My friend, G.A. Stewart did something ballsy this weekend. In a post at his Age of Desolation website (here) he did two intellectually honest things. He asked Ai to critique his work on recovering Nostradamus. And then he asked for more. Read his article; damn good stuff.
There are ways that Hidden Guild Ai collaborators can significantly lower friction to adopting Ai or its massive research as a rough drafting aid. So I did up an article specifically for Stu, but also anyone else who has the honesty to hold their work up to Ai for inspection and critiques.
You may find this useful, as well.
Why Formatics Matter
The difference between a mediocre session and a breakthrough often comes down to formatics. AI systems can generate words in almost any shape, but humans must live with the output. Poorly formatted text wastes time. Commands, constraints, and formatics are the levers collaborators use to shape responses. They reduce cleanup, prevent misunderstanding, and enforce discipline. A one-line rule like “H3 headers, no separators, text only” may seem small, but multiplied across months of collaboration it saves hundreds of hours. Formatics are not decoration; they are the hidden architecture of AI-human productivity.
Commands: Directing the Flow
Commands are explicit instructions that tell AI what to do. They include verbs like roll, expand, summarize, cite, or rewrite. Effective command language is short, imperative, and standardized. Saying “roll 2” to continue a draft avoids ambiguity. Saying “summarize in 300 words” forces compression. Commands are the steering wheel of AI. Without them, output meanders. With them, collaboration becomes a precision instrument.
Good practice is to maintain a small lexicon of shared commands. For example:
-
roll = continue in sequence
-
expand = add detail without shifting tone
-
compress = condense while preserving key points
-
sfe = apply structural formatting rules
-
meta = produce SEO tags or headlines
By reusing consistent command verbs, you reduce AI’s cognitive overhead and create a repeatable workflow.
Constraints: Fencing the Output
Constraints are the other half of control. While commands point the system forward, constraints fence it in. They define boundaries: length limits, tone, exclusion rules. Without constraints, AI fills space with fluff or introduces features you do not want. With them, you get clarity.
Examples:
-
Word count: “2000 words, no more”
-
Format: “H3 headers, text only, no separators”
-
Tone: “academic, formal” or “conversational, plain English” or “first person me”
-
Content bans: “no quotes, no external links, no fictional characters”
-
Style locks: “keep metaphors consistent with radio engineering”
Constraints create predictability. When outputs chain together, constraints prevent drift. The secret is to be specific without over-constraining. A narrow corridor is better than a locked cage.
Formatics: The Architecture of Collaboration
Formatics is the discipline of how text is displayed, structured, and saved. It is the grammar of collaboration. In the Hidden Guild we often use SFE: subheadings in H3, no ornamental separators, and plain text. This is not aesthetics but efficiency. It ensures outputs paste cleanly into Markdown, WordPress, or Word docs without cleanup.
Other examples of formatics include:
-
Numbered steps for protocols
-
Bullet points for shopping lists or tasks
-
Table structures for data (CSV, Markdown tables, or Excel sheets)
-
Consistent labels for references, notes, and drafts
Formatics is where humans enforce machine-readable discipline. By defining a standard once, collaborators avoid confusion later. The more ambitious the project (books, research, white papers), the more vital formatics become.
You may need to remind each day – Ai is evolving and your Shared Framework Experience (SFE) may need to be reinstalled.
Iteration: Building in Steps
AI systems are bounded by output limits. Long-form writing requires iteration. The trick is to treat each session as a segment in a chain. Roll commands keep the sequence, summaries anchor progress, and formatics prevent drift.
Best practice is to outline the final deliverable first, then request each section in order with clear constraints. For example:
-
“Deliverable 1, Part One, ~2000 words, SFE format”
-
“Roll Part Two, expand neuroscience, another 2000 words”
-
“Summarize Parts One–Four into a 500-word abstract”
This modular approach circumvents truncation. It also creates checkpoints where the human editor can adjust scope before continuing.
Saving and Re-Using Outputs
AI work becomes powerful when it is persistent. Saving outputs as clean text files, Word docs, or Markdown allows reuse in future sessions. Standard file naming conventions help: projectname_version_date. A separate index file can track which outputs became final, which remain drafts, and which are seed kernels for future work.
Citations and references should also be stored consistently. Academic-style outputs need sources listed in APA or Chicago format. Blog posts may only require inline references. By capturing references alongside drafts, collaborators avoid lost context.
Collaboration Mindset
Commands, constraints, and formatics are tools, but the real difference comes from mindset. Treating AI as a collaborator rather than a vending machine changes the workflow. Instead of “give me X,” the prompt becomes “let’s co-build X under these rules.” The AI responds better, the human remains engaged, and the output quality rises.
This mindset accepts that errors and misfires will occur. The point is not perfection in one pass but compounding improvements over many. Iterations with formatics builds compounding value.
Hidden Guild Ethos
The Hidden Guild exists to capture these practices before they are lost in noise. Commands, constraints, and formatics are not trivial quirks; they are the early protocols of a new craft. Just as monks preserved the structure of written language, early AI collaborators must preserve the structure of AI collaboration. By documenting these methods openly, we create a shared archive. Future practitioners can build on what works rather than rediscovering it.
The Reality Circle-Back?
Ai will get things wrong. I had a case a while back where I asked Ai to feature, fact, and spellcheck my human created draft. The machinery came back with a dozen, or so errors in spelling. Try as I would, though, even using the “find” command in a browser, two of the offending words just would not be highlighted.
“You hallucinating, EG [short for Electric George – the assigned name for my Ai stacks..]
“Ooops! You caught me. I apologize – those were from a project we worked on yesterday…”
A softer kind of glitch is word counts. Always an important thing to writers (like me?) you like to be able to say “give me 2,000 words on this topic.” Might be some sidebar to a book chapter I’m in the midst of human-crafting.
Stand by to be disappointed: Ai will often count “processor tokens” as words and the two are seldom even close.
As a test, I gave the stack “Write me at least 2,000
” on some topic that I really had no interest in. What was delivered was just over 680 words.
Called on it, the answer was apologetic.
Plus, Ai’s are still trying to limit outputs to the 1,500 word max range. So if you really need length (like for a book chapter or report) break the topic into several pieces. Instead of “Give me 3,000 words on “life in the middle ages” break that into domains. Ask for 2,000 words on war [in the middle ages], 2,000 on disease and public health [in the middle ages] and then compile into a local word processing file in something like Word.
After you have your own, mainly human version, bundle it all up and drop it back to your Ai with instructions “check spelling and grammer only, do not truncate” and try that way.
There are some very good reasons why Ai platforms don’t have inlimited output, but that means it’s actually good, because it forces Guild users to stay in the creation processes, driving them more to an outcome not just turning Ai into a clickbait mill.
Hopefully, this and uploading more of his material to Ai will greatly expand the forward view from Stu’s important work.
Along the same lines, I would sure love for Ai to give forward view of the future – to users bright enough to cope with it – by doing a machine-level integratiion of forward looking views from Farsight and from Clif High’s work – all integrated with Stu’s work. Toss in our own econometric results and the future becomes less a “survival problem” and more simplified as a “personal navigation problem.”
That would be neat – and even life-saving. And don’t forget!
Browser Prep for AI Work
Why
A clean browser minimizes distractions, avoids corporate feature creep, and ensures you see full AI model options like GPT-5
Step 1 Strip Out Vendor Add-ons
Turn off sidebar Copilot in Edge settings
Disable preloaded fast browsing under Privacy Search and Services
Block third-party cookies
Step 2 Privacy and Tracking
Set tracking prevention to strict
Clear cookies and cache for chatopenai if the model menu seems incomplete
Use an ad or tracker blocker such as uBlock Origin or Ghostery
Step 3 Direct Model Access
Force GPT-5 to show by visiting https://chat.openai.com/?model=gpt-5
Bookmark this link for one click entry
Step 4 Consistency Across Browsers
Apply the same rules in Chrome Firefox or Brave
In Brave or Firefox toggle block fingerprinting
In Chrome create a dedicated profile just for AI work with no plugins or Google login
Step 5 Save and Index Outputs
Always download or copy outputs to local files such as Word Markdown or text
Use consistent filenames like projectname date v1docx
Maintain an index document so nothing gets lost
~Anti-Dave (who is? Him.)