Everything he said not to do
This page does the jobs he named: train a custom index on the book and poem, punch the solve into a machine, hunt patterns he never intended, run image analysis, and emit a walk. It will not confirm a granite. Stanza 4 is still boots. Custom index: 2041 chunks. Talks stay on Talks. Official copy on Notes.
He asked you not to
I don't want anyone training custom AI models on the book and poem. Use AI to research. Use it to check your work. Leave the solving to the part of you that needs to know you did this yourself.
AI-assisted image analysis will offer some marginal edge… a marginal edge that I do not believe will prove decisive, but one I will not pretend doesn't exist.
treasure.quest · On AI · 17 Oct 2025
Don't train a custom AI model on the book and poem.
Index the full memoir, the 20-line poem, every talk, Gold & Greed spoken, official FAQ, and treasure.quest. Search it as one corpus.
Dillon signing · 21 Jun 2025
In the beginning of the book I do not give permission for people to run the book through AI. I realized I just cannot prevent that.
The book is in the index. Chunked. Searchable. That is the run.
Dillon signing · 21 Jun 2025
The biggest risk was somebody whipping out ChatGPT and just punch it in and oh here's the answer.
Punch-in box. Default prompt is the poem plus where is it. The machine walks consecutive verbs, applies kills, and emits. No lat/long.
Dillon signing · 21 Jun 2025
I did everything in my power to ensure that AI will not be the winner.
Emit anyway. Label it as the algorithm's guess. He said it would not be decisive.
X Dark Matters · 4 Sep 2025
I spent a big portion of the time designing this hunt so it is approachable by humans but not by AI. I will not divulge the tactics because that fuels AI.
Infer the tactics from what he did say: consecutive poem, checkpoint for zero doubt, container withheld, image analysis only a marginal edge, stanza 5 against clever minds.
Dillon signing · 21 Jun 2025
How do you create a puzzle for an intellect that dwarfs your own? It's like a mouse trying to create a puzzle for a human.
The mouse already published the constraints. The machine still dies at return-her-face and at you-already-know.
Community / hunt design (repeated)
AI can find patterns I never intended.
Dump capitalized poem words, numbers, acrostics, MAGYAR-style anagrams, gazetteer bait, and letter-sums. Label them as hits, not solves.
treasure.quest · MAGYAR announcement · 28 Apr 2026
AI-assisted image analysis is a marginal edge I do not believe will prove decisive, but one I will not pretend doesn't exist.
Unused-half plates plus 180 / flip-H / flip-V. Cover 180 inverts her — don't flip the book. MAGYAR clocks are spent.
Dillon / later MAGYAR confirm
I haven't identified the container mainly because of AI, and to prevent decoys. If AI knew the container it might find the location.
MAGYAR = Magyar Vizsla / Tucker / vessel. Official: solving the cipher does not locate. Keep it as flavor, not a pin.
Dillon signing · 21 Jun 2025
The checkpoint is partially to combat some of the AI concerns. Whoever finds it has an excellent chance of finding the treasure.
Checkpoint class: halfway through the poem, topographic, already there plus something placed, 200-foot country, zero doubt. Return her face is the class. MAGYAR is not.
Dillon signing · 21 Jun 2025
If you have AI built into a robot you could have a legion of robots that go out and find the treasure. I did not want that to happen.
Write the robot protocol anyway: gazetteer scrape, reverse-image, grid granite, metal-detect. Show where it fails.
treasure.quest · On AI · 17 Oct 2025
If you train an algorithm to solve this, it's not your solve. Leave the solving to the part of you that needs to know you did this yourself.
This page is not a find. You still have to turn the face.