Picture this you walk into a nondescript warehouse.

@atlasberry008

34.2K followers · usually 5.1K views a post

Views

18.4K

Likes

845

Comments

177

Selling · SaaS › AI Tool

Frontier

https://frontier.m1c.vc/ ↗

Why it worked

The video taps into a controversial and ethically charged topic (destruction of books for AI training) which generates strong reactions and discussion. It also connects this to the broader AI race and the value of human-created data, making it relevant to current tech trends.

Summary

This video discusses Anthropic's "Project Panama," an initiative to scan and destroy millions of books to train AI models. It highlights the legal and ethical implications, contrasting Anthropic's destructive approach with Google's method of scanning and returning books, and touches upon the increasing demand for verified human text data in AI development.

Structure

1Introduction to Project Panama and the scale of book destruction
2Explanation of the process: bulk buying, cutting, scanning, pulping
3Comparison with Google's book scanning methods
4The legal justification (fair use) and the value of pre-2022 books
5The AI race and the demand for human-generated text
6Call to action to join the Frontier community

Original caption

"Project Panama is our effort to destructively scan all the books in the world." That's Anthropic's internal codename, revealed in unsealed court filings. Here's the constraint behind it: there's a finite amount of text humans wrote before 2022, and that number never goes up. Everything published since is contaminated with AI-generated writing, and models trained on that degrade. So printed books became premium inventory — vendor proposals discussed up to 2 million of them in six months. Compute keeps getting cheaper. Verified human data doesn't. If you like this kind of thing and want to stay close — join our community Frontier. Link in bio. anthropic, project panama, ai training data, claude, copyright, fair use, destructive scanning, data scarcity, llm training, venture capital #venturecapital #startup #founder #ai #copyright

More from @atlasberry008

Their other posts in the index, biggest breakout first.

On-screen text

Anthropic's "Project Panama": Shredding Books to Train AI PICTURE THIS NONDESCRIPT WAREHOUSE ARE SCANNING AND PAGES TURNING YOU SEE MILLIONS ON THE SHELVES STACKED ON THE FLOOR WAITING FOR THEIR DESTRUCTION IT FEELS LIKE YOU'RE WALKING INTO FAHRENHEIT 451 EXCEPT THIS IS REAL THIS JUST CAME OUT IN AN UNSEALED COURT FILING HAD TO HAND IN THE DISCOVERY AND THE JUDGE TO MAKE THEM PUBLIC AND HERE'S THE LINE INTERNAL PLANNING DOCUMENT 'Project Panama is our effort to destructively scan all the books in the world.' PROJECT PANAMA IS OUR EFFORT TO DESTRUCTIVELY SCAN ALL THE BOOKS IN THE WORLD SAME DOCUMENT THE EMPLOYEES LIKE SSSHHH Inside Project Panama, Anthropic's Secret Effort To Scan and Shred the World's Books Anthropic spent tens of millions of dollars secretly buying, scanning and destroying millions of physical books to train its Claude chatbot, according to more than 4,000 pages of unsealed court documents reviewed by the Washington Post. HERE'S THEY BULK BUY PHYSICAL BOOKS CUT THE SPINES OFF WITH HYDRAULIC CUTTERS LOOSE PAGES PRODUCTION SCANNERS AND THEN PULP WHATEVER IS LEFT THEY EVEN HIRED In February 2024, Anthropic hired Tom Turvey, the former head of partnerships for Google's book-scanning project, to develop a lawful route to a much larger research library. OF PARTNERSHIPS FOR GOOGLE BOOKS TO RUN THIS THING VENDOR PROPOSALS A vendor proposal referenced an ability to convert between 500,000 and two million books over six months WE'RE TALKING ABOUT 2 MILLION BOOKS IN LIKE SIX MONTHS AFTER 2022 IS CONTAMINATED WITH ALL THIS GENERATED THE MODELS TRAIN ON THAT TO DEGRADE SO PRE 2022 That is why books published before 2022 have suddenly become premium commodities in the AI industry. PRINTED BOOKS BECOME PREMIUM INVENTORY NOT BECAUSE BUT BECAUSE HUMANS DEFINITELY WROTE THEM From Piracy to Purchase, the $1.5 Billion Pivot ANTHROPIC ANTHROPIC JUST PAID A $1.5 BILLION SETTLEMENT DIGITAL BOOKS SO NOW THEY WENT LEGIT THE DESTRUCTION IS LEGAL ONE COPY GOES IN AND ONLY ONE COPY COMES OUT IF YOU KEEP THE BOOK AND THE SCANNED THEN YOU HAVE TWO COPIES AND THE FEDERAL JUDGE CALLS IT FAIR USE Publishers sue GOOGLE over copyrighted books allegedly used to train Gemini ACTUALLY DID WITH CAMERAS AND THEN GAVE THEIR BOOKS FROM LIBRARIES Google ANTHROPIC WAS OWNING THEIRS AI companies are now destroying warehouses full of rare books I THINK THIS IS WRONG FEEDING LIKE RARE EDITIONS HYDRAULIC CUTTER THIS SAYS A FRONTIER LAB IS RUNNING WAREHOUSES HYDRAULIC CUTTERS AND PULPING MILLIONS OF BOOKS THAT JUST SHOWS YOU THE DEMAND IS FOR HUMAN TEXT AND I THINK THIS IS CAUTIONARY TALE Cautionary tale AS WE START SETTING REGULATION AND RULES AI Firms Are Quietly Buying and Destroying Millions of Printed Books to Train Their Models • Your Property: Destroying a book you legally purchased and own is typically treated as destruction of your own private property. NO LAW INTENDED THIS CENTURY JUST MADE THIS THE CHEAPEST PATH A RATIONAL COMPANY BUT FOR FOUNDERS THIS IS YOUR SIGNAL COMPUTE START TO BUT VERIFIED HUMAN DATA AND AGAIN THIS SORT OF THING Frontier Signal from the edge. A private network for frontier builders and backers. JOIN OUR COMMUNITY FRONTIER LINK IN BIO

Transcript

Picture this you walk into a nondescript warehouse. And all you hear are scanning machines and pages turning. You see millions of books on the shelf, others stacked on the floor waiting for their ultimate destruction. It feels like you're walking into Fahrenheit 451 except this is real. This just came out in an unseen court filing. Antic had to hand these over in the discovery and the judge decided to make them public. And here's the line from thetic internal planning document. Project Panama is our effort to structurally scan all the books in the world. And this same document tells all the employees like keep this quiet. Here's what they did. They bulk-bodied physical books, cut the spines off with hydrochloric cutters. Run all those loose pages through production scanners and then pulp whatever's left. They even hired the former head of partnerships for Google books to run this thing. Some of the vendor proposals were talking about converting 2 million books in like six months. The internet after 2022 is contaminated with all this AI generated writing. And so if the models train on that then they start to degrade over time. So pre-2022 printed books become premium inventory. Not because they're rare, but because humans definitely wrote them. And Trapi just paid a $1.5 billion settlement over pirating digital books. So now they went legit. And actually the destruction is what made them legal. It's because one copy goes in and only one copy comes out. If you keep the book and the scan, then you have two copies. But when you shred it, then the federal judge calls it fair use. Google books actually did the same thing with cameras and then gave the books back. But Google was boring their books from libraries. And was owning their. Now look, I think this is wrong. No one should be feeding like rare additions of books into a hygienic cutter. But look what this says about the race. A frontier lab is running warehouses, hygienic cutters and pulping millions of books. I mean that just shows you how insaneble the demand is for human text. And I think this is also a cauchy tale as we start to think about setting regulation and rules for AI. No law intended this, century old doctrine just made this the cheapest path. And so a rational company took it. But for founders, this is your signal. Compute is going to slowly start to get cheaper. But verified human data is not. And again, if you like this sort of thing, I want to stay close, join our community frontier, Lincoln Bio.