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。

有观察指出 Jev 可能基于 Qwen 构建,但关于其训练方式、是否从头预训练及所用骨干模型,公开信息仍然很少。
Interesting finding: Jev appears to be based on Qwen. Despite being presented as a new “decision model,” there’s very little public information about how Jev was actually trained, whether it was pretrained from scratch, or what backbone it uses. Some signs point toward Qwen