HackerNews中文版

我一直在测试4o分析对话的能力，通过将截图上传到移动应用。当我问：“这段对话的语气是什么？”时，我总是得到这样的回复：“请重新上传支持的格式，如.jpg、.png或.webp。当前文件类型无法查看，因此我无法评估语气。”然而，当我使用完全相同的图像，而是问：“人们在这段对话中感觉如何？”时，它却能正确地对截图进行情感分析。更奇怪的是，当我输入类似“我看到回应的语气发生了变化”时，应用只显示一个黑色闪烁的方块（好像在思考），但从未返回任何回复。看起来“语气”这个词可能在输入解析中出现了问题，或者触发了一个静默失败。有没有其他人遇到过这个问题？这可能是一个提示分类的边缘案例，还是图像处理中的一个bug？

查看原文

I've been testing 4o's ability to analyze conversations by uploading screenshots to the mobile app. When I ask:"What's the tone of this conversation?"I consistently get this response:"Please re-upload the image as a supported format like .jpg, .png, or .webp. The current file type can't be viewed, so I can't assess the tone yet."However, when I use the exact same image and instead ask:"How are people feeling in this conversation?"it responds correctly with an emotional analysis of the screenshot.Even stranger one is when I type something like:"I see the tone of the response has changed"the app just shows the black flashing square (as if it's thinking), but never returns a response.It seems like the word "tone" might be breaking something in the input parsing or triggering a silent failure. Has anyone else run into this? Could this be a prompt classification edge case or a bug in the image pipeline?

请问HN：在移动设备上，"tone"这个词会导致ChatGPT-4o的图像分析出现问题吗？