A couple of days ago, I helped a friend with her master's thesis. Her supervisor required her to re-implement and validate the modeling for some of the studies cited in her paper using code. This kind of request is quite common in research, but she was completely stuck at this step because she had no coding experience at all.
She had already struggled for half a day, trying various domestic AI tools, but the code either wouldn't run or produced nonsensical results. In the end, she had no choice but to ask me for help.
By the time she reached out to me, she was already on the verge of a breakdown. I recommended she try ChatGPT again and refine her prompts, but after another round of attempts, it still didn't work.
So, I asked her to send me the materials. When I opened them, it was a PDF. At that point, I had a basic suspicion: the issue likely wasn't the model's capability but the input format. I threw the file into Antigravity and ran it through the Claude model inside. The model quickly provided a rough modeling plan but also noted that the current environment couldn't directly parse the PDF content. Essentially, it was searching for information online based on the title and generating inferences using common research methods. In other words, it hadn't actually read the paper.
So, I did something very simple: I used PDFgear to export the PDF as images and re-input them into the model. After running the code a few times and fixing the bugs that emerged, the mathematical model ran smoothly. If you exclude my intermittent complaints and her emotional breakdown on the other end of the internet, the whole process took about half an hour.
This experience gave me a strong feeling about a saying: the AI era is the best of times, but also the worst of times.
From an efficiency perspective, this is indeed a very good era.
If we rewind a few years, completing mathematical modeling often required reading papers, understanding research methods, looking up information, writing code, and repeatedly adjusting parameters. Even for someone skilled, it could take a full day. Now, the process is much simpler: let the model handle understanding the paper structure and generating the code framework, while humans only need to do some checking and debugging. Many tasks that only professional researchers could complete in the past can now be done by ordinary people with the help of tools. As long as you know how to ask questions, verify results, and fix a bit of code, you can break down complex tasks and solve them step by step.
At the same time, this is also a somewhat frustrating era.
My friend was stuck simply because of an "incorrect input format" issue. The model didn't read the paper content, fabricated a bunch of nonsense, and naturally, the entire process couldn't proceed. Before she reached out to me, she had spent a lot of time on various "magical" AI tools, but none of them clearly pointed out the real problem. Only Claude noted that the current environment couldn't parse the PDF. All it would have taken was a simple step: asking the user to convert the PDF to images before inputting it, or directly generating a PDF parser, and the problem would have been solved. But this step wasn't proactively explained by most tools.
This incident has made me increasingly feel that the real gap between top-tier models may not necessarily lie in intelligence itself. People like to discuss model parameters, leaderboards, and benchmarks, as if being "smarter" would lead to a completely different experience. But in real-world usage, the difference in experience often comes down to engineering details.
Whether a system can recognize input format issues, explain the limitations of its current reasoning, or provide actionable suggestions for fixes—these details are crucial for users. Often, users don't need more complex reasoning capabilities; they just need a clear prompt: "Your input might be problematic."
Lately, many people like to talk about AI Agents, imagining a system that can automatically understand requirements, call tools, and complete complex tasks. It sounds wonderful, but in my view, reality is still far from that stage.
In this small case, the entire process still involved a human first identifying where the problem might lie, then selecting tools, adjusting the input format, verifying the model's results, and finally fixing the code. AI did make many steps faster, but the control of the process remained in human hands. In the short term, this is unlikely to change fundamentally.
What's even more interesting is that the truly important abilities in the AI era haven't actually changed. It's still about understanding problems, breaking them down, judging the boundaries of tools, and combining different tools to solve problems.
Many early internet users are actually quite familiar with this way of thinking. In the days when internet speeds were only a few dozen KB, finding a single image might require scouring forums, FTPs, and various resource sites. Back then, everyone naturally understood that tools had limitations, so they had to constantly try different methods and piece together various tools to get things done.
But today's internet environment is completely different. Most people from Generation Z have grown up with recommendation algorithms and streaming platforms, where content is automatically pushed to them, and they only need to click to play. All tools are perfectly packaged and ready to use out of the box. Over time, people's understanding of tools has actually diminished.
Tools have become more powerful, but many people are increasingly unfamiliar with their limitations. When they encounter a problem not covered by mainstream tools, they often have no choice but to get frustrated or give up.
That's why I increasingly feel that the real gap between people in the future may not be about who has AI and who doesn't, but about who understands tools better. More and more people will know how to use a single tool, but few will be able to combine multiple tools to solve problems.
In the AI era, this ability will become even more important. After all, AI itself is just one of the tools, and tools never solve problems on their own. It's always people who truly solve problems.