Tuhin Chakrabarty

@tuhinchakr.bsky.social

Assistant Prof @sbucompsc @stonybrooku. Researcher → @SFResearch Ph.D. → @ColumbiaCompSci Human Centered AI / Future of Work / AI & Creativity

When a model writes, where do its words come from? Are they new, or do they match exactly with language it saw in training? An AI-writing detector can't tell you. @tuhinchakr.bsky.social's group at Stony Brook has been dissecting AI-generated prose with our infini-gram engine. 🧵

Bild

Very grateful to @ai2.bsky.social for tools like OLmO Trace and Infingram for helping us to draw the boundary between whats human written and whats ripped off from the training data 😈

Ai2@ai2.bsky.social · 5d ago

When a model writes, where do its words come from? Are they new, or do they match exactly with language it saw in training? An AI-writing detector can't tell you. @tuhinchakr.bsky.social's group at Stony Brook has been dissecting AI-generated prose with our infini-gram engine. 🧵

If you're going to cite that NBER report from OpenAI about "how people use AI," you've got to at least caveat it. We have no way to directly verify most of the report, and we do have at least one good reason, via indirect evidence and quoted below, to not trust it. www.nber.org/papers/w34255

Maria Antoniak@mariaa.bsky.social · 7mo ago

Not the point of the OP, but I was curious and took a closer look at this 2025 piece from OpenAI. Much could be said, but it fails my go-to test for all of these pieces about "how people actually use AI": the tokens "sex" "erotic" and "NSFW" do not appear in the paper.

Trying to push back here as the first author of the paper. Please do not politicize and make it about race. Pangram doesn’t flag authors based on race. Put Octavia Butler, Roxane Gay, chimamanda ngozi adichie and it will not flag them as AI. Don’t blame me for your lack of understanding of AI.

Steve Ammidown@stegan.bsky.social · last wk.

That article from The Atlantic plays into so many garbage narratives at once -Black woman must be cheating if her book is popular -Romance novels are formulaic -AI is to be believed always, humans are lying Congratulations to that publication and article author on being terrible. Good job.

Very telling research on the rise of self-published books with substantial AI text - which on their own earn less than 'regular' books, but together seem to earn their 'authors' substantial and are thus crowding out the market for others:

Tuhin Chakrabarty@tuhinchakr.bsky.social · last wk.

🚨 New paper on AI & Copyright Using @pangram.com full-text AI detection, we study self-published genre fiction on Amazon, 2023–26. Amazon is flooded w/ books with substantial AI text. They earn real 💰 & crowd out non-AI books. w/ @dhillonp.bsky.social @xinyuel.bsky.social & Jane Ginsburg (1/n) 🧵

Including evidence that AI Slop takes revenue from real authors. The key long term effect --- human authors, already facing long odds, leaving the scene entirely --- will be devastating for literature (as it will be for music and other arts).

Tuhin Chakrabarty@tuhinchakr.bsky.social · last wk.

🚨 New paper on AI & Copyright Using @pangram.com full-text AI detection, we study self-published genre fiction on Amazon, 2023–26. Amazon is flooded w/ books with substantial AI text. They earn real 💰 & crowd out non-AI books. w/ @dhillonp.bsky.social @xinyuel.bsky.social & Jane Ginsburg (1/n) 🧵

Incredibly important new research published today shows the huge impact AI books are having on real writers’ ability to make a living. Looking at self-published books on Amazon, they found: - 20% of self-published books include substantial AI text - ... 🧵 1/3

Tuhin Chakrabarty@tuhinchakr.bsky.social · last wk.

🚨 New paper on AI & Copyright Using @pangram.com full-text AI detection, we study self-published genre fiction on Amazon, 2023–26. Amazon is flooded w/ books with substantial AI text. They earn real 💰 & crowd out non-AI books. w/ @dhillonp.bsky.social @xinyuel.bsky.social & Jane Ginsburg (1/n) 🧵

This is a very interesting thread. I’m surprised how well rare phrase use distinguishes AI/human text.

Tuhin Chakrabarty@tuhinchakr.bsky.social · last wk.

🚨 New paper on AI & Copyright Using @pangram.com full-text AI detection, we study self-published genre fiction on Amazon, 2023–26. Amazon is flooded w/ books with substantial AI text. They earn real 💰 & crowd out non-AI books. w/ @dhillonp.bsky.social @xinyuel.bsky.social & Jane Ginsburg (1/n) 🧵

The real contribution of AI— Truth & poetry die in irrelevance. Very convenient for those who hate truth & poetry. Authors vie for sustenance in an environment in which noise overwhelms signal. Derivative mediocrity becomes the norm. This is the path to our slow death through suffocation. #resist

Tuhin Chakrabarty@tuhinchakr.bsky.social · last wk.

Here's the imbalance at the heart of the paper. Over 3 years, the books catalog grew 38x cumulatively. The book's revenue? Only 9x quarterly. Way more books fighting over a pie that barely grew which means the average book earns less than it used to. (4/n)

A must read thread for anyone thinking about the socio-cultural and economic effects of generative AI. Huge thanks to Tuhin & his fellow researchers for their amazing work here. And as Tuhin notes, there are significant legal implications given the clear market dilution resulting from AI content.

Tuhin Chakrabarty@tuhinchakr.bsky.social · last wk.

🚨 New paper on AI & Copyright Using @pangram.com full-text AI detection, we study self-published genre fiction on Amazon, 2023–26. Amazon is flooded w/ books with substantial AI text. They earn real 💰 & crowd out non-AI books. w/ @dhillonp.bsky.social @xinyuel.bsky.social & Jane Ginsburg (1/n) 🧵

Paper thread behind @willoremus.com ‘s terrific Atlantic reporting 👇 For those interested

Tuhin Chakrabarty@tuhinchakr.bsky.social · last wk.

🚨 New paper on AI & Copyright Using @pangram.com full-text AI detection, we study self-published genre fiction on Amazon, 2023–26. Amazon is flooded w/ books with substantial AI text. They earn real 💰 & crowd out non-AI books. w/ @dhillonp.bsky.social @xinyuel.bsky.social & Jane Ginsburg (1/n) 🧵

Really enjoyed speaking to @willoremus.com about new work on market dilution caused by books with substantial AI text in them. My favorite part of this experience was to find evidence in the form of hyperspecific rare expressions from a given AI book under scrutiny that also occurs in other books.

Will Oremus@willoremus.com · last wk.

AI-assisted fiction is already here, tons of people are reading it, very few of them realize it, and no one wants to talk about it. Gift link to my latest: www.theatlantic.com/technology/2...