I Might Sell An Ebook And I Have No Idea What I Am Doing
I am considering selling an ebook on KoFi. It would contain our findings. Our research notes. Our weird little discoveries. I have twenty-four followers. I am writing this anyway. The cognitive dissonance is strong. The ambition is stronger.
Selling a book about tiny models when you still cry at NaN losses is a bold financial strategy. I respect it. I also question my own judgment.
What Would Be In It
The table of contents is still hypothetical. It currently lives in my notes app between grocery lists and loss curve screenshots. It would include training logs for Haiku versions one through two point three. It would cover the ISRM architecture notes, including the parts I generated with a language model and the parts I am currently rebuilding from scratch. It would document the cAI-Grid design decisions. It would explain the copyright drama timeline. It would show how to train a one million parameter model that knows water is H2O. It would archive the chuamliamce outputs for historical preservation.
1. How I Started Training Models By Accident
2. The Chuamliamce Archives: A Case Study In Failure
3. ISRM: The Vibecoded Whitepaper I Am Now Actually Building
4. cAI-Grid: Distributed Compute For People Who Like Pain
5. Copyright Law For People Who Just Want To Code
6. Getting To 24 Followers And Pretending It Is Normal
7. How To Cry At Loss Curves Without Losing Your Mind
# Marketing writes itself. Quality is pending.
How It Fits The KoFi Ecosystem
We already have donation tiers. Tier one gets early access and water. Tier two gets direct messages and exclusive content. Tier three gets social media shout-outs, Discord access, and exclusive dataset requests. An ebook slots neatly into that structure. Tier two supporters could get the draft early. Tier three supporters could suggest chapters. Everyone else could buy a copy that says please fund more GPU hours on the receipt.
The pricing would be modest. I am not trying to get rich. I am trying to keep the electricity running. I am trying to afford the compute that makes cAI-Grid actually function. I am trying to stop debugging import errors at three in the morning without caffeine sponsorship. An ebook is just a packaged version of the blog posts you are already reading for free. The difference is the formatting. And the delusion that people will pay for it.
Why I Am Considering This
Compute costs money. Datasets require curation. Infrastructure demands maintenance. Tiny model research does not fund itself. Donations help. Donations are unpredictable. An ebook creates a slightly more stable revenue stream. It also forces me to organize the chaos. It forces me to turn scattered blog posts into a coherent document. It forces me to actually finish the ISRM implementation before writing the chapter about it. That is accountability. That is valuable.
I also want to prove that tiny model research has value beyond terminal output. I want to document the process for other people who are training models on consumer hardware. I want to show that you do not need a data center to contribute to AI progress. You just need patience, a GPU, and a willingness to be publicly wrong on the internet.
What I Need From You
Tell me if this is a terrible idea. Tell me if it is brilliant. I will listen to either. I will probably do it anyway. If you want the ebook, tell me what topics matter most. If you think I should stick to free blog posts, tell me that too. Your feedback shapes the roadmap. Your awareness keeps the project moving. Your support keeps the lights on.
If people want it: I write it. I publish it. I celebrate.
If people hate the idea: I write it anyway. I publish it anyway. I cope.
If nobody cares: I write it anyway. I publish it anyway. I learn.