Working on Adult Thinking System

If anybody thinks that writing a book with chatGPT consists of prompting it and sitting back, either they haven't tried it or they hadn't read what it produced. I've spent about 8 hours on 2 pages, getting a production pipeline set up, re-evaluating older pages after learning better alternatives, and especially getting chat to render my face without distorting it. We (I have to refer to the combo of me, codex, and chatgpt as "we" because it couldn't be done without all of us) got through 35 pages when I became dissatisfied with the relationship between storytelling and information presentation.

I asked chat to re-evaluate the first 35 pages, based on what we learned about narrative/infographic ratio. A symbiotic version of the "Recursive Self Improvement" that frontier AI developers seek and fear. Critique of its own output, steered by my prompts and queries, is followed by a detailed plan for revision of the critiqued draft. In response to my request for self-critique, it presented a detailed plan for redoing the whole thing. Now codex is involved on the production side. I spend time copying and pasting from chatGPT into codex, then copying and pasting the response. The process involves many checkpoints at which I need to decide between options, and I often need to remind/redirect. It's as complicated and time-consuming as writing a book without AI.

I asked chatGPT to use images of me to be found on the Web as input for the Howard model codex uses to produce pages that include my likeness. Note the part about permission to reproduce,

Next step only: create a Howard web-reference and pose strategy.

Do not generate images.

Do not rebuild any page.

Do not use web images as final page assets.

Do not generate Howard’s face or hat.

Do not use Git.

Do not ask me to run Terminal commands.

Purpose:

There are many public Web images of Howard Rheingold showing different angles, poses, gestures, hats, and expressions. We may use these as pose/angle references and shot-planning references, but not automatically as final production assets.

Create:

docs/HOWARD_WEB_REFERENCE_AND_POSE_STRATEGY.md

Include:

1. Purpose of using Web images:

- pose planning

- angle reference

- expression reference

- continuity comparison

- shot-list creation

2. What Web images must NOT be used for:

- final artwork unless rights are cleared

- automatic pasted photo assets

- generated “identity approximation”

- replacing the approved Howard asset policy

3. Rights / permission rule:

- If a Web image is to be used directly in final art, it must have permission, license, or clear rights.

- Otherwise it is reference only.

4. Recommended reference categories:

- full face

- three-quarter face

- side/profile showing aquiline nose

- hat close-up

- speaking gesture

- listening pose

- hands-on work pose

- seated laptop pose

- writing pose

- walking/full-body pose

- over-shoulder/looking-at-board pose

- two-person conversation pose

5. How this should change page planning:

- Howard appears only when his presence adds meaning.

- If no approved pose exists, either make the page Howard-free or request a new real source photo.

- Do not force Howard into every page.

6. Create a proposed new Howard source-photo shot list for future pages:

- writing/downward desk pose

- laptop/typing pose

- standing at evidence board

- over-shoulder board/screen pose

- two-person conversation

- listening pose

- walking/full-body pose

- hands-on making/testing pose

- table-sorting/redaction pose

- pointing/gesturing toward object or screen

7. Add columns for a future rights/reference ledger:

- reference image description

- URL or source

- photographer / rights holder if known

- intended use: reference only / final asset candidate

- permission status

- pose category

- notes

8. Update the production rule:

Web images may guide pose planning, but final Howard imagery must come from:

- approved source-backed Howard assets,

- user-provided photos,

- newly captured photos,

- or explicitly licensed/cleared images.

Then report back:

1