 900 个决策与 Opus 5 对比,发现 Jev 已提前将 128 个错误中的 122 个标记为不确定,强调其自知能力。
Joined the same week. What converted me was not the speed, it was checking 900 of its decisions against Opus 5 and finding Jev had already flagged 122 of the 128 misses as unsure. It knows when it does not know. The run is here: