Spent the Weekend Tinkering with Kimi and Hit Plenty of Pitfalls

I messed around with the Kimi smart assistant over the weekend and hit quite a few snags. This thing really does have some chops when it comes to long-text processing, but it's not the kind of product you just open and use—you've got to know how to feed it instructions.

The first day was basically blind trial and error. I logged in with my phone number, and the interface was very clean—one input box, a row of buttons below. My first instinct was to just dump a 20,000-plus-word industry report PDF into it, tap the plus sign to upload, and then ask whether it could extract the core points. After a few seconds, it actually listed them out for me, and even picked out the contradictions between several reports while it was at it. That step was a pleasant surprise—I'd assumed I'd have to feed it in chunks.

But then I made a mistake. I asked it to "write me a review script," and the draft it gave was very flat, utterly safe, the kind of thing anyone could write. Only later did I realize the problem was on my end. That line in the source material is right—AI isn't a person, so just telling it "I believe you can do it" does nothing for it; you have to give it a role, skills, and a task. I changed it to "You are a tech reporter who's covered the AI beat for three years; write a colloquial hands-on review, no summarizing or grand conclusions," and what came out was immediately different.

The third day I started to get the hang of it. It has this @-to-call-tools design—typing @ brings up things like a PPT assistant. I tried having it turn an interview transcript into an outline, then convert that into a PPT draft, all in one go without a break. But I got stuck here too—at first I didn't know what @ could do, and it took me ages of poking around to figure it out. That entry point isn't very friendly to newcomers.

After a week my usage has basically settled into a pattern. I throw long documents at it to read, cross-compare multiple materials—this kind of work it does faster than me, and I don't have to watch over it. Writing I still do myself; at most it helps me bang out a draft or rephrase something. I also tried image recognition—uploading screenshots and having it read the text inside. It works, but the accuracy is so-so; if it's even a bit blurry it can't make it all out.

Let me be clear about the downsides too. Its strengths are concentrated in long text and organizing materials—if you're expecting it to be an all-purpose chatting buddy, you'll be disappointed. The instruction barrier is real; people who don't know how to ask questions will find it pretty mediocre. The tool entry points are tucked away a bit deep, and you have to figure them out yourself.

My verdict is: it depends. For people who constantly have a pile of PDFs, reports, and interview transcripts to read and organize, it's worth using—this kind of work really does save time. If you just want an AI to chat with casually or write WeChat Moments copy, something else is less hassle. This thing is like a capable intern—you've got to know how to assign the work.