I want to share two examples of using AI that didn't go as planned. Initially, they seemed promising, but after some exploration and sharing, they fell flat. Here's what happened:
Example 1: The Short Video
One day, I saw a trending video on Twitter from a famous Chinese TV show. It was a dramatic meme recreated as a learning video. Why not try this in another language? So, I took the famous scene from the movie "Taken" and used it to introduce generative AI. I spent about an hour creating, testing, and sharing it on LinkedIn and other social media. The result? Zero interaction. People didn't engage with it like they do with my other posts 😞
Example 2: The Infographic
I learned about a prompt that generates cartoon-style images from realistic photos. After testing, I created an image using Nvidia's CEO, Jensen Huang's profile picture. It would be interesting to assemble a shareable piece, so I made a timeline for NVIDIA featuring this image as the primary visual element. I shared it on social media, but my performance on Twitter and LinkedIn could have been better. Considering I spent almost two hours on this(a whole night's time for a working dad), the return on investment (ROI) was disappointing.
Here are my learnings from these two failed experiments:
1. Noise Over Signal 📢
In the fast-growing AI industry, there's a lot of hype. New tools and tips pop up daily, and the fear of missing out (FOMO) is real. Emails touting "8 must-have resources for learning AI" or "12 game-changing tools for your productivity" flood our inboxes. But in reality, identifying genuine needs is rare. This means that failed experiments, like my video and infographic, are inevitable. I spent time testing the possibility of the tool rather than providing value from the posts to the audience. That's a good lesson.
2. Shorten the Test and Fail Loop ⌛️
For these projects, I spent 1-2 hours each to quickly put them together and tested them on social media. The results weren't great, but I learned from them and moved on. If I had spent a whole week perfecting the workflow, finding fancy prompts, or setting up automation, only to discover the projects weren't valuable, it would have been a more significant waste of time. I'm glad I only invested a few hours and got quick feedback. I'll keep this short-feedback-loop style for upcoming AI experiments as my primary approach.
That's what I learned about AI this week.
从失败的 AI 实验中学到的两课
我想分享两个使用 AI 的例子，这些例子并没有按计划进行。起初它们看起来很有前途，但在实际执行后，结果却平平无奇。事情是这样的：
例子1：短视频
有一天，我在 Twitter 上看到一段来自著名中国电视节目的热门视频（容嬷嬷教你学习网线接口格式）。这是一个被重新制作成学习视频的梗视频。“为什么不用另一种语言试试呢？”于是，我选取了电影《飓风营救》(Taken)中的经典场景，并用它来介绍生成式 AI (Generative AI)。我花了大约一个小时来制作、测试并在 LinkedIn 和其他社交媒体上分享。结果呢？没有任何互动。人们并没有像平时对待我的其他帖子那样积极反馈 😞
例子2：信息图表
我了解到一个可以将现实照片生成卡通风格图片的提示。在测试后，我用 Nvidia CEO 黄仁勋 (Jensen Huang) 的个人照片创建了一张图片。将其制作成一张可分享的作品会很有趣，于是我用这张图片作为主要视觉元素，为 NVIDIA 制作了一张时间线。我在社交媒体上分享了这张图表，但在 Twitter 和 LinkedIn 上的表现并不理想。考虑到我花了将近两个小时（对于一个忙碌的爸爸来说是一整晚的时间），回报率（ROI）令人失望。
以下是我从这两个失败实验中学到的经验：
1. 噪音胜于信号
在快速发展的 AI 行业中，充斥着各种炒作。每天都有新工具和新技巧出现，错失恐惧症 (FOMO) 的感觉非常真实。我们的收件箱里充满了“学习 AI 的8个必备资源”或“提升生产力的12个革命性工具”。但实际上，识别真正的需求是很少见的。这意味着像我这样的视频和信息图表的失败实验是不可避免的。我花时间测试工具的可能性，而不是为观众提供有价值的内容。这是一个重要的教训。
2. 缩短测试和失败循环
对于这些项目，我每个都花了1-2个小时快速完成并在社交媒体上测试。结果虽然不理想，但我从中学到了经验并继续前进。如果我花整整一周的时间来完善工作流程、找到更好的提示或设置自动化，最后发现这些项目没有价值，那将是更大的时间浪费。我很高兴我只花了几个小时并获得了快速反馈。我将继续采用这种短反馈循环的方式，作为我未来 AI 实验的主要方法。
这就是我本周关于 AI 学到的东西。
全贴地址：
https://bearwith.ai/lessons-from-failed-ai-experiments/
