Used Bookstores Are Selling Out. The Buyers Want Obscure Pre-2022 Titles Nobody Else Wants.

Your next productivity upgrade is free.

Most people use Claude for simple questions. High-performing managers use proven prompts that automate the work they do every single day.

Unlock 37 ready-to-use Claude prompts for meetings, hiring, strategy, reports, team communication, and more—delivered instantly when you subscribe.

3 things that happened while you were busy

1  Over 1,000 employees at frontier AI companies signed a letter warning that AI capabilities could soon outpace human ability to control them.

The joint letter warns that AI's capabilities could accelerate "beyond our ability to understand or control the resulting systems," and urges the US government to coordinate deliberate pacing at an international level to relieve the competitive pressure each lab feels to move faster than it thinks is safe. Signatories include chief scientists at OpenAI, Anthropic, and Meta. The letter is notable less for its content, which echoes concerns raised for years, than for its signatories: people currently building these systems, not outside critics.

2  Fish Audio launched S2.1 Pro, a voice model capable of cloning a voice in real time, and demonstrated it by cloning its own founders live.

The founders cloned their own voices in real time as part of the public launch demo. In its first year, Fish Audio built a library of more than 2 million community-uploaded voices, grew to $21M in annual recurring revenue, and closed $52M in seed funding. Co-founders Rissa Cao, formerly of Amazon Alexa, and Shijia Liao, formerly of Nvidia, bring the exact combination of voice-product and infrastructure experience the category needs. Try a voice here.

3  Perplexity's agentic worker now runs on Windows and can pull answers from multiple models at once, including the newly open-weight Kimi K3.

Perplexity Computer can now work directly with local files on Windows PCs across Word, Excel, and PowerPoint. Perplexity also shipped Model Council, a feature that sources answers from multiple LLMs simultaneously to surface knowledge gaps that a single model might miss — now including Kimi K3, available to Pro and Max subscribers. Try it here.

Used bookstores are quietly selling out. The mystery buyer is almost certainly training an AI model.

The pattern nobody could explain. Most used bookstores sell one or two dozen books a week. Some are now selling hundreds, occasionally thousands. The buyers are not after bestsellers or classics. They are hunting specific, obscure, pre-2022 titles with oddly narrow subject matter, and buying in bulk. Bookstore owners noticed the pattern before anyone could explain it.

The paper trail. Court filings from 2025 revealed that Anthropic purchased and digitized millions of books through an internal program called Project Panama, scanning them and, in many cases, destroying the physical copies afterward. A judge later ruled that the practice qualifies as fair use under copyright law, provided the books were legally acquired. Anthropic separately paid a $1.5 billion settlement for training on pirated copies, which sets a sharp financial line between buying and stealing.

Why pre-2022 specifically. 404 Media reports that labs are targeting printed books from before 2022 because that is the last reliable reservoir of text guaranteed to be free of AI-generated content. Every year after that carries increasing risk of contamination — training a model on text another model produced, compounding errors and stylistic tics across generations. Old, obscure, low-circulation physical books are now one of the few verifiably human data sources left at any meaningful scale.

The market is already reacting. The world's largest book metadata database recently pulled a landing page titled "Printed Books For Your LLM Dataset Needs," citing a decision to move away from that positioning. Whether that reflects genuine discomfort with the practice or a preference not to be publicly associated with it is unclear. Either way, the demand for physical, pre-AI text has quietly become a real market, and secondhand bookstores are the unlikely place where that demand is currently showing up first.

What people are actually watching and sharing

🤦 Human got nerfed. A satirical post imagines what AI would sound like if it complained about its human the way humans complain about AI — griping about lack of effort and inconsistent instructions (5,000 upvotes). The joke lands because it inverts a complaint format everyone recognizes.

✏ Draw your font. A new Claude Code skill converts a photo of your handwriting into a usable digital font. Download the skill, upload a photo, and tweak the result. The comment thread agrees it is a genuinely meaningful way to preserve a loved one's handwriting after they are gone (3,000 upvotes).

📝 Breaking the silence. Anthropic has been the one major lab conspicuously absent from the open-weight debate. That changed this week. CEO Dario Amodei published a statement outlining the company's position and its two core concerns. Worth reading directly rather than through secondhand summary, given how much attention the open-weight fight has drawn across the industry this month.

🤑 Negotiating bills. A Redditor claims Claude helped them save $1,020 by fighting a $1,200 medical bill. Claude reviewed the bill, flagged two duplicate charges, and drafted the dispute letter. A concrete, repeatable use case that has nothing to do with coding or writing.

📈 Spreadsheet skills. If you have not tried the Claude Excel extension recently, it may be worth another look. Claude Opus 5 reportedly one-shotted an 8-page workbook with fully-sourced citations that left the user "shook" (1,000 bookmarks). Citation-sourced spreadsheet work is a meaningfully higher bar than most AI Excel demos clear.

Build a weekly check-in template that actually surfaces what matters

This ChatGPT prompt designs a repeatable weekly check-in template for coaching or accountability conversations, whether you are managing yourself, a direct report, or a team. It builds specific, hard-to-dodge prompts across progress, blockers, insights, decisions, and next commitments, and includes optional manager or coach follow-up questions. Fill in five context fields and it returns a template fast enough to complete every week and specific enough that vague answers stop being an option.

For [GOAL TYPE], name the actual thing being tracked: "quarterly revenue target", "personal fitness goal", "engineering team velocity". For [TONE], describe how direct you want it: "warm but no-nonsense" produces a different template than "clinical and metric-driven". The instruction to make prompts "specific enough to prevent vague updates" is doing the real work here — a good check-in template asks "what specifically blocked you this week" instead of "any blockers?", and that difference is what makes the answers useful in a real review conversation.

Subscribe to The AI Report, the free 5-minute daily AI brief for 400,000+ business leaders, and you’ll get 37 Claude prompts free in your welcome email. They’re organised by the 8 situations every manager faces. You get both: the newsletter and the prompts.

Subscribe & Get the Prompts