每一次互动都是一次图灵测试

2作者: scottfalconer大约 1 个月前原帖
我每天至少看到一封邮件或一条LinkedIn帖子,明显是“由AI撰写”的。 使用AI没问题,但如果没有人类的痕迹,我就会在中途失去兴趣……就像发现一张酷炫的照片是假的一样。 我们为什么能如此快速地识别出机器人的声音? * 我们会根据受众的反应进行调整,而模型则是对所有人进行平均。 * 真实的人会留下痕迹……比如恰到好处的笑话,或者不小心的错别字。 * 精致而通用的文风就像是一个本地新闻主播闯入你的群聊。 我如何保持草稿的人性化: 1. 明确说出谁在说话,为什么。“客服代表,对一个感到沮丧的客户。” 2. 粘贴客户的真实抱怨。这让模型感到压力。 3. 加入一段我自己的声音,要求它进行修改。这比遵循规则效果更好。 4. 将温度调高一点,然后删掉任何奇怪的内容。 5. 大声朗读。如果听起来像企业的墙纸,我就会让它变得粗糙一些——把逗号换成破折号,加入一个“哎呀”。 这就是我的日常。你的呢?我在收集一些技巧,以便我的收件箱不再像办公室天花板的掉落瓷砖。
查看原文
I see at least one email or LinkedIn post a day that screams &quot;written by AI&quot;. Using AI is fine, but if there’s no human fingerprint I drift off midway...same reaction as if I realize a cool photo is a fake.<p>Why do we spot the bot voice so fast?<p>* We riff off our audience. Models average everyone.<p>* Real people leave fingerprints...jokes that land, typos that don’t.<p>* Polished‑generic prose feels like a local news anchor gate‑crashing your group chat.<p>How I keep drafts human:<p>1. Spell out who&#x27;s talking and why. &quot;Support rep, one frustrated customer.&quot;<p>2. Paste the customer’s actual rant. Makes the model sweat. 3. Drop a paragraph in my voice, ask it to revise. Works better than rules.<p>4. Crank temperature a notch, then chop anything weird.<p>5. Read it aloud. If it sounds like corporate wallpaper, I rough it upswap a comma for an em dash, toss in a &quot;whoops.&quot;<p>That’s my routine. What’s yours? I’m collecting tricks so my inbox stops looking like office ceiling drop in tiles.