I am someone who much prefers speaking to typing when I interact with AI. Every time I try to explain what I want to Gemini on the web, Claude Code on desktop, or Antigravity on desktop, I feel annoyed.
Do people really like typing on keyboards?
Some do. Just not me.
Though whether you need to type or not does depend on the situation~
So I asked AI to help me solve this pain point.
Now I only need to open the terminal — or ask AI to open it — hold down a key, speak, and let go. What I say appears wherever my cursor is!
When it comes to voice input, I really have to praise ChatGPT. I doubt any voice-recognition tool can beat its accuracy, especially when Chinese and English are mixed together.
I used to use Apple Keyboard's voice recognition as an in-between way to input Chinese, but its accuracy with mixed Chinese and English, and with some words, was far below what I wanted. Compared with ChatGPT, none of the other tools I have used can compete.
Wispr Flow is also very accurate! That was the first thing that struck me when I tried it, though I do not know whether it uses the same speech-recognition model as ChatGPT. 🤔
All in all, the recognition accuracy of the little program I made is pretty good so far! Thank you, AI, for helping me solve another pain point of mine.
Recently I have been immersed in information from the English-speaking world, and the huge gap makes me uneasy. Some people already use AI to work for them all day, while some people have not used AI at all — especially coding AI.
What coding AI in China is really worth mentioning... I thought of TRAE? But when I first used it, the models were freely available and maybe too many people were using it. I would enter one instruction, it would work for a while and then stop; continue working, then stop again... I gave up, and have not opened it since.
Both AI and AI coding have truly empowered many kinds of work. They have also brought many people — designers, creative people, PMs, even professionals who know nothing about code — into the world of building software.
This kind of development was beyond what anyone expected. Who knew that a little internal tool used by the Claude team would become something developers around the world use like crazy? 🤔
It reminds me of an article I read before, ‘Why We've Tried to Replace Developers Every Decade Since 1969,’ about why people try to replace developers every ten years.
The analysis was interesting to me as an outsider. It gave me a glimpse of development history. The article begins in 1969, moves through COBOL in the late 1960s and 1970s, CASE tools in the 1980s, Visual Basic and Delphi in the 1990s, then web frameworks, low-code, and no-code after 2000... It ends by saying that AI has changed how developers work, but has not removed their judgment.
AI does not only augment developers. As I said above, it also empowers professionals in many fields to take part in building software.
I used the word ‘professionals’ there. Yes — I think that is also why I saw news a few days ago that tldraw no longer accepts external PRs. Some people simply do not know what they are doing with AI. They think they understand, and the result is a lot of unnecessary trouble for maintainers.
AI lets many people understand how to make certain things. But AI still cannot make some people know whether what they made should exist. People say everything that exists has a reason, but that feels... like a strange reason. I still do not know how to argue against it. 🤔
That is why most of the AI tools I make are only for solving my own pain points or for fun. The tools I really want to make, the ones that could count as empowering, are related to audio. But I am not an expert in that field. I still cannot fully know what technical problems around audio I do not know yet... But I really have tried to make something. The MVP just has not come out yet 😂
Anyway, trying is a good thing. I have learned many things I had never touched before through vibe coding — at least for me.
Over the past month or so, I have grown very fast in getting to know and using AI. But I cannot reproduce everything I have learned one by one for friends who want to learn, and that troubles me. It became the next pain point I tried to break through with AI. My progress is 40% — an imaginary number, but it can turn my vibe-coding process into visible traces. That is quite a big step, even though it is still barely useful.
Here I want to add two ways I use AI to learn.
[A] Say I see a very inspiring article, podcast, or video on social media. It mentions learning paths or practical steps. The simplest way is not to copy it myself. It is to feed the material to AI and let it point the way — find a shortcut — and tell me how I can get the result I want faster. This process also makes me clearer about what I actually want.
[B] Or someone mentions something they recently made that you find interesting, but they do not explain how. Take a screenshot to AI. Even if you are completely confused at first, after a few rounds of questions and answers, you will definitely gain something.
Whenever I see someone move their eyes and hands but not their brain, then get criticized for just asking others to hand them the answer... Well, you need to consciously break through your own limits. Awareness leads to action, action changes results, and the results feed back into awareness... I understand the logic, but I have fallen into the same trap too...
Of course, for example A, it is not impossible to follow everything from beginning to end. But sometimes you find that an author's words carry context. For example, they give you a long list of CLI commands and ask you to follow them... Wait, what is a ‘CLI command’?
CLI means Command Line Interface. What we call CLI commands are commands entered in a terminal.
Honestly, even though I graduated in electronic information engineering, I hardly touched a terminal during university 🥲 I do not know whether that was my problem or a problem with the education I received. When I tried to connect Claude Code in the terminal, I looked at other people's tutorials and my eyes went dark: bro, what are you even saying??!
I wanted to cry. Really! I wanted to cry.
It reminds me that early this morning I was struggling with my vibe-coding project when I suddenly found that GitHub had suspended my account — even though I had been using it a few minutes earlier. I looked for the reason and guessed it might be because I had registered for a Gemini Hackathon-related platform the day before. A few minutes before I noticed the suspension, I had received an email saying my GitHub account had successfully authorized Devpost...
I started an appeal, then found that GitHub does not support Chinese phone numbers. I had to ask a friend in the U.S. to lend me hers to get a verification code. My chest felt full of a dull frustration through all of it, because it was delaying my project work.
Luckily, the manual review passed that afternoon. They explained it was a false positive and returned my account. But I still felt a serious structural unfairness. Who created this situation? GitHub cannot carry the whole blame.
A few days ago I listened very carefully to Weng Jiayi's interview on the WhynotTV podcast. I found the fatalism he talked about interesting — he was so certain, and he researches AI and reinforcement learning. I have also heard people around me talk about supposedly real mystical events... In the end, maybe I have just seen too few samples. I cannot see through it. I do not know...
How did the mood get so strange? Let me force a turn. Let's be a little sun with energy.
The unknown is so interesting, and it feels good to keep doing things with both feet on the ground.