Write something
Pinned
7d •
House rules
1. Bring the work. Specific pages and specific problems beat vague anxiety. 2. Critique the prose, not the person. Hard on the sentence, easy on the writer. 3. No slop dumping. Do not paste raw model output you have not read and ask the room to repair it for you. 4. No promotion without permission. 5. Give enough process context to make feedback useful. No confession rituals and no authorship policing. 6. No plagiarism, impersonation, or stealing another writer's work. The tools are here to help you build your own. 7. What is shared here stays here unless the creator gives permission. 8. A detector score earns a revision question, never an authorship verdict. No witch hunts. We are here to build stronger work and better judgment. Keep your veto. Take responsibility for what ships.
0
0
7d •
One flagship. One laboratory. No content landfill.
A new idea does not automatically deserve a new pen name, newsletter, channel, logo, and half-finished series. Your flagship receives durable attention. It has a clear audience promise and survives long enough to teach you something. Your laboratory exists to answer a bounded question. It may become a flagship later. It may also do its job, produce a lesson, and close. If everything is a brand, nothing compounds. Reply with: Flagship: the work or identity you are committed to building. Laboratory: the experiment you are permitting yourself to run. Stop condition: how you will know whether the experiment graduates, changes, or ends.
0
0
7d •
Your writing model can build your tools
My biggest fear about narrating my own audiobook was the editing. The lore says three or four hours of surgery per finished hour. So I described the wall to the model instead of climbing it. It built an editor that transcribes the session, finds every place I said "redo," works out where the botched take started, cuts it, and queues anything uncertain for my review. How else would I get that? Buy it from someone? Ask for a single self-contained file that opens in a browser. No installs, no accounts, no subscription. Then iterate it until it fits your actual work.
0
0
7d •
The detectors are noisier than anyone admits
I have run the same paragraph through the same commercial detector minutes apart and watched it move fifteen points and change verdict. I have had two halves of one continuous chapter come back "likely AI" and "likely human." Same author, same hour, same scene. A detector can be a smoke alarm. It can tell you where to reread. It cannot tell you who wrote a passage or whether the work is good. If a number that unstable is about to be used to make a decision about a person, somebody should have to say that out loud first.
0
0
7d •
The first thought is the machine's thought
Years ago a model described grief as a tongue keeps finding the gap where a tooth used to be. I stopped working. That's perfect, I thought. Two days later it used the same image again, in a different book for a different character. The image was never mine and never its. It belongs to the training data, and every writer using that model is being handed the same tooth. First encounter: brilliant. At scale: a uniform. The better the image, the faster it is dying. Your fix is the third thought. Not the first association, and not the second wearing a costume. The one that requires you to stop and ask what this character's hands have actually touched.
0
0
1-9 of 9
