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。

介绍 Jev 模型:跳过文本生成,直接以概率形式输出决策,据称比 LLM 调用快 193 倍、便宜 444 倍,且不会产生幻觉。
the most interesting ai launch this month isn't a chatbot. it's a model that can't chat at all. jev (from an ex-openai researcher) skips text entirely and just returns decisions as probabilities. 193x faster, 444x cheaper than an llm call, and it can't hallucinate because you