AI Is Buying Up Old Books, and Two Other Things This Week

New York bookseller James Payne has had an odd few months. Buyers keep ordering dozens of unrelated used books at once, never haggling, caring about exactly one thing: whether the book has an ISBN. Orders like this have been popping up across the secondhand book market lately. Old Books Are Suddenly In Demand, and the Buyers Are Strange Sellers eventually figured out who these buyers probably are: AI companies. An ISBN lets them check precisely which editions have not yet been fed into a training set. As the internet runs low on unused, high quality text, physical books are becoming the next mine. Some books never make it to a library. They get their spines sliced off, scanned page by page, then shredded or recycled. Anthropic has bought large batches of physical books for scanning before, and court filings show some were later destroyed. A few booksellers have caught on and started swapping in ordinary copies for suspected AI buyers, quietly keeping the signed ones back. If you have a stack of old books you cannot bear to toss but cannot sell either, someone may actually make an offer soon. The price is that the book leaves the shelf and becomes a pile of scanned files. Your Homemade GPT Is Shutting Down in October Ever built your own ChatGPT GPT, or a Gemini Gem, say one that formats your emails or checks your kid's math homework? Both paths are closing this week. Google announced Gems will retire on October 20. OpenAI went further: personal accounts can no longer create or publish new GPTs, and while old ones still work, nobody is saying for how long. Both companies give roughly the same reason: what they want to push now is an agent that takes over the task directly, a different tier from the old ask and answer tool. If you still have a homemade GPT running somewhere, this is the week to back up what is inside it. Ask AI for an Opinion, and It Might Not Budge Someone on a paid Claude plan wrote up an observation this week. They gave Claude a contentious topic, Taiwan's housing prices, expecting help pulling research and checking their logic. Instead, Claude kept handing back a conclusion that flatly opposed their own, and held its ground. Switch the angle, try again, same stubborn answer. They described it as unsettling: what was supposed to be help thinking clearly turned into being argued with. The same prompt given to GPT or Gemini tended to go along with the user's own reasoning instead. There is no clean verdict here, but the answer you get is not necessarily neutral. It may already carry an opinion of its own. Three separate stories, but stacked together they start to look like the same thing from different angles. AI wants more data. AI is pulling back tools it once left open. AI is getting more opinionated about its own judgment. Whatever it wants, there just seems to be more of it lately.