When ChatGPT was first launched, I tried using it to plan a trip, but the experience was quite poor. It often hallucinated, got addresses wrong, or recommended non-existent attractions. Since the interface was entirely text-based, I couldn't see the locations of the attractions on a map. I eventually had to resort to manually arranging the itinerary and confirming accommodations using maps. It felt like these applications were more gimmicky than practical.
As I was planning a trip to Vancouver and had spent two years delving into how AI could improve my life and work, I decided to give these tools another try. Surprisingly, the experience was significantly better this time. I shared this chat history here.
In short, I first asked the AI to recommend types of attractions and suitable areas to visit in Vancouver. I then informed it that we had seniors and children, which helped narrow down the options. Next, I asked it to mark these attractions on a map, allowing us to visually grasp the distribution of the sites and determine the final itinerary sequence and transportation methods. Ultimately, with the AI's help, we generated an overall itinerary.
The entire chat might seem unremarkable, but if I had attempted it two years, or even a year ago, I likely wouldn't have been able to complete it so smoothly. During this process, I primarily used the following techniques:
- Using different AI for different tasks, such as using web search AI to reduce hallucinations and employing more intelligent AI like Claude for reading documents or generating code.
- Using document management to efficiently teach the AI new content it previously didn't know. For example, by pasting Bing Maps documentation to teach the AI how to generate a URL, allowing us to view all the attractions' locations directly on Bing Maps.
- Using Chain of Thought to avoid potential pitfalls. For instance, guiding the AI to use addresses rather than coordinates to mark locations on the map.
But in reality, these technical tricks just scratches the surface. The improved AI experience and powerful functionality are not because AI is smarter than before, or because the tool I use, Poe, is better than other clients, but because I have adopted a Builder's Mindset.
The Builder's Mindset refers to our attitude towards tools shifting from passive to active. We are no longer just users of tools; we are builders of tools. When there are no ready-made tools, we build them; when existing tools aren't user-friendly, we improve them. In the AI era, this building and improvement have become particularly simple. For example, although AI currently lacks a feature to directly display recommended attractions on a map, we can instantly teach AI to perform this function by merely pasting the URL of Bing Maps' help documentation. Reading documents, understanding parameter meanings, constructing the correct URL—these tasks that previously required our manual effort can now be handed over to AI.
Therefore, those with a Builder's Mindset may be the biggest beneficiaries in the AI era and will likely distance themselves significantly from passive users. However, this shift in mindset is not achieved overnight. On one hand, it's a specialized field with many pitfalls and tricks; on the other hand, it requires time and experience to accumulate. For instance, the reason I know about this feature of Bing Maps is because I learned related knowledge while building other tools in the past.
If I were to teach you just one thing from my AI usage experience over the past two years or even longer, I would unhesitatingly choose the Builder's Mindset. This is also one of the reasons why Kedaibiao and I co-created the From Users to Builders AI course. Interested students can join our latest free trial lesson, or check out the feedback from participants in our live course.
Comments