The author reflects on the anxiety sparked by AI’s rapid progress: machines far surpass individuals in knowledge and efficiency, so the real competitive edge lies in choosing the right questions, having access to exclusive data, conducting experiments and validating results, executing across disciplines, and adjusting course in time. The author is still exploring how AI can advance academic work and wonders whether to change careers, but sees recognizing the problem and trying to adapt as progress in itself. Attending CCP4 Shanghai School deepened their understanding of crystallography, though they feel they have little to share and that the official tutorials are already comprehensive. Professionally, the author believes that structural biology alone offers too little of a competitive edge. Without straying from their core work, they plan to use AI to explore computational biology, including molecular dynamics, docking, and virtual screening.
It's been half a year — time to share a few recent thoughts.
The AI Era
Over the past six months I haven't updated much — no travel posts, no academic posts. I could make excuses and say I've been too busy, and that's objectively true. But the main reason is that these days I find it hard to write content that is both high-quality and irreplaceable.
A travel post without photos isn't a good travel post, and as someone who fusses over photo editing, retouching and picking shots eats up a lot of time — which is why the photos from several trips are still untouched. Besides, a plain travel post is just a record of my own life. Honestly, readers would get too little out of my posts; spending time on my travelogues is less rewarding than searching Xiaohongshu for a more comprehensive take (after all, some people do this for a living).
Writing academic content is an even bigger headache. ChatGPT 3.5 came out around this time three years ago, and back then generative AI was riddled with hallucinations and utterly untrustworthy. It couldn't go online either, so there was no way to verify its output; and without reasoning ability, its logic was often muddled, which made its output basically unusable except for writing text that reeked of AI.
But things are very different now. In coding, Cursor and Claude Code are incredibly powerful, and the array of MCP integrations is dizzying; in text-to-image, Nano Banana is jaw-dropping; and in research, ChatGPT with web access and reasoning has become a real boost for me too. It makes me marvel at how strong AI has gotten — strong enough that $20 a month can dramatically raise a person's competitiveness.
Just a couple of years ago, the news still often featured people who insisted on not using AI at work, and their companies didn't require it either. Today, an employee who doesn't use AI is almost guaranteed not to be favored by any company — possibly even disliked. Some say AI is developing far slower than expected, but the truth is it's only been three years since ChatGPT. AI moves so fast that every so often a flood of new knowledge appears and there's too much to read; every so often a product comes along that reshapes an industry (Cursor, Nano Banana, and so on). This pace constantly makes me feel out of step with the times, and I can't help worrying that I'll miss my chance in this wave — or even lose my job (though the odds are quite low).
Which brings me to a question: in this era, what exactly is my moat? It's been the question nagging at me lately.
“
A moat comes from 'picking the right problems + owning exclusive data and samples + building scalable experimental/computational platforms + rigorous validation and compliance + the ability to execute cross-disciplinary integration.' AI is an accelerator, but 'problem selection, causal inference, experimental constraints, and quality systems' are what determine irreplaceability.
In terms of knowledge and reading speed, machines have such high throughput that humans can't compete no matter what. Pose the same question to a PhD student and to AI, and today the AI will most likely give a more thorough and complete answer. And such a machine costs just $20, while a human needs over twenty years of training. That inevitably makes me pessimistic, and it forces me to think about how to use AI well.
I just finished watching the podcast with Lao Luo and the man in the red shirt, which also touched on the so-called moat problem several times, mostly in the context of companies. The man in the red shirt argued that being able to adjust strategy in time is itself a company's moat. Applied to personal development, I think that makes sense too.
Unfortunately, my exploration of AI for an academic career has so far been limited to chatting with ChatGPT — the more I think about it, the more I feel I should quit and move into another industry. That kept me anxious for a while, and I still don't have a great solution. But on second thought, just being aware of this and trying to make a change is already something.
CCP4 Shanghai School
Apparently, the last CCP4 Shanghai School was five years ago. It was pure luck that I got to attend this time — it happened to be held nearby, which was a rare opportunity.
I'd already gained a lot by the end of the first morning, so I stuck with it and attended nearly all ten days of lectures, even sneaking into a few days of tutorials. I got quite a lot out of it and came away with a deeper understanding of crystallography. Even without this knowledge, I could still solve protein crystal structures and complete the deposition, but there was always a lingering feeling of being undertrained.
I'd thought about refining what I learned at the conference and sharing it, but once I created a new folder I realized there was only so much I could share. For one thing, I only know the surface — I might know a bit more than most, but it's still just a smattering; and in the foreseeable future I'll have very few chances to actually do crystallography, which is a real pity. For another, the official CCP4 tutorials are already quite detailed, and I don't want to reinvent the wheel. A post like that wouldn't add much value, so I'd rather spend the time elsewhere — I deleted the folder.
Of course, if there's anything you'd like to discuss with me, I'd be very happy to — feel free to get in touch.
Computational Biology or Bioinformatics
This is another question I've been mulling over lately.
Structural biology on its own is starting to feel a bit underwhelming to me. For one thing, a structure itself often doesn't say all that much, handy as it is as a tool. For another, if I want to stay competitive, I think there are two things I need to consider: branching out into more skills, and moving closer to drug discovery. Otherwise, doing structural biology without moving toward applications feels like letting a lot of opportunities slip by.
Quite a few of my friends work in bioinformatics, and their lifestyle is one I envy; the seniors in that field have also landed on good career paths. That naturally makes me wonder whether I could combine bioinformatics with structural biology.
But exploring the idea left me a little disappointed. The overlap between bioinformatics and structural biology is quite limited, whereas computational biology fits perfectly. So I'm planning to dip a toe into computational biology, including but not limited to:
Molecular dynamics simulations
Molecular docking / virtual screening
I'll hold off on anything deeper for now, since my main job is still structural biology — I don't want to put the cart before the horse. Let's see if I can make something interesting out of it.
Of course, I plan to have all of this powered by AI — that'll make it faster.
Wrap-up
I feel like there's still plenty more I could talk about, but I'm out of writing steam. Let's stop here for now — I don't know when the next post will be.
Long Time No See
It's been half a year — time to share a few recent thoughts.
The AI Era
Over the past six months I haven't updated much — no travel posts, no academic posts. I could make excuses and say I've been too busy, and that's objectively true. But the main reason is that these days I find it hard to write content that is both high-quality and irreplaceable.
A travel post without photos isn't a good travel post, and as someone who fusses over photo editing, retouching and picking shots eats up a lot of time — which is why the photos from several trips are still untouched. Besides, a plain travel post is just a record of my own life. Honestly, readers would get too little out of my posts; spending time on my travelogues is less rewarding than searching Xiaohongshu for a more comprehensive take (after all, some people do this for a living).
Writing academic content is an even bigger headache. ChatGPT 3.5 came out around this time three years ago, and back then generative AI was riddled with hallucinations and utterly untrustworthy. It couldn't go online either, so there was no way to verify its output; and without reasoning ability, its logic was often muddled, which made its output basically unusable except for writing text that reeked of AI.
But things are very different now. In coding, Cursor and Claude Code are incredibly powerful, and the array of MCP integrations is dizzying; in text-to-image, Nano Banana is jaw-dropping; and in research, ChatGPT with web access and reasoning has become a real boost for me too. It makes me marvel at how strong AI has gotten — strong enough that $20 a month can dramatically raise a person's competitiveness.
Just a couple of years ago, the news still often featured people who insisted on not using AI at work, and their companies didn't require it either. Today, an employee who doesn't use AI is almost guaranteed not to be favored by any company — possibly even disliked. Some say AI is developing far slower than expected, but the truth is it's only been three years since ChatGPT. AI moves so fast that every so often a flood of new knowledge appears and there's too much to read; every so often a product comes along that reshapes an industry (Cursor, Nano Banana, and so on). This pace constantly makes me feel out of step with the times, and I can't help worrying that I'll miss my chance in this wave — or even lose my job (though the odds are quite low).
Which brings me to a question: in this era, what exactly is my moat? It's been the question nagging at me lately.
In terms of knowledge and reading speed, machines have such high throughput that humans can't compete no matter what. Pose the same question to a PhD student and to AI, and today the AI will most likely give a more thorough and complete answer. And such a machine costs just $20, while a human needs over twenty years of training. That inevitably makes me pessimistic, and it forces me to think about how to use AI well.
I just finished watching the podcast with Lao Luo and the man in the red shirt, which also touched on the so-called moat problem several times, mostly in the context of companies. The man in the red shirt argued that being able to adjust strategy in time is itself a company's moat. Applied to personal development, I think that makes sense too.
Unfortunately, my exploration of AI for an academic career has so far been limited to chatting with ChatGPT — the more I think about it, the more I feel I should quit and move into another industry. That kept me anxious for a while, and I still don't have a great solution. But on second thought, just being aware of this and trying to make a change is already something.
CCP4 Shanghai School
Apparently, the last CCP4 Shanghai School was five years ago. It was pure luck that I got to attend this time — it happened to be held nearby, which was a rare opportunity.
I'd already gained a lot by the end of the first morning, so I stuck with it and attended nearly all ten days of lectures, even sneaking into a few days of tutorials. I got quite a lot out of it and came away with a deeper understanding of crystallography. Even without this knowledge, I could still solve protein crystal structures and complete the deposition, but there was always a lingering feeling of being undertrained.
I'd thought about refining what I learned at the conference and sharing it, but once I created a new folder I realized there was only so much I could share. For one thing, I only know the surface — I might know a bit more than most, but it's still just a smattering; and in the foreseeable future I'll have very few chances to actually do crystallography, which is a real pity. For another, the official CCP4 tutorials are already quite detailed, and I don't want to reinvent the wheel. A post like that wouldn't add much value, so I'd rather spend the time elsewhere — I deleted the folder.
Of course, if there's anything you'd like to discuss with me, I'd be very happy to — feel free to get in touch.
Computational Biology or Bioinformatics
This is another question I've been mulling over lately.
Structural biology on its own is starting to feel a bit underwhelming to me. For one thing, a structure itself often doesn't say all that much, handy as it is as a tool. For another, if I want to stay competitive, I think there are two things I need to consider: branching out into more skills, and moving closer to drug discovery. Otherwise, doing structural biology without moving toward applications feels like letting a lot of opportunities slip by.
Quite a few of my friends work in bioinformatics, and their lifestyle is one I envy; the seniors in that field have also landed on good career paths. That naturally makes me wonder whether I could combine bioinformatics with structural biology.
But exploring the idea left me a little disappointed. The overlap between bioinformatics and structural biology is quite limited, whereas computational biology fits perfectly. So I'm planning to dip a toe into computational biology, including but not limited to:
I'll hold off on anything deeper for now, since my main job is still structural biology — I don't want to put the cart before the horse. Let's see if I can make something interesting out of it.
Of course, I plan to have all of this powered by AI — that'll make it faster.
Wrap-up
I feel like there's still plenty more I could talk about, but I'm out of writing steam. Let's stop here for now — I don't know when the next post will be.