How Sausages Are Made (and How a Vibe Research Was Born)

It started with an idea.

Months ago, long before there was a paper or even a draft, I found myself circling around a question: why do some reasoning models seem to “fail” in ways that feel strangely human? I kept turning it over in my head on walks, in the car, in those in-between moments when you’re not really working but your mind refuses to let go.

Then one night I stumbled upon the Apple paper, The Illusion of Thinking, and one week later came The Illusion of the Illusion of Thinking. It was like someone had dropped a match into a pile of dry twigs. It didn’t give me the idea, it sparked the one I’d been nursing quietly for months. Suddenly, I couldn’t stop thinking.

I opened a new project, started pulling papers, sketching diagrams, whispering to myself while driving. I even used ChatGPT in voice mode during commutes, pouring my thoughts out loud so I would not lose them. I spinned up every deep search product I had access to, I wrote, rewrote, used Claude and ChatGPT to expand, clarify, and critique the ideas. It was like vibe coding, but in academic writing. For me, this collaboration with an AI was new, so I started a diary.

For me, this collaboration with an AI was new, so I started a diary and invited friends to read in in real time.

My diary shows 17 hours logged, spread over a few weeks. But those 17 hours? They are not the full picture. They are just the visible part of a much longer story, the part where I sat in front of a keyboard. The invisible part was the months of thinking, discarding, distilling, connecting dots that only started to make sense later.

Then came the reviewing process.

If you’ve ever sent a paper out for peer review, you know that’s when the sausage grinder really starts. Reviewer 1 was constructive, nudging me to explain how my work went beyond prior critiques. Reviewer 2, let’s just say their feedback was less than kind. They questioned the contribution, the clarity, even the usefulness of the whole thing. Two striking remarks stayed with me: Reviewer 2 wrote

“There is no contribution of the paper, the discussion is low-level quality and the conclusions do not seem to have practical or theoretical usefulness.” Another line hit just as hard: “The paper proposes a discussion on an extensive subject, but fails only analyzing shortly specific topics of that subject.”

For an hour or two, I was angry. Then I got to work. I rewrote the introduction, reframed the entire contribution, tightened every paragraph. I added design implications I hadn’t planned to include. I went back through decades of references to make sure every citation pulled its weight. Working on the original draft and the submission took 17 hours (MDPI.com’s author process is smooth as silk). Answering the reviewers took 14 more hours.

And then, something surprising happened. I realized that Reviewer 2’s harshness had actually made the paper stronger. Their criticism forced me to look hard at the messy, half-ground meat of my argument and turn it into something people could actually digest. Reviewer 2, by the way, was only slightly impressed and still recommended not to publish the paper, even after the improvements.

Now there’s a screenshot sitting on my desktop, full of tracked changes. It’s the clearest picture of what vibe research really is: an idea sparked, shaped by months of invisible labor, tested in the fire of critique, and finally molded into something you can hold in your hands.

So yes, my diary says 17 hours. But the real work started months before that, and it never really stops.

That’s how sausages are made.

What’s your own behind-the-scenes story? 👇 I’d love to hear it.