Jev 进模拟塔台：语音模型对话，Jev 判定放行
开发者把 Jev 放进模拟 JFK 塔台，语音模型负责与飞行员通话，Jev 在后台逐条评估放行指令，置信度接近 0.5 时选择犹豫或转人工。
A developer put Jev inside a simulated JFK control tower. Realtime voice models talk to the pilots. Jev sits behind them and evaluates every clearance while the airport keeps moving. That separation is the interesting part. The voice model doesn’t need to make the safety decision. Jev doesn’t need to speak like a human. Each system handles the job it was actually built for. Voice handles the radio. Jev returns the judgment. Code controls the simulation. When confidence falls near 0.5, the system can hesitate, escalate or hand the decision to a human instead of confidently inventing an answer. This is what real AI automation may look like: not one enormous model pretending to run everything but several specialized forms of intelligence working inside the same loop. The chatbot talks. The decision model decides. The surrounding software remains in control. That architecture is the reason I spent days mapping the most useful JEV systems in the article below.
这条是引用着另一条写的：x.com/xmyttle/status/2103904144289390899