The Better AI Gets at Finding Information, the More Judgment Matters
And the only thing I still need to do is make judgments.
When I first did research while I was in school, I remember that even at my most efficient, I could only read four or five papers a day.
But I had one particularly strong ability back then: finding information through all kinds of channels. I could find information faster and more efficiently than most people.
That ability went on to run through much of my work over the following years.
For example, I could always find the information I needed faster, bring it together, assess it, and analyze it, then produce a relatively high-quality business analysis report in the shortest possible time. That was valuable for sales-driven and data-driven businesses and companies.
All in all, for a long time, this was a very desirable ability to have.
As I interact with AI more and more, I’ve found that this ‘very desirable ability’ can already be replaced by AI.
These days, I hand almost everything that involves finding an answer or information over to AI, and it can produce a result in minutes.
The more I talk to it and get to know its personality and temperament, the better I understand how to ask it questions that get me information more efficiently and accurately.
And the only thing I still need to do is make judgments.
I’ve been thinking: if AI had existed back then, I could have read dozens of papers a day, becoming many times more efficient.
I have to concede: people shouldn’t try to compete with AI on the efficiency of this kind of ‘mechanical work.’
At the same time, though, I’m very clear that it still can’t truly replace people when it comes to reading papers.
You can use it to go through and summarize a paper very quickly, or to quickly find the most relevant literature. But researchers still have to understand and assess the experimental methods, conditions, and data in important papers themselves. AI still isn’t up to that task.
So this is interesting: what will produce the best results in the future is not whom AI replaces, but what kind of collaboration people form with AI. Complementing each other seems more important.
A few days ago, I was listening to a podcast and heard that top-ranking token spender whom Claude Code was ‘hunting down’ say that he was working with a professor to launch an AI product called Citely.
It’s a tool for working with research papers, focused on the problem of ‘fabricated references’ in paper writing in the AI era, making sure references are checked for authenticity and assessed for quality.
The product doesn’t try to be ‘all-purpose.’ It just takes the single function of reference verification as far as it can, saving a great deal of time spent checking references by hand.
At first, most people might think this product sounds simple. But when you think it through, it’s actually far from simple.
I imagine it involves working with many databases, handling and cross-checking information sources from different channels, tracing cross-references, and so on. There should be a substantial amount of engineering involved.
But I think this highly specialized product will have great value for universities and research institutions in the future.
So when Liu Xiaopai (刘小排, the speaker mentioned above) said that this product was something ‘other people can’t do, but I can keep doing for twenty years,’ I suddenly felt warmth and respect toward him.
Compared with making ‘traffic-driven products,’ perhaps this is the kind I like more.
Still, he said something that I think is quite right, and I agree with it: ‘人生是旷野，有矿才能野。’ (‘Life is an open wilderness, but you need resources to roam free’—a pun on 矿, ‘mine,’ and 野, ‘wild.’)
Hahahaha.. Keep going!