How to modify chatGPT work flow to operate in Telstra Mobile Data Dark patches.

 

# ============================================================
# Operating AI Workflows in Telstra Mobile Data Dark Patches
# ============================================================

CORE REALITY
- Assume comms are unstable.
- Tools will fail quietly and continue sounding fluent.
- Fluency is inertia, not lift.

FAILURE ORDER
- STT fails first.
  - Desktop browser STT dies before mobile.
  - iPhone STT is more resilient (tighter audio + buffering).
- Text output may continue after capacity is exceeded.

EARLY WARNING SYMPTOMS (AIRSPEED BLEED)
1. Constraint slippage
2. Glue language appears (“overall”, “in summary”)
3. Verbosity up, information density down
4. Rephrasing replaces progress
5. Partial compliance with confident tone
6. Silent revision of frozen outputs
7. “Done” tone with missing elements

RULE
- No symptoms ≠ safe.
- Assume stall risk under load or bad comms.

THREE TEST MODULES
[TEST 1: CONSTRAINT PIN]
- Reassert one non-negotiable constraint mid-task.
- FAIL: acknowledged, then violated.

[TEST 2: DELTA INJECTION]
- Add a small change dependent on prior state.
- FAIL: ignored or whole task restarted.

[TEST 3: FREEZE & RECALL]
- Freeze output A. Require exact reference to A.
- FAIL: paraphrase, drift, or silent edits.

TERMINAL STATES
9. COMA   – fluent, hollow, unresponsive to tests
10. DEATH – confident nonsense; runway assumed, terrain confirmed

WORKFLOW MODIFICATIONS
- Chunk ruthlessly: one task, one objective
- Freeze early and often; capture externally
- Externalise state (notes outside the chat)
- Write fragments first (nouns, verbs, decisions)
- Avoid nesting and long dependency chains
- Stop early; checkpoint before it looks impressive
- Defer synthesis to clean networks

OPERATIONAL SPLIT
- Bad comms: thinking, fragments, note-taking, humans
- Good comms (Men’s Shed / landline): synthesis, summaries, structure

BOTTOM LINE
- You can fly in dark patches.
- Just don’t pretend it’s VFR.
# ============================================================
 

Assumption: comms will drop, stall, or gaslight you. Design accordingly.

Principles

• Treat AI output as volatile, not persistent

• Fluency ≠ capacity; assume silent failure under load

• Interruptions add overhead, not insight

Workflow Adjustments

• Chunk ruthlessly: one task, one objective, short hops

• Freeze early, often: capture outputs immediately; don’t rely on scrollback

• Externalise state: notes live outside the chat (Mark’s Log, notebook, text file)

• Fragments first: jot nouns/verbs/decisions, not prose

• Avoid nesting: no long dependency chains in one pass

• Checkpoint mindset: stop while it’s still working, not when it’s impressive

• Defer synthesis: heavy summaries only when comms are clean

Operational Split

• Bad comms: thinking, note-taking, human conversation

• Good comms (Men’s Shed / landline): synthesis, summarisation, structured output

Bottom Line
You can fly in bad weather—
just don’t pretend it’s a sightseeing trip.