I saw this AI Blog Question Challenge1 on blog.imalan.cn and thought it was pretty interesting, so I'm joining in for fun and filling it out.
1. How was your first experience with AI models?
My first experience with a cutting-edge AI was probably around 2023, when GPT-3.5 was just starting to become popular. Since the AI I had encountered before was mostly things like voice assistants on phones, GPT-3.5 did seem much smarter by comparison. But essentially, it was still a "not very easy to use" model mainly for entertainment, let alone for programming. At the time, we even had a course dedicated to using AI to write both front-end and back-end code. This kind of work, which is now considered one of the most suitable tasks for AI, was actually a terrible experience back then. Most of the front-end code still had to be manually adjusted before it "almost looked acceptable," and the whole course took quite a bit of effort. The final result can be seen in this video. But now that Claude Code exists, the same task might be finished in a single day.
2. Do you use AI or are you completely against using it?
Without a doubt, I use it. In the face of the latest technology, insisting on "old-school programming" is, in most cases, just self-congratulatory. What matters more is knowing how to harness AI and understanding the boundaries of AI's capabilities.
3. Do you have any preference among different models, for example Claude vs ChatGPT? If yes, how do you choose?
For everyday tasks, writing-related tasks, or search tasks involving factual information, I tend to prefer Gemini. After all, Google has a natural advantage in these areas thanks to its search engine business. For programming tasks, I tend to prefer Claude Opus or GPT-5.5. Since Claude has lower usage limits and is more prone to banning accounts, in practice I mainly use GPT. I rarely use other lower-tier substitute models.
4. What aspect of AI models do you like and what do you not like?
What I like about AI models—LLMs, specifically—is their humanistic care. You can chat with an LLM as if you were talking to an unbiased friend. What I dislike the most is the unique "AI feel" of AI models, along with all kinds of trashy AIGC that comes with it.
5. How do you feel about AI generated images? Does it annoy you if someone use them in a blog post?
I think this depends on whether they are used with restraint, the quality of the generated images, and their relevance to the article's content. Since I'm not very good at drawing, I use AI-generated images for the covers of some blog posts to illustrate the topic, but at the very least, they need to not be unpleasant to look at. Too many AI images, in my view, are just filler and go against the original intention of using a blog for self-expression.
6. Internet is flooded with AI slop now, full of generated text, images, audio and videos. How do you filter it from authentic human creation? Do you have a strategy?
I usually think AI is just a tool. If users can fully express their creative intent through AI and produce aesthetically pleasing works, that is not actually a bad thing. But the reality is that most AI works flooding the internet are produced by authors trying to cut corners, and compared with works created through traditional methods, they generally lack aesthetic sense. For me, the method of identifying them is simple: feel them with care. For most AI works, I almost always develop an instinctive aversion, even disgust. In text, there are large amounts of formulaic, repetitive phrasing such as "not… but…," "underlying logic," and wheel-spinning clichés. In images or videos, there is obvious smearing, unnatural geometric relationships, uncoordinated colors and lighting, and similar-looking character models. Most of them are "AI at a glance."
7. Are you hopeful for a better future with A.I. or a dystopian one?
I am relatively optimistic about the future brought by AI, because at its core, it breaks knowledge monopolies and brings a certain degree of technological freedom. Thinking back to my freshman year, even writing a VS Code configuration file required searching online for ages. Now I can simply let Claude Code or Codex write it locally, without even having to copy and paste anything myself. If I have a small need that requires writing a script or modifying an open-source project, I can easily accomplish it with the help of AI, which used to be hard to imagine. On the other hand, however, I remain reserved about the future of AI itself—LLMs in particular. It's not just the astonishing number of academic papers published every year with questionable quality; at present, the industry has also placed the focus of LLM research on piling up computing power and training datasets, while some are even enthusiastic about hyping up concepts. Current LLMs mostly provide answers within existing knowledge and rules, the copyright issues of training datasets remain ambiguous, and their innovativeness is questionable. Imagine training an LLM on materials from at least ten years before the discovery of special relativity. Could the current model architecture allow it to break away from the existing knowledge system and independently derive special relativity? If not, then concerns about "AI replacing humanity" are still premature.
Reference
[1] https://blog.rishabhps.com/posts/2026-05-28-ai-blog-question-challenge