@Phoenixyin13: Today, I will thoroughly teach you how to read a paper efficiently and with high quality in the AI era. S. Keshav's 'How to Read a Paper' is indeed a classic that has stood the test of time in academia, directly solving the problem with a three-step method. Check and see if you usually read papers like this: First pass, funnel. A…

Today, I will thoroughly teach you how to efficiently and effectively read a research paper in the era of AI.

S. Keshav’s How to Read a Paper is indeed a classic that has stood the test of time in academia. It solves the problem in a straightforward three-pass method. Let’s compare: Is this how you usually read a paper?

First Pass: Funnel.
AI is responsible for compressing 50 pages of noise into 5 minutes of signal. But whether this article gets into your brain’s memory still depends on human intuition and topic selection taste.

Second Pass: Scaffolding.
AI explains complex formulas and interdisciplinary background, reducing cognitive load so you can focus on examining the core logic and figures without distraction.

Third Pass: Adversarial Training.
Treat AI as a free, 24/7, and extremely harsh reviewer. Through multiple rounds of prompt bombardment, force yourself to find flaws in the paper.

When I was young, I loved reading The Heaven Sword and Dragon Sabre. I personally like to compare scientific research to the “Great Shift of the Universe.” AI is like a peerless divine weapon; it can make you strike a hundred times faster. But without the underlying internal skills like Keshav’s “Nine Yang Divine Skill,” newcomers can easily go astray.

What truly lifts the ceiling of research is precisely those things AI cannot replace: taste and skepticism.

A few years ago, when I was doing research at Zhejiang University, a professor told me over dinner that while mentoring undergraduates, he had observed an interesting paradox:

The more advanced the tools, the easier it is to create academic illusions.

In the past, if a student didn’t understand a paper, the report they submitted would be messy and awkward.
Now, with the help of large language models, they can easily produce a perfectly structured, elegantly worded summary that looks flawless.
But this is like eating a beautifully coated sugar bomb. Once you peel off the pretty AI-generated shell, the insight into the study’s limitations and the intuition about baseline flaws are often non-existent.

Therefore, in AI-era research training, the second step to teach newcomers is active wall-breaking.

In the second and third passes, don’t just look at sentences like “we improved by 10%.” Let AI help you abstract the logical framework of the experiment setup, control variables, and ablation studies.

Close your eyes and think:
If I were the author, facing the same dataset and hardware constraints, how would I design this baseline? Why did the author miss that obvious control group?
Was it due to objective limitations, or was it intentionally avoided?

In short, never ask AI:

“What is good about this paper?”

That’s foolish and won’t help you find the focus.

Drawing from my experience at Zhejiang University, I habitually push AI to be your adversary. Let it apply the strictest academic standards to stress-test you.

Only when you can counter each of AI’s challenges—finding answers or refutations in the original text—have you truly completed what Keshav calls the Third Pass.

All rivers flow to the sea. Every published top conference or journal paper represents only the successful 5% of the author’s efforts. The remaining 95%—failed attempts, painful tuning, and compromises after rejection—are hidden between the lines.

Research tools change, paradigms shift—from hand-copied literature to Google Scholar, and now to ChatPDF and agent ecosystems. But the essence never changes.

Research is the art of asking questions, not the repetition of answers.

In the age of AI, may you save the mental bandwidth freed up for the purest skepticism, the most unconstrained conjectures, and the most stubborn pursuit of truth.

This is perhaps the compass every new researcher should hold tightest on the journey to finding the one true way.

paperpaper (@paperpaper886):
Recently, I’ve been guiding undergraduate interns joining the lab. I’ve found that how to read papers is often the most overlooked step in research training.
I recommend a classic short article every new researcher should read: How to Read a Paper by S. Keshav.

The article proposes a very practical “three-pass method” for reading papers:
First pass: 5–10 minutes of quick scanning: title, abstract, introduction, section headings, conclusions, and references.
Goal: answer the 5 C’s?