Recently, I made a decision to give up Devin, an AI development tool costing 500 USD per month (because I can replicate it with cursor), but renewed my subscription to ChatGPT Pro for 200 USD per month. The reason is that I gradually realized that o1 Pro has aspects that other models simply cannot replace. While the value of other models primarily lies in making me aware of things I didn’t know before, the value of o1 Pro is in helping me gain insights that I wouldn't have thought of even with sufficient information.
If traditional AI models (like GPT-4o and Claude) are like a secretary helping me with information gathering and organizing such physical labor, then the regular o1 subscription is like a seasoned assistant who can spot gaps in my viewpoints and calculations and identify obvious blind spots in my thinking. My expectations from them are more about refining the details.
However, o1 Pro has elevated my AI experience to a whole new level. It’s no longer just a tool; it feels more like a mentor who can guide me to delve deeper into thinking and expand my cognitive boundaries. To put it simply, my attitude towards GPT-4o is “help me get this done,” towards regular o1 is “this AI is good and knows these things,” and my attitude towards o1 Pro is “Mentor, teach me,” genuinely feeling like I’ve learned something.
Let me give two examples: The first is in our group of seasoned master photographers, where there are very specialized and complex quantitative computation problems related to PQ curves. One of the experts tried to replicate a classic result using standard calculations but always ended up two off, so he asked in the group what was going on. This is a very specialized field, and even for someone like me who’s an old master but not a professional in the field, just knowing the basic principles doesn’t help identify issues in the calculation process.
I also tried asking various AIs, including GPT-4o and the regular o1 model, but the answers were all very superficial. Even when provided with specific technical documents (to avoid hallucinations), they couldn’t figure out the exact reason. They could only vaguely mention that it might be some details or parameter differences, without pinpointing the specific issue.
But after trying o1 Pro, after about ten minutes of thinking, it pinpointed a conceptual error in our understanding, explained a fundamental issue with our calculation method’s framework, and then, through detailed calculation steps, arrived at the official result. After seeing o1 Pro’s results, the experts in the group also agreed that its calculations were correct.
The second example is that, inspired by the class representative, I’ve recently become interested in Funnel Analysis and wanted to understand it further. Similarly, I first asked the regular o1, then the Pro mode, and made a comparison. The regular o1’s response indeed systematically and thoroughly introduced the basic concepts and applications of Funnel Analysis, and provided a detailed example analysis. This is useful in itself, but it’s like a summary that an ordinary practitioner could give after appropriate preparation. Clear and well-organized, it’s good, but it feels like something is missing—the surprise factor.
However, o1 Pro’s result felt particularly deep. It was more like an industry veteran sharing their profound professional experience. I read it several times and felt a sense of “Wow, I’ve learned something new.” I also asked the class representative about his feelings on this analysis, and he felt that the depth of this analysis could surpass 95% of data analysts.
But one thing to note is that the techniques for prompting o1 Pro differ from those used with other LLMs. Often, to achieve deep and insightful results, we not only need to provide specific information and requirements but also inspire and motivate it to truly invest time in thinking. For example, in the prompt, we need to tell it: “Don’t treat what I say as fixed requirements; you need to have your own independent thinking and analysis.” Or say, “Think backwards about what constitutes a good answer, then self-check against this standard, preferably giving me some surprises.” Or, “Analyze the assumptions behind my questions, especially consider whether my question might be flawed. If we ask the right question, the insights or notices it gives me would be greater than answering a wrong question.” Other models might not understand or implement such prompts, but o1 Pro often provides very deep and sometimes surprising results after such prompts.
So, in summary, I decided to renew the ChatGPT Pro Plan because o1 Pro indeed offers thinking depth that other models cannot provide, even helping me break through my original cognitive limits. In my use case, I have also felt that the value it brings me indeed surpasses the two hundred dollars. This value is something that other models currently cannot provide.
Comments