如果你用过Advanced Voice Mode，GPT-Live流畅了一些，但依然不能当真正的助手用，有硬伤。
用户对OpenAI的GPT-Live语音模式给出初步印象：对话更自然、更具表现力，但相比Advanced Voice Mode（AVM）仍有限制。它无法访问GitHub、Linear或邮件等应用，没有唤醒或挂断语音触发，通话每60分钟超时一次。在相同推理水平下，它无法正确计数字母，而文字聊天可以。用户希望加入语音自定义指令、24/7运行、屏幕共享和硬件设备，认为整体只是过去一年内的微小改进，距离个人助手还有很大差距。
Some first impressions on the GPT-Live voice mode.…
Some first impressions on the GPT-Live voice mode.
Positives:
- Way more natural - More expressive - Conversation flows well, but interrupting it is still tricky sometimes
Negatives:
- Voice mode doesn’t have access to apps, so it can’t read my GitHub, Linear, or email - No wake-up trigger (of course), and no end-call voice trigger either - The “call” times out after an hour (same as AVM), so you still can’t have it on all day - Still can’t count letters, even on high intelligence. At the same reasoning level, text chat can count letters fine. - No transcript on screen - we used to get this [EDIT: it's hidden by default, you swip down to see it, should be visiable by default)
Requests:
- We need voice custom instructions. For now, I’ve used the general setting to change Vale’s voice from posh to a normal South East England accent - Once it’s on 24/7, it could start alerting me about schedule updates, pulse updates, etc. It could also start alerting me about Codex updates. - This would go well with the screen sharing feature we never got on the Windows desktop app - Hardware device, please
Overall, it's a small improvement within the last year or so, but it has a long way to go before it can be a personal assistant. Give it more intelligence and wake up/end triggers.
Just from re-reading my observations from advanced voice mode over a year ago below, I think a lot of that is still relevant too: