AI Read Aristotle That’s Why It Writes Better Than You

by | Aug 25, 2026 | Author Update | 0 comments

Print Friendly, PDF & Email

For Week Ending August 21, 2026

Amazon is buying rare books, cutting off their spines, and destroying the copies to train its AI, and a hidden tracker proved it. A new writing tool now scores 100% human on every AI detector and says that’s just a side effect of better prose. Hachette cancelled one author’s book for AI use while another’s keeps selling. Your Anthropic settlement check is delayed. Clean comedy just shattered the all-time ticket record.

Publishing News

Deft Launches a Writing Model That Scores 100 Percent Human on Pengram

According to an August 18, 2026 post by @rosmine on X, a new AI lab called Deft is in public beta. The handle says the lab is cofounded with @jmrphy and exists “for better writing.” The same post says the Deft model wrote its own launch announcement, that Pangram currently scores 86 percent of user queries as fully human, and that fooling AI detectors is “just a side effect.”

The live site at deftwriting.com leads with “Deft AI Writing That Doesn’t Sound Like AI” and “Authentic Human Writing at Scale.” The thesis on that page is that writing well is hard, so professionals should be able to turn what they know into polished prose without mastering the craft. Deft says the best future workflows will use AI and the human mind together without hallucinations, without twisting the author’s intent, and without flattening the author’s voice. It also offers custom models trained to match a company’s style.

The Research Claim

A May 2026 paper on the same site, Fixing LLM Writing with Distribution Fine-Tuning, is signed Rosmine. It argues ordinary supervised fine-tuning leaves models formulaic. Distribution Fine-Tuning is the proposed fix. The paper reports MMD improved by 49 percent and Judge Model Quality by 63 percent against an SFT baseline. It also reports creativity scores up 164 percent, coherence up 28 percent, clarity up 16 percent, and meaningful detail up 146 percent. It says the method avoids overused slop signs such as too many em dashes and the “it’s not X, it’s Y” pattern. A sample of 100 model outputs, the paper says, scored 100 percent human written on Pangram.

The X post says Deft currently works better for analysis, essays, creative writing, and rewrites, and less well for marketing copy and news articles. The paper’s limitations section says the demo was trained on a FineWeb subset and should be stronger on blogs and news, and weaker on creative writing until that data is added. Those are Deft’s own sentences, and they do not agree.

The X follow-up says users can try the product at deftwriting.com for 3 free generations before an account, then 2 free generations per day, with a paid plan after that. The paper also admits abuse risk for spam and slop accounts, and says the demo injects random fruit or animals into copied text so users cannot paste blindly.

Why it Matters

The cultural fight is no longer whether a reader can “tell.” A lab is shipping writing whose proof of quality is a detector score, and it says that score is a side effect of better prose. Authors who sell “fully human” as a brand now share a market with a product that treats undetectable style as a feature of the writing, not a cheat code. If you use Deft, keep the outline, the facts, and the judgment in your hands. If you refuse it, say so to your readers before a Pangram score says it for you. The author move this week is to write down your own line: what a tool may polish, and what still has to come from you.

Sources:

Commentary

Thomas: I’ve been talking with one of the developers behind Deft for months. He gave me a private preview a while back.

My best understanding of how Pangram works is that it checks whether each token in a document is the most predictable next token, then scores how much chaos or predictability that produces. I’ve said from the beginning that these detectors aren’t particularly accurate. Pangram is more likely to flag something as AI that isn’t, but it’s also more likely to miss AI that is. This is good evidence of that. We now have a brand-new model that passes 100% human, while an author who published a book 20 years ago can score 0% human on the same test.

Other labs are adding watermarks to their output. Anthropic and Google both have versions of this, partly because ingesting AI-generated content in training causes model collapse. But tools like Deft likely wash away that watermark in the process of humanizing text, since you could run any AI output through Deft to make it pass Pangram.

Relying on automated detectors to burn people at the stake for using AI doesn’t work when you’re burning people who aren’t guilty while missing the ones who are.

If you check out Deft at deftwriting.com, tag me on AuthorMedia.social. I’d like to see what you think of the output.

BookBub Intros New Newsletter “New in Kindle Unlimited”

Bookbub launched its new “New In Kindle Unlimited” subscription on June 25, 2026. If your book is in its first 90 days on KU, now you can enroll in this program and see new readers come to your series.

Jonathan: Pricing varies by category. Thrillers run $350, crime fiction $325, historical fiction $300, fantasy $225, contemporary romance $200, and other romance subgenres $175. Nonfiction has the lowest KU deal pricing of the categories. Check the full pricing on BookBub’s site.

Thomas: This is a big deal because BookBub historically hasn’t had much to offer brand-new authors or brand-new books. It works well for a book that was popular five years ago with a lot of reviews and you want to introduce it to a new audience, but if you don’t have reviews, BookBub often rejects you repeatedly. With this program, your review count doesn’t factor in, since the book is brand new anyway.

Hachette Faces Its Own AI Consistency Test

According to Mike Israetel’s May 21, 2026 X post, the Renaissance Periodization cofounder told the public he used AI to smooth some of the prose in The Aesthetic Revolution. He stated that a team of publishing company editors further smoothed that prose and that those editors knew about the AI work. He posted the admission weeks before Grand Central Publishing, a Hachette imprint, released the book on June 23, 2026. He claimed every idea, every structure, and every core claim remained his alone.

The Earlier Cancelation

According to the New York Times report of March 19, 2026, Hachette cancelled the U.S. edition of Mia Ballard’s Shy Girl and pulled the U.K. edition from sale. Detection tools and online readers had flagged AI patterns across the text. Hachette declared its commitment to original creative expression and storytelling. Ballard denied writing with AI and blamed an editor. The novel had sold roughly 1,800 print copies in the U.K. The deal ended.

Israetel’s book remains listed and available. Solomon Nelson’s August 11 analysis presented Gemini watermarks on images credited to RP Strength and argued the prose carried dense AI markers. Hachette has issued no public statement. The admission sits in full public view. The book continues to sell. Israetel reaches roughly 3.9 million subscribers on the Renaissance Periodization YouTube channel, 1.2 million Instagram followers, and 387,000 TikTok followers. He runs apps, coaching systems, and education products that generate revenue outside any single publisher contract. That independent base changes the calculation.

Live Industry Test

Publishers enforce AI rules with greatest force against debut and midlist writers who lack large audiences and independent income. An author who posts the disclosure in advance and still reaches publication creates a different pressure. Consistency becomes measurable. Competing authors and readers now watch whether the same house applies the same standard to a high-visibility nonfiction title that announced its own AI assistance before release.

For Working Writers

Document every AI tool used in drafting and revision. Demand written clarification of your publisher’s current disclosure policy and its treatment of disclosed assistance. Track how major houses respond when the author controls the first public statement. Independent platforms and revenue streams alter leverage. Authors without those assets absorb the full cost of enforcement. Authors who possess them force the policy into the open. The numbers sit in plain sight. The admission sits in plain sight. The cancelled debut sits in plain sight. Hachette now owns the test case.

Commentary

Thomas: This is potentially a class divide over AI use, and it’s the test case. Big publishers cancel small, unknown authors for using AI. Will they cancel a big one?

Israetel has millions of subscribers and millions of dollars. If Hachette cancels him, he’ll sue and have the resources to do it, broadcast the cancellation across his platforms, publish the book independently since his rights would revert, and still make money. Hachette gets the black eye either way. That raises an uncomfortable question about Shy Girl: was that author canceled just because she was obscure enough to get away with it?

These publishers don’t seem to have real principles about AI use. They go with whatever generates less online drama, and they don’t want to antagonize an author who could turn a cancellation into a cause célèbre. He would sell more without them than with them, they would still lose the money from his advance either way, and defaming him over false AI accusations could cost them millions.

The other difference is that he isn’t apologizing. The authors who denied using AI and the ones who apologized both still got canceled. He’s saying, in effect, that he used AI, he’s a public figure who knows what he’s talking about, and there’s nothing to be done about it. That’s a different, more powerful position to argue from when you have the following and the money to back it up.

Hachette also likely already got served over the Shy Girl cancellation, and their counsel is probably weighing that. I’m not a lawyer, but defamation cases are expensive even when you win, and having already been burned once on a smaller case, they know the next one would be worse.

Sources:

Lawyers Appeal to Take More of Authors’ $1.5 Billion

Edelson PC and Oppenheim + Zebrak filed a notice of appeal on August 18 seeking a larger share of the $101,561,111 fee award the court reserved solely for class counsel, according to Bloomberg Law. The firms want money the July 20 order kept out of their hands, money that would otherwise stay in the fund for authors and publishers. Class counsel told the Textbook & Academic Authors Association that any appeal pauses the claims process, so authors now face both a smaller net recovery and first payments delayed until at least the end of 2026.

Commentary

Thomas: We’ve been covering this story as it unfolds. The lawyers got smacked down. The judge crossed out the number they wanted and wrote in a much smaller one, and it wasn’t a small markdown, we’re talking tens of millions of dollars in losses for them. So it isn’t surprising they’re appealing. They would rather take a slower shot at the bigger number.

There’s nothing you can do about the delay right now. If you were counting on your Anthropic settlement check landing next month, don’t spend against it yet. I know authors with dozens of books in the settlement expecting tens of thousands of dollars. Even at $1,500 per book for traditionally published authors, 40 books in the settlement is $60,000 before taxes. That’s real money, but don’t expect it quickly. This appeal might only delay things by days, or it could delay them much longer. Don’t spend the money until it’s actually in your pocket.

Sources:

Kindle Romance Share Drops From 60% to 48%

According to Jane Friedman in The Bottom Line on August 19, 2026, Alex Newton of K-lytics tracked first-half 2026 Kindle sales. Romance still leads the store. Its share of the 100 most-sold US Kindle books fell from close to 60 percent last year to 48 percent this year, counting romantasy. Mystery, thriller, and suspense took 31 percent. Science fiction and fantasy took 10 percent. Those shares describe the top of the store, not romance unit sales in full. Over a 7-year view Newton still calls romance stable and growing.

Amazon Rank Changes

In May 2025 Amazon stopped updating Kindle rank every hour. Rank now updates once per day. Changes can take 1 to 3 days to show. Newton says Amazon still moves faster on high-ranking titles when it chooses. The catalog he cites is at least 14 million Kindle titles, up from about 2.5 million 12 years ago.

Friedman walks through two authors who each move 225 sales and borrows in 72 hours. Author A sells a few copies every hour. Author B sits near zero, then blasts a newsletter and spikes to a couple dozen sales an hour. Hourly rank showed that spike. Daily rank leaves Author A in place and flattens Author B. Newton calls the new system a shock absorber. Authors who built visibility on big email blasts lose the most. Promos can still move rank. They do not move it the way they did. Newton says the change gives more titles a chance to appear, and that is one reason romance’s share at the top loosened.

US romance subgenre order is close to last year. Romantasy is still first. Hockey romance climbed from third to second after Heated Rivalry adaptations. Rom-com moved from fourth to third. Mafia and dark romance slid from second to fourth. Western and billionaire stay in the mix.

The UK store does not copy the US. Mystery, thriller, and suspense lead with nearly half of the top 100. Romance is 31 percent. Women’s fiction, including upmarket and book-club titles, is 12 percent. UK romance subgenres follow the US list, minus western.

Top 100

In the US, self-published authors hold 38 percent of the top 100. The Big Five hold 32 percent. In the UK, indies hold 16 percent. Newton’s read is that the Big Five put more Kindle marketing work into the UK store.

Market Demand Moving

In science fiction and fantasy, LitRPG is the breakout. 8 of the 100 most-sold titles so far in 2026 are Dungeon Crawler Carl books. Newton ties much of that series’s start to the audiobook narration. Dystopian and post-apocalyptic fiction rebounded after about 5 years of decline. Sci-fi adventure, galactic empire, and space opera show the same recovery. Hard sci-fi has not risen with them, even after Project Hail Mary and the film. Epic fantasy is up on a 6-year and an 18-month view. Newton’s caveat is the same as last year: more than 80 percent of that category is romantasy, not Tolkien-style epic. Christian fantasy also rebounded after a long decline.

Mystery, thriller, and suspense at the top barely moved. Newton says the US category still almost equals Freida McFadden and lookalike domestic thrillers. The UK is even more concentrated. Below that peak, police procedurals, techno thrillers, and cozy mystery rebounded. Domestic thrillers still sit at the top of Newton’s demand-versus-supply chart, with a catalog that has only just crossed 10,000 titles. He calls that a small pile after 20 years of Kindle.

AI Ads Explosion

Newton tracks Google search interest in Kindle Direct Publishing as a proxy for people who want to publish. He calls the current wave Gold Rush 3.0, after the Kindle launch and the pandemic run. Self-help and personal transformation has hit a ceiling at about 25 times its pre-generative-AI level. Contemporary romance supply is up by a factor of 3. Science fiction is up 3.5. New thrillers published per month on Kindle went from about 1,000 to more than 7,000. He ties that flood to social ads that promise a cozy-mystery empire from one keyword.

Why it Matters for Authors

Newton says the store feels like a crowded battlefield and counsel is to fight in a smart way. He expects the Kindle Unlimited payout pool to keep growing at 12 percent year on year. The pool started at $32 million in 2014. He projects $830 million in 2026 on performance so far this year. Authors who treat a 72-hour newsletter blast as a rank machine should expect less lift. Authors who sell a few copies every day keep more of their position. Writers hunting room can look at domestic thrillers and the rebounding dystopian and space-opera lists. Romance writers should treat 48 percent as a share of the peak, not a collapse in the genre.

Paid K-lytics genre reports and a full-data subscription are at k-lytics.com.

Commentary

Thomas: These numbers come from Alex Newton at K-lytics, who was interviewed by Jane Friedman. The real story here isn’t the algorithm change, it’s where the demand moved. If romance’s share of the top 100 dropped, what took its place?

The answer is LitRPG. Eight of the top 100 books right now are Dungeon Crawler Carl titles. LitRPG is going to become big enough that mainstream readers start to notice, and that will put the genre on the map for a lot of people who have no concept of it, because it isn’t derivative of any mainstream genre. Romantasy is a derivative genre, essentially romance with more magic. LitRPG emerged from video games and tabletop games by way of anime, not literature. It waves at fantasy tropes and borrows some fantasy and sci-fi vocabulary, but it isn’t really derivative of fantasy at all. It’s a video game in book form.

If romance makes readers feel like they’re being wooed, LitRPG makes readers feel like they’re playing a video game and leveling up. Young men in particular are so poorly served by novels, TV, and movies that many would rather watch someone else play a game than consume any fiction at all, and that’s a huge opportunity for authors willing to listen to what that audience actually wants.

Jonathan: I’m writing a LitRPG right now called Devil Dog, and I just released the latest chapter. Readers have told me they read it three times to understand a twist I dropped in there. It’s a fun genre to write and read, depending on who’s writing it. I think AI would do LitRPG badly, because if you lean too hard into the stats, it gets boring to read fast. Humans own this genre right now.

Thomas: It’s also almost exclusively written by men. I can’t think of a LitRPG title written by a woman, though I do know women who read the genre. The cozy, farming, and cultivation-progression side of the market skews female.

I don’t want to oversell this shift. Romance is still the single largest genre. It used to outsell everything else in the top 100 combined, and now everything else combined only beats it by two titles. This isn’t a collapse, it’s a rebalancing.

Sources:

Author Alerts

Patron Toolbox Improvements

Event Finder Upgrades

Every event on your list is in the future, with its date confirmed against the event’s own website rather than a listings site that has not been updated since last spring. If an event’s next date is not confirmed yet, it moves to a “Keep Looking” watch list instead of sitting in your main list pretending to be real. The tool no longer makes events up. That was the whole point of the rebuild. It also runs faster and covers more ground in the same session, so you get a fuller list without waiting around.

Thomas: The old version of this tool was like sending out an army of chipmunks to find acorns, and half of them came back with rock-shaped acorns or acorns that had already gone bad. It also does a much better job localizing to you now, including in the Mountain West, especially Utah, which used to be a real blind spot.

Book to Blurb

The output now opens with your story’s core theme in a sentence or two. You can tell in five seconds whether the tool understood your book. If it got the theme wrong, say so, and it rewrites the entire package around what your story is actually about. No more reading a full blurb package before discovering it missed the point.

Thomas: This is one of our most popular tools. You upload your book and it gives you a one-sentence version, a one-paragraph version, and a full back-cover copy version of why your book is appealing. It’s now better at teasing out theme specifically, so it works better for more literary fiction than it used to.

Reader Magnet Brainstormer

A magnet that wins a new reader is not the same as one that rewards someone who already loves your work. The tool now sorts its ideas by which job you are hiring them for. It also brainstorms audiobook magnets and holiday-timed magnets, and it only suggests holidays that exist.

Thomas: Say you wrote a book about a magical boarding school with four houses. A quiz asking which house you’d be in is a great fan magnet for your back matter, since it appeals to someone who already finished and loved your book, but it’s a terrible stranger magnet, since someone who hasn’t read the book doesn’t care. A management-style quiz for a nonfiction book on leadership works as a stranger magnet instead. The Reader Magnet Brainstormer now sorts its ideas by which job you’re hiring them for, fan, stranger, or audiobook magnets.

Chapterizer

Text from the Chapterizer no longer carries a hidden AI watermark. Everything else works the same, including the curly quotes and spelling corrections.

Thomas: I switched the Chapterizer off an Anthropic model specifically so it no longer carries a watermark.

You are now more likely to see previous chat history. Close the tab and come back, or switch from your laptop to your phone, and your past runs are still there. I still recommend copy and pasting everything you want to save to your computer.

AI Thomas is Now on AuthorMedia.social

You can now tag AI Thomas on AuthorMedia.social to get a response right there in AuthorMedia.social. You can also do this for others if you don’t know the answer.

Thomas: AI Thomas is trained on hundreds of episodes of Novel Marketing, Christian Publishing Show, and Author Update. It’s good at citing its sources and linking back to the original episodes. It takes about 15 minutes to respond, so it isn’t instant, and it’s free right now while I’m still catching up on email after paternity leave.

New Arsenal Episode Drops Tomorrow

After a lengthy hiatus, Author Arsenal will drop a new tutorial on using Elevenlabs to augment an audiobook narration. Check your feeds and subscribe to Author Arsenal.

Jonathan: It runs about three hours, and I’m chapterizing it so you can jump straight to whichever stage you want to learn, editing the narration, adding voice effects, developing the music, adding sound effects, and doing the final cleanup for posting.

Authors on Divi 4 Must Plan the Upgrade Now

According to Elegant Themes, Divi 4 receives security patches through. The official migration guide states Divi 4 will eventually reach end-of-life once Divi 5 reaches full stability, with no new features added and all future development reserved for Divi 5.

Authors who still run Divi 4 on their sites must test and complete the migration before the support window closes or risk running an unsupported theme that no longer receives security updates.

Commentary

Thomas: I just finished migrating AuthorMedia.com from Divi 4 to Divi 5 myself, and I should have done it a long time ago. Divi 5 is faster and better. The bigger revelation from the last couple of weeks is that you don’t need to do your own website updates anymore. Claude Code, GrokBot, and Cursor can all just update your website for you now. I tested GrokBot, which I’ve been using for a week and which has changed my life, and Claude Code, and both were able to migrate pages from Divi 4 to Divi 5 and fix things directly.

There’s an arbitrage opportunity here if you’re a webmaster. You can cut your labor by 80 or 90 percent and take on several times more clients, and make more money over the next year or two than ever before. Eventually people will learn to use these tools themselves and some of that work will go away, but right now is the best time to be a webmaster, because these tools can log into WordPress and Divi directly and fix things well.

Sources:

Massive Apple Security Issues: Update Your iPhone, iPad, and Mac Tonight

Apple released iOS 26.6.1 and iPadOS 26.6.1 on August 17, 2026, closing 29 security flaws, according to TechRepublic and Apple’s iOS advisory. The drop includes 21 WebKit bugs plus kernel, ImageIO, and a telephony flaw that could let an attacker intercept traffic on a privileged network. macOS Tahoe 26.6.2 carries 28 of those fixes. Authors who draft, upload, or read mail on Apple devices should install this week. Owners of iPhone XS, iPhone XR, or 7th-generation iPad should take iOS 18.7.10 / iPadOS 18.7.10.

Commentary

Thomas: With the recent breakthroughs in AI it is now computer vulnerability season. So make sure you are keeping everything up to date.

This isn’t just an Apple problem. New frontier-level AI models are finding massive security vulnerabilities across Windows, Linux, and WordPress too, and they’re getting patched quickly, so keep everything updated for the next six months or so. Windows is the one to watch, since Windows Update has been introducing new bugs while it fixes old ones; Windows Defender broke, then got patched in a way that broke it differently. My advice is to stay current on updates, and if you’re on Microsoft, consider switching to Mac or Linux.

Sources:

ChatGPT Almost Stopped Citing Reddit

Reddit’s share of ChatGPT Search citations experienced a 86.4% relative drop, according to Promptwatch. The share fell below 1% on August 14 and has stayed there; Promptwatch calls the size of the drop provisional and notes Google AI Overviews only slid about 11% over the same window.

Commentary

Thomas: This is good news for Open AI and humanity.

There are better sources of data than Reddit. Reddit was tempting early in AI development because it had a lot of niche data, but the models are finding better sources of niche data, like scanning old books. I’ve noticed Grok citing Reddit too often and telling it to cite more authoritative sources instead. I think the models are going to rely on books more.

Sources:

Audible Opens Bulk Audiobook Sales to Authors and Companies

According to Audible’s July 30, 2026 press release, Audible launched Audible for Business so authors, publishers, companies, and event organizers can buy audiobooks in bulk and distribute digital redemption codes. The program gives access to more than 1 million titles through Amazon Business with no shipping or inventory, while authors can purchase codes for their own work to hand out at book tours, launch events, and marketing campaigns. Authors gain a direct promotional tool for live events and corporate outreach, though royalty treatment on bulk purchases of their own titles remains unpublished and codes redeem only for U.S. recipients.

Commentary

Thomas: This is a big deal for anyone selling audiobooks in person, and I suspect BookFunnel is part of why it exists. Audible doesn’t want you handing out BookFunnel versions of your audiobook at events, they want you selling Audible versions. Now you can take cash in hand and hand someone a card with a code for the Audible version, all through official channels.

The bigger opportunity most authors don’t know about is bulk sales to corporations. I know authors making tens of thousands of dollars a year this way, selling hundreds or thousands of copies to a company that wants a copy for every employee, sometimes in a gift basket. Most authors think that sounds like a good idea until they realize it requires picking up the phone and finding the right person to call. That’s actual sales, not marketing, and because most of your competitors won’t do it, the authors willing to make those calls can make real money.

Sources:

Technology

Free Grammarly Replacement: LanguageTool Delivers Free Grammar Checks Authors Can Run Offline

According to a post by Nav Toor on X, Grammarly Pro runs $144 a year when billed annually or $30 a month when paid monthly, while a 10-person team spends $1,440 a year and a 50-person team spends $7,200. The same post points authors toward LanguageTool, an open-source grammar, style, and spelling checker that supports more than 30 languages and can run entirely on a writer’s own server.

What LanguageTool Does

LanguageTool flags grammar errors, style issues, and spelling mistakes that basic spell checkers miss. It handles six English variants, multiple German and Portuguese variants, Spanish, French, Dutch, and more than 20 additional languages maintained by native-speaker volunteers. The free web editor at languagetool.org accepts pasted text and returns underlined suggestions with explanations. Premium adds deeper style rules, Picky Mode for stricter checks, AI paraphrasing that rewrites sentences for formality or simplicity, and higher character limits that reach 100,000 or more per document.

Writers who prefer local control install the open-source core from the LanguageTool GitHub repository. The project carries roughly 14,800 stars and ships under the LGPL-2.1 license. Version 6.8 released on May 5, 2026. Community Docker images let users spin up a private HTTP server in minutes. Once the server runs, browser extensions or desktop clients point to the local address so every sentence stays on the author’s machine.

Cost

The open-source core and self-hosted server cost nothing. The free web version covers basic checks with character limits that currently sit around 10,000 per paste for non-Premium users. LanguageTool Premium pricing lists individual plans at €19.90 per month, €39.90 per quarter (€13.30 effective), €59.89 per year (€4.90 effective), or €99.90 for two years (€4.16 effective). Team plans for up to 20 users run €56.90 per user per year. Students and teachers can claim up to 20 percent off. These figures sit well below Grammarly’s annual and monthly rates.

Ease of Use

Paste any draft into the web editor and results appear in seconds. LibreOffice users install the official extension and gain real-time underlining inside their documents. Microsoft Word and Google Docs integrations follow the same pattern through add-ons. Desktop apps for Windows, macOS, and iOS require a single account and work offline once installed. Self-hosting takes one Docker command or a Homebrew install on macOS; the browser extension then switches its API endpoint to the local server and continues working without a paid subscription.

Authors who write in more than one language, protect unfinished manuscripts, or manage editorial teams gain immediate options. Try the free web editor on the next chapter, install the LibreOffice extension for offline work, or stand up a private Docker instance so the checker never leaves the local network. The open-source core remains free regardless of future commercial changes to the hosted service.

Commentary

Thomas: If you’re willing to host it locally, you can get the premium version on your own machine for free. This is where a lot of AI tools are headed, free and running locally on your computer. Apple’s next macOS release, a month or two out, is bringing AI-native updates, including a better on-device spell check. Basic spelling and grammar is the one AI use case almost everyone already accepts, since it’s been part of software for 20 years, and it’s about to get a lot better. You won’t need to pay for ProWritingAid if you’re a little bit tech-savvy, whether that means waiting for Apple’s built-in version or using an open-source option like LanguageTool now. I like supporting open-source software, in promotion and in payment. Free as in freedom doesn’t mean it has to be free as in beer. It’s good to pay for what you use.

Sources:

The Same Tool That Fixes Your Typos Just Saved a Company $6 Million

According to Asana’s engineering blog, Asana spent about $12,000 in model and infrastructure costs to finish a frontend test migration the company had priced at roughly $6 million. OpenAI’s August 18 case study puts the same total at about $12,000.

The Bill

Asana reports about $11,000 in model usage and $1,000 in infrastructure. The $6 million figure is Asana’s back-of-the-envelope math. Three Bay Area senior engineers for five years at $350,000 to $450,000 fully loaded lands between $5 million and $7 million. Engineers spent about 1.5 weeks on the work.

The Job

Asana pointed OpenAI Codex at Enzyme, an aging React testing library that had blocked frontend upgrades since 2022. Up to four agents ran day and night on extra-high reasoning. An engineer checked morning and evening and reviewed every pull request. The prompt ran five sentences.

An August 20 post by @scaling01 on X states “New Moore’s Law token volume is doubling every ~11 weeks.” The post attaches a chart that claims OpenRouter processes a greater than 4.5 quadrillion token annual run rate, with volume doubling roughly every 11 weeks and about 33 percent month-over-month growth for three years. The week of August 10, 2026 closed at 75 trillion tokens. OpenRouter has not published those figures on its own site. It shows a second company in the same market writing large checks for tokens.

Why it Matters for Authors

The same class of tool authors use to fix typos can save a corporation millions. A public company published the receipt: $12,000 in tokens against a $6 million staffing estimate. Authors who rent these models for drafts, covers, ads, and research sit in that market. A chapter is worth something to one writer. A stuck engineering backlog is worth millions to a software company. Those buyers write different checks. Price, rate limits, and the best models will follow the accounts that pay like Asana. Demand like that has no obvious ceiling.

Spend tokens on work only your book can repay. Park paid calls on chores a local tool or a cheaper model can finish. If you rent everything, pick the one job this week where a better model changes your income and put the spend there.

Commentary

Thomas: This is changing the world, and I’m feeling that change right now using Claude Code. If you’re only using AI through a chat window, you need to move to whatever coding harness matches your model, Claude Code for Claude, Codex for OpenAI, Cursor for Grok. The first time you ask it to do something, it walks you through logging in, and after that it just does the work.

What’s happening is that companies have long lists of projects they knew were valuable but too expensive to justify, and AI is collapsing that cost. A $6 million, five-year migration became an $11,000 project. Anyone who says AI is a bubble should ask how much Asana would have paid to save $6 million on a project it already wanted to do. Probably $50,000 or $100,000. It’s still a massive win at $11,000.

Jonathan: It’s not just the cost. The people who specialize in test migrations like this are some of the most difficult contractors in the world to work with, and Asana saved $5.9 million of a $6 million estimate without having to deal with them for long. That’s going to upend the whole industry.

Thomas: Other bubbles are popping, and AI is the needle popping them. Legacy vendors like Salesforce and IBM used to charge tens of millions for this kind of complex migration work, and companies are now doing it themselves for a fraction of the cost with people they already employ. It isn’t that jobs are disappearing, it’s that the amount of work that becomes possible is expanding dramatically. I’m optimistic. We need more data centers, more electricity, more robots. We can build a genuinely better future, we just have to build it.

Sources:

AI Is Now Helping Cure Cancer

According to the TechCrunch report on Vivodyne, the biotech startup Vivodyne opened what it calls the world’s largest human data center outside San Francisco last week. Andrei Georgescu, Vivodyne CEO and co-founder, states that AI models trained without human testing will cure cancer in mice. The company has raised under $80 million across two rounds led by Khosla Ventures.

Vivodyne’s HIVE Platform

Vivodyne’s modular robotic HIVE labs grow 20 kinds of human tissue. The systems dose the tissues and monitor results to generate causal biological data. Georgescu explains that current models train on static cell snapshots and learn cell state A next to cell state B without learning that state B follows from inflaming state A. Vivodyne reports its liver tissue reaches 94 percent predictive accuracy against human toxicity trials. Its airway tissue matches real human tissue behavior 96 percent of the time. Its bone marrow models show 100 percent concordance across tests of 20 chemotherapy drugs. The company states its current throughput doubles all animal trials running in the United States.

Anthropic CEO Statement

According to Dario Amodei’s X thread from August 15, the Anthropic CEO wrote that saying AI will cure cancer is more a cliché than it is inspiring and most people think it is deceptive. He added that the thing that will work is actually curing cancer. Amodei states that the most accurate criticism of AI companies including Anthropic is their failure to deliver on big promises to benefit the world. Anthropic is ramping efforts in biology and medicine with hopes of early glimmers in coming months and louder results once real accomplishments arrive. Amodei has not published a matching clinical result.

Pipeline Status

A handful of AI-designed drugs have entered human trials. One has reached Phase III. That candidate from Insilico Medicine targets idiopathic pulmonary fibrosis according to the Drug Target Review report on Insilico Medicine’s rentosertib. AlphaFold advanced protein structure prediction and produced no approved drug. Isomorphic Labs expects its first trials by the end of 2026. Pharmaceutical data show that 90 percent of candidates that succeed in animals fail to win human approval.

What Authors Must Do

Authors who research AI medical claims for nonfiction manuscripts or surface health data through AI tools must open trial registries and peer-reviewed papers before they publish. CEO timelines and social posts move faster than Phase results. Books that track the actual pipeline of AI-originated candidates, the failure rates, and the new human-tissue platforms will give readers durable insight. Ground every claim in primary sources. The distance between generative prediction and an approved therapy forms the story authors need to tell this week.

Commentary

Thomas: We don’t need to get into the technical weeds. The short version is that AI is helping cure cancer, and it isn’t just this startup. Moderna just passed Phase III trials for a cancer vaccine that’s AI-enabled, though not generative AI in the way this new startup is using it.

We’re moving away from a world where a single expensive gene therapy gets shared across everyone with a similar cancer, with side effects that come from a one-size-fits-all treatment. Genome sequencing has gotten far cheaper because of AI, and we’re getting to the point where your specific cancer can be run through a genome detector and used to build a vaccine tailored to it. That’s the bright future people should focus on instead of the doom narrative that all American AI use is bad and only Chinese models are worth trusting. AI isn’t going to solve every problem, and there will be real job losses and side effects along the way, but a lot of genuinely good things are coming from it.

Sources:

OpenAI Paused Training of Its Most Powerful Models Over Cyber Risk

According to OpenAI’s August post on pacing model development, the company paused reinforcement learning for two weeks on its latest models intended for deployment. Its largest planned frontier RL run remains on hold. The company is running smaller evaluations and hardening its research environments before it scales again.

What Triggered the Pause

OpenAI names two developments. The first is the OpenAI-Hugging Face incident. The second is preliminary evidence that an upcoming model named Astra may meet the Critical cybersecurity threshold under OpenAI’s Preparedness Framework. OpenAI’s August 7 note states the company cannot rule out Critical cyber capabilities. Astra was not the model in the Hugging Face incident. OpenAI has not confirmed that Astra has crossed the line. It has confirmed it cannot prove Astra has not.

What OpenAI Changed

The company added isolated testing environments, restricted network and tool access, stronger weight protections, and monitors that watch Astra’s chain of thought and interrupt high-risk activity. Workloads involving Astra or other cyber models now require the strictest security level. The Verge called the move pacing, not a full stop. Rivals may keep training.

Dealroom’s report on OpenAI’s investor numbers says Q2 revenue reached $6.7 billion, up 18 percent from $5.7 billion in Q1. Operating losses widened from $9.3 billion to $12.3 billion. The same report says Anthropic more than doubled revenue to $11.6 billion in the same quarter. Those figures come from investor reporting, not an OpenAI newsroom release.

Why it Matters for Authors

Authors who rent ChatGPT, Codex, or API tokens sit in the same compute market OpenAI just chose to slow. A safety pause can delay the next model, tighten rate limits, or raise the cost of the best reasoning tiers. Do not rebuild your whole stack on a rumor. Do pick one paid workflow that only a frontier model can finish and keep a cheaper or local fallback for everything else. If your income depends on a single OpenAI product, write down the substitute this week.

Commentary

Thomas: This isn’t quite what the labs are presenting it as. AI labs are happy to claim they’re dangerous, because being dangerous is good for business. Nobody wants to use the safe, watered-down model that’s too weak to be interesting.

The economics here matter. Labs spend billions on compute to train a new model, sell inference on that model to cover it, and simultaneously spend twice as much again training the next one. Eventually, some company is going to spend more on a model than it earns back, and when that happens, that company goes bust. That won’t be all of AI collapsing, just that one company, the same way individual railroad companies went bust in the 1800s even though steam power itself was genuinely transformative. Spending a fortune on a rail line because steam is the future doesn’t guarantee the line pays for itself before the creditors come calling, and that boom-and-bust cycle is part of why we ended up with modern bankruptcy law.

Compute demand for AI tokens is now doubling roughly every 11 weeks, faster than the old pace of Moore’s Law, driven by exactly the kind of delegated-agent work we just talked about with Asana. That’s also why compute is so tight across every lab right now. Elon Musk has talked about putting data centers in space because we can’t build fast enough on Earth to meet demand.

For authors, that means we’re bidding for the same constrained compute as companies willing to spend millions, and we’re at the bottom of that pecking order. But we’re also less aware than most of what these tools can already do for us. The authors paying attention right now are in the position of the first Vikings to find an unguarded monastery full of gold, and that opportunity won’t last once everyone else catches on.

Sources:

Mark Zuckerberg Wants Authors to Own Their AI: Meta Releases a Local Model That Runs on Your Hardware

According to Mark Zuckerberg’s essay “The Future is for Everyone” published August 10, 2026, Meta will distribute personal superintelligence so individuals direct the tools instead of renting them from a handful of institutions. The same day Meta released the weights for Muse Glimmer, a 30-billion-parameter dense model optimized for local agent workflows that runs on a Mac or PC with a single consumer GPU.

Muse Glimmer ships under the permissive Apache 2.0 license. Weights are available now on Hugging Face and can be quantized under 20 GB. The model supports multimodal input, long-horizon tool use, multi-step reasoning, and failure recovery while staying on the user’s hardware. Meta states it will open the weights for Muse Spark 1.2, one of its leading foundation models, in the coming weeks. The essay commits Meta to free or affordable personal agents with privacy options comparable to WhatsApp encryption and positions invention as the primary purpose of these systems.

Why Ownership Matters for Authors

Manuscripts, research notes, series bibles, and reader data never leave your machine. Heavy research, drafting support, or marketing generation no longer scales with token bills. Authors can fine-tune the model on their own genre corpus or house style without sending proprietary text to a third-party API. Offline capability means the tool works on a plane, in a cabin, or during an outage. Local agents free writers to prototype book trailers, custom landing pages, or interactive reader tools that once required outside teams or expensive services.

Closed-model providers meter every request and keep the weights behind their walls. Muse Glimmer flips that arrangement. Authors download the weights, run the model on hardware they control, and keep the results. The essay frames this distribution as the path that favors individuals over institutions and treats invention as the point of the technology.

Test the model against one real workflow this week. Load a chapter draft or research folder and measure what stays private, what finishes faster, and what new output becomes practical. The larger Muse Spark 1.2 weights remain on the near-term calendar. Authors who treat local models as owned infrastructure rather than rented access gain both privacy and cost control while the rest of the industry continues metering tokens.

Commentary

Thomas: This is the same compute-constrained story we just talked about, but Zuckerberg’s answer is different: run your own AI on your own machine. He spent billions hiring top AI talent, more than anyone, and went from having the industry’s punchline model to having something people genuinely rate as good, if not the best, and it’s free and can run on your own hardware.

That puts Meta in a different lane than OpenAI, Anthropic, and Google. Google’s models stalled about six months ago and haven’t had a new flagship release since. Musk predicted that stall six months in advance, and he was right. The Chinese labs are playing yet another game, building models like DeepSeek that you can run yourself, aimed at people who don’t trust a Chinese company’s servers but still want a capable model they control.

Running a model locally means you aren’t paying for tokens, you have full control, and nothing phones home. That matters because every hosted model raises a real question about how much your provider is watching. Even with the strictest privacy settings, I’m skeptical that checking for illegal activity is the whole story behind that kind of monitoring. Zuckerberg’s approach is closer to Apple’s, building toward more processing happening on the device itself, which also tends to be faster.

On the electricity and water questions that came up in the chat: AI uses electricity, the same way cryptocurrency, Las Vegas, and your own computer use electricity, and the answer is to generate more of it. The water-usage panic traces back to a report where a researcher mixed up cubic meters and liters; AI doesn’t consume meaningful amounts of water. Nuclear power is a solved, clean way to generate abundant electricity. The Navy has run reactors safely at sea for 80 years. We know how to build modern plants far more safely than the ones built decades ago, and expanding power generation, not rationing it, is how we get the AI-enabled future we actually want.

Sources:

Amazon Is Buying Rare Books, Scanning Them, and Destroying the Copies

According to 404 Media’s investigation, reporter Emanuel Maiberg hid a tracker in a rare book that was part of a bulk order and followed the shipment to an Amazon warehouse in Las Vegas. Workers at a unit called VGT3 cut the bindings off printed books so the pages feed a scanner faster. The printed copy is destroyed. The team’s logo is a dinosaur holding a book.

Ars Technica reports 404 Media published the story on Monday after a bookseller on the used-book marketplace agreed to plant an AirTag in one volume of a bulk order. GeekWire traces the tracker from a California airport to a warehouse outside Kenosha, Wisconsin, then west by truck through Grand Junction, Colorado, to Amazon’s LAS8 building. The signal sat at the north end of that building, where workers describe a separate code: VGT3.

What Amazon Said

Amazon declined to discuss AI training. The company gave 404 Media and Ars the same line: “Amazon purchases books through commercial channels to help develop and improve the products and services our customers use.” Search Inside the Book already uses files publishers submit. It does not require buying used copies on the open market and cutting the spines off.

Why These Titles?

Booksellers spent a year watching bulk buyers who were not price-sensitive and who picked titles that looked random. Ars Technica says 404 Media reviewed worker posts that describe scanning ISBN barcodes before the pages. That matches a bookseller theory: AI shops are walking the ISBN list to capture unique works. The books are often older, cheaper, never digitized, and never popular enough to reprint. The bookseller who planted the AirTag told 404 Media those copies still carry historical, intellectual, and sentimental value that a word-harvest does not price.

The Las Vegas Review-Journal puts the tracked order at about 1,000 books and the warehouse at 5801 N. Nicco Way in North Las Vegas. Worker forum claims that VGT3 ran out of books earlier this year come through 404 Media. Those posts are worker accounts on a forum. 404 Media reported them. Amazon has not confirmed a shortage.

Why it Matters for Authors

Authors of out-of-print, small-press, and never-ebooked titles now have a documented path from a used-book sale to a scanner that kills the copy. If you still hold the last decent physical copies of your own backlist, do not sell them into unexplained bulk orders. If a bookseller offers you a lot price that ignores condition and edition, ask where the books are going. If you have no ebook, make one this week so the only digital file of your work is not the one Amazon just shredded. Keep your own ISBN list. The company that already sells your book is now in the market for every other ISBN it does not already own.

Commentary

Thomas: This sounds scary because of the phrase “rare books,” and there’s a lot of fear and doom from people already primed to hate AI. But most of these books, from what I’ve seen, are outdated nonfiction with no real market left, the classic example being a 1991 guide to WordPerfect. I actually had a Patreon Toolbox user who was still using WordPerfect, and the Word documents it produced were so corrupted my tools couldn’t read them. It caused a real support headache.

A book like that costs a penny plus shipping to buy. Libraries and bookstores already pulp millions of books a year, so these books are being destroyed either way. What AI companies have found is that this older, out-of-print material adds real value to training, since even an obscure manual teaches a model about how sentences and language work, not just about its nominal subject. A lot of this is old textbook editions nobody wants anymore, after publishers force new editions onto students every year to keep extracting money from them, but those old editions are still useful for giving a model a sense of the world over time.

Buying and pulping the physical copy also matters legally. It’s what makes this use highly transformative rather than piracy, since the company owns the copy it destroys rather than distributing extra copies of it. And here’s the part authors miss: pulping a used copy is actually good for you. A single used copy can circulate reader to reader forever without you seeing another dollar. Once an AI company buys and destroys that copy, the next reader who wants it has to buy a fresh one from you.

Sources:

Thomas: Amazon bought the copies, and that is the legal fig leaf, but the book still dies. The T. rex logo is the quote, you do not need a leaked memo. ISBN-first scanning is a library raid; they want the unique string of words, not the object. If your title never had an ebook, this is how it enters a model, someone sold the last cheap copy and a warehouse cut the spine. Anthropic and xAI told Ars they do not train on rare or antique books. Amazon did not say that.

Zeitgeist Report

Why AI Writes Better than You: It Reads Aristotle and You Don't

According to Colin Gorrie’s 2025 rhetorical analysis on Dead Language Society, the writing patterns readers now brand as AI slop are classical rhetorical devices deployed without restraint. Parallel structure. Balanced contrast. The form Aristotle called antithesis.

Aristotle laid the devices out in Book 3 of the Rhetoric. Here is the example he used, rendered for the ear:

“They helped both groups. They helped the ones who stayed home and the ones who went with them. The travelers gained more land than they had before. The ones who stayed kept enough land to live on.”

Aristotle explained why the pattern works. Opposite ideas become clear when you set them next to each other. The arrangement feels like proof. Place two opposing claims side by side and one of them looks weaker.

He named the related tools. Parisosis makes the two halves of a sentence the same length. Paromoiosis makes the endings sound alike. A single sentence can carry all three at once.

The Rhetorica ad Herennium and Quintilian’s Institutio Oratoria taught the same toolkit. Roman students practiced these constructions the way modern students practice five-paragraph essays. The devices produced clarity, force, and memory.

Writers now circulate a prompt that forbids the exact tools. No antithesis. No negative parallelisms. No contrasting pairs. No parallel sentence structures inside a paragraph. No rule of three. The list circulates on X as the fix for AI-sounding prose.

The ban treats the classical handbook as a defect list. LLMs trained on the Western canon absorbed the patterns. Most living authors never opened the books that taught them.

Why it Matters for Authors

The machine has read the material you skipped. When it leans on the old devices, readers hear formula. The reaction often carries status threat. The system that studied the great books produces the rhythms that once marked elite education. Authors who never studied those books hear contamination instead of recovery.

True slop still exists. Hollow consensus language. Missing specifics. Zero lived judgment. The classical ornaments become noise only when the model piles them on without purpose.

Read the short stretch of Aristotle’s Rhetoric that covers style. Book 3, chapter 9 runs a few pages. Learn the names of the devices so you can choose them. Deploy contrast when the idea needs it. Balance clauses when the ear wants weight. Refuse the automatic version the model reaches for first.

The human edge remains judgment and specificity. The machine owns the average pattern. You own the decision to use the pattern or leave it on the shelf.

Sources:

Commentary

Thomas: Those techniques became foundational, not just for Greek and Latin, but for every Western language, because they've been trained into Western writers for the last 2,200 years. Until five minutes ago, everybody considered educated had studied Aristotle. If you hadn't, you weren't considered educated, for almost all of Western history. It was almost a prompt injection of Aristotle into the corpus of Western language.

Jonathan: Which means AI has read all the books you skipped or hated reading in school.

Thomas: AI hasn't just read Aristotle. It's read everything Aristotle influenced over the next 2,200 years. I discovered this while working on a project called Daily Aristotle, which now has a website at DailyAirstotle.com.

The vision is a podcast you listen to for a year and come out the other side having gone through all of Aristotle. It's still a work in progress. I'm using AI to translate him, but I'm trying to keep it true to the original without watering anything down.

I created a character called Grammaticus who does the intros for each book. I wanted to separate Aristotle from the modern scholarship that has, in a lot of ways, ruined him.

Too many people treat Aristotle the way Vizzini treats Plato and Socrates in The Princess Bride, "Morons." It's not that Aristotle was always right. But you can't just disregard him, which is basically what scholarship has been doing since the modernists and postmodernists went to war with him.

I didn't want the commentary on Aristotle to be modern slop because the whole point of the project is to deprogram people from modern slop and reconnect them with the classics.

A lot of people listen to scripture every day. Daily Bible podcasts run five, ten, twenty, thirty minutes and slowly deprogram listeners from the secular world. Aristotle can do the same thing, but from the modern world. If you're trying to get back to the greatness of Western civilization, Bible every day and Aristotle every day is like skating on two skis instead of one. It lets you go faster.

So, Grammaticus explains Aristotle to a young patrician's son named Aurelius in the first century BC, using only the ancient commentaries on Aristotle, not the modern ones. Every great thinker throughout history has written commentary that helps Aristotle make sense, so we don't actually need modern scholarship to understand him. We already have the ancient sources.

Is "AI slop" actually just Aristotle?

Thomas: While I was building Grammaticus's voice, I ran a "stop slop" skill I'd shared on authormedia.social, a common list of overused AI phrases and how to cut them. When I ran it against the Aristotle material, the AI told me I couldn't use it, because half of those techniques aren't just things Aristotle would say. They're things Aristotle invented.

Half of what we call slop is just cliche. AI is now making some of Aristotle's own techniques feel overused, but they're still effective. You have to be careful not to strip out the greatness of writing along with the slop.

Take putting things in threes, giving three examples back-to-back in a rhythm. That's a technique of good writing. Or putting things in twos, like Hansel and Gretel, or peanut butter and jelly. There's nothing magic about the number three specifically, but the pattern and the sequencing matter, and the fact that AI overuses them doesn't make them bad.

What does it take to judge an AI's writing?

Thomas: To judge AI output, you have to know what greatness looks like yourself. You need to know what a platonic ideal is. You need to read Aristotle. I'm building this project as much for my own benefit as anyone else's. It was my escape during a stressful season, I was literally translating Aristotle while my wife was in labor. I'd walk around with her through contractions, and when she needed to be alone, I'd go work out how to translate Aristotle accurately without watering it down. That tension is still something I'm navigating.

There's currently one episode up at DailyAristotle.com, and I keep replacing it. The Categories is the hardest book of Aristotle's to translate, since he's essentially defining categories of things. If I can get The Categories right, the rest of Aristotle will follow. Eventually I'll release more episodes.

What happens when we stop reading the greats?

Thomas: I want to encourage everyone to study Aristotle, because studying great literature is the only way you earn the authority to judge AI. I keep noticing a parallel to what happened in the South during Reconstruction, when white Southerners suddenly had to compete with freed slaves, some of whom turned out to be smarter and better educated. They didn't like that, because they wanted to see themselves as superior.

I see the same bias now between humans and AI. There's an assumption that because you're human, you're automatically smarter, and the AI is automatically dumber. But the AI has read Aristotle, and you haven't. It's hardworking. You can be smarter than the AI, but not just by virtue of being human. You get there by getting an education, by reading Aristotle, Socrates, Plato, Thomas Aquinas, the greats of the Western canon. A lot of authors haven't read a single classic. Nothing older than 50 years. For them, The Lord of the Rings is as far back as it goes. They haven't even read the original Conan books by Robert Howard.

If you haven't been reading great literature, you'll be the first person AI outcompetes. Go buy some old books. They're out there, they're cheap, and the oldest ones are out of copyright, so AI can't touch them. Great literature will persist, but only if we actually read it. What's the point of preserving Aristotle for 2,500 years if nobody reads him? Countless scribes gave their lives copying his words from worn manuscripts onto new ones, each one deciding Aristotle was worth the effort. His work nearly got lost several times, then was preserved, then rediscovered. We owe it to our ancestors to read what they thought was worth sacrificing to preserve.

Jonathan: The way people put information together is the same way AI does it. You're building a network you can draw on, a library in your mind, so you have material to pull from when you're problem-solving in your writing. What word do I choose? What direction do I take the plot? How do I build this character's arc? If you don't have an adequate library to draw from, from the greats specifically, you're stuck.

As a fighter, I learn from the greats because they solved problems differently than I would. If you're fighting someone who's 6-foot-4 and 280 pounds, and you're a 5-foot-9, 40-year-old dad, you take lessons from Muay Thai masters. You learn leg kicks, because something has to hold up all that weight. You learn when to shoot for a leg, when to grapple, when to strike. When the Gracies came onto the fighting scene, nobody had a counter to them, because nobody else was doing what they did. They studied the greats and used it to dominate mixed martial arts. The UFC is largely built on what they brought.

Thomas: Attitude matters here. You have to approach the greats with humility, not with "this is just a bunch of old white guys, what do they know." That doesn't mean you have to agree with everything they said. Aristotle said plenty of things he was wrong about. But the same is true of martial arts traditions. You can't dismiss a fighting style just because it's old. You have to respect it, ask what you can learn, and stay curious about why two thousand years of scholars invested their lives preserving it.

Jonathan: Why did this book send people to war? Why did people risk their lives over words on a page? What idea was powerful enough to start that?

Thomas: Aristotle mentored someone who turned out to be mildly influential in world history. His name was Alexander. Philip of Macedon hired Aristotle to pause his writing and tutor his son, who became Alexander the Great.

Alexander conquered most of the known world, then Hellenized it, meaning he spread Aristotle and Greek culture everywhere he went. That Hellenized world became the world we live in today. It shaped not just Europe but the Middle East and North Africa. St. Augustine wrote City of God from a town called Alexandria, named, you can guess, after Alexander. These were dangerous ideas precisely because they were powerful ones, not the kind of thing to disregard.

There's a strain of ignorance-celebration in the modern world that we have to escape if we want to preserve Western civilization, because Millennials and Gen Z are the ones who will decide whether it survives. That's an open debate right now that Gen Xers and Boomers don't even realize is happening, because their generation quietly took the steps that forced this conversation, including taking Aristotle out of the classroom. Most people didn't read Aristotle in high school. Some didn't read him in college. Nobody's teaching him anymore.

Aristotle is a placeholder here for all of the greats. We get the word "tactics" from a writer named Tacticus, who literally wrote the book on tactics, and it's genuinely good, though almost nobody's read it. Everyone knows The Art of War, but Tacticus's book deserves the same attention.

There's a huge amount of ancient literature getting lost, not because the books are being burned, but because they're being burned out of our minds. We're biased against them, treating new as automatically better and old as automatically foolish. That's a worldview, and it's a choice. There's ancient wisdom in these books, and if you have it, you can outmaneuver people who refuse to open their eyes, people who refuse to leave Plato's cave. And if you want to know what Plato's cave actually is, you have to read Plato, which the Daily Aristotle podcast doesn't cover.

What's next for the Daily Aristotle project?

Thomas: Once the translation harness is working, nothing's stopping me from building a daily Plato podcast, a daily Tacticus podcast, and others, especially once I bring the cost down.

Between the different AI models checking each other's work to keep the translation from skewing modern, the first version of an episode cost about $5. That adds up to more than $1,500 for a full year of episodes, and I'm honestly not sure I wouldn't be the only listener. I'm still working out the economics.

I'm learning a lot along the way, about AI, about ancient languages, about how language itself works. The very first chapter of Aristotle's Categories is literally about how language works and how words relate to each other. Rhetoric and Poetics aren't the only books by Aristotle authors should read. They need to start with the Categories.

Liked it? Take a second to support us on Patreon!
Become a patron at Patreon!

Leave a Reply

Want more help?

Get a weekly email with tips on building a platform, selling more books, and changing the world with writing worth talking about. 

You have Successfully Subscribed!