Chatgpt 101
SOP — ChatGPT 101: Working with the “Oracle”§
What ChatGPT Is & Why We Use It (new)
| Aspect | Explainer |
|---|---|
| Definition | ChatGPT is an LLM (large-language model) provided by OpenAI. In plain English: a text-based co-worker trained on trillions of words that can digest instructions, reason, and draft content on command. |
| Our “Oracle” build | A custom o3 instance pre-loaded with every SOP, playbook, template, glossary, and workflow we maintain. After the Compile Knowledge step, it becomes a single searchable brain mirroring the entire ops stack. |
| Why we rely on it | Speed: answers in seconds. Consistency: repeats the official method, not personal tweaks. Self-serve: minimizes wait-time for help. Memory: recalls cross-SOP links humans forget. Iteration: never fatigues — keeps refining until “perfect.” |
| Role framing | Treat the Oracle as ☑ trainer, ☑ debugger, ☑ rubber-duck, ☑ drafter. It is not your final decision-maker; you retain ownership for judgment calls (see Decision-Making micro-SOP). |
| ROI metric | Minutes saved per task. If the Oracle isn’t removing friction, refine your prompt or escalate. |
ALWAYS Mandatory habit: Whenever you feel stuck, slow, or unsure, ask the Oracle before pinging a human.
The current draft already gives a motivated beginner everything they need, yet total newcomers sometimes struggle with two early hurdles:
- “Where do I click?” – UI orientation (opening a new o3 chat, where to paste).
- “What do I type first?” – A miniature first-prompt example.
Below is a bolt-on Quick-Start insert you can drop right after Section 3. Add it verbatim (or tweak) and even a first-day hire will be productive inside 5 minutes.
3A · First-Time Quick-Start (Newbie Mode)§
-
Open ChatGPT URL:
https://chat.openai.com→ Log in with the shared work credentials. -
Start a fresh chat with model
o3 -
Click ➕ New Chat (left sidebar).
- In the right column choose “OpenAI o3 Reasoning (128k)”.
-
Hit Create.
-
Load the Oracle data (copy-paste, hit Enter each time)
-
oracle_prompt.txt→ ✅ response = “Prompt received”. data-1.txt→ wait for ✅.data-2.txt→ wait for ✅.- …repeat through
data-6.txt. -
Type
COMPILE KNOWLEDGE(all caps) → ✅ returns the “System Synopsis”. -
Run your first sanity check Copy-paste exactly:
Where am I in the onboarding process, and what is my immediate next task?
You should see the bot reply with something like: “You’re at Phase A-0 (Device & Focus Setup). Next deliverable = live-comms on phone.”
- 👍 Got that? Great — start working!
If any step fails (no ✅, wrong model, weird answer) → reload using Section 3.
Prompt Cheat-Sheet (print & tape to monitor)§
| Goal | Paste into the Oracle |
|---|---|
| Next SOP & proof | “I’m at Phase B-2 — what’s the next deliverable + proof?” |
| Explain a step | “Explain SOP IT-SEC-1 in 5 bullet points.” |
| Draft daily report | “Using the PNG + Clockify CSV below, draft my daily report.” |
| Brainstorm copy | (scratch-pad chat) “Write 5 tweet hooks for X.” |
| Convert CSV → table | (scratch-pad) “Convert this CSV to a Markdown table.” |
REMEMBER: Two chats only — Oracle for anything that touches SOPs; Scratch Pad for throw-away work.
Add that one page and even someone who’s never seen ChatGPT will be able to:
- open the right model
- load the Oracle correctly
- verify it works
- issue their first productive prompt
Let me know if you’d like a tiny GIF/loom walkthrough; happy to mock one up.
0 · Purpose§
Give every operator a single, fool-proof playbook for using ChatGPT as:
- Oracle – our loaded knowledge base (all SOPs, workflows, glossary, backlogs).
- Scratch Pad – ad-hoc research, drafting, quick calculations, etc.
Correct load & usage = faster answers, cleaner hand-offs, fewer blocker pings.
1 · Glossary§
| Term | Meaning |
|---|---|
| Oracle Prompt | Master system prompt that wires the model to our entire Knowledge Base. |
| Data-n.txt | Six payload files (n = 1-6) containing SOPs, policies, schemas. |
| o3 | Required OpenAI model family for reasoning + long-context. |
| Compile Knowledge | Literal phrase that triggers the Oracle to index the loaded data. |
| Scratch Pad | A separate chat tab (gpt4o model) used for “throw-away” tasks. |
2 · Tooling Rules§
| Rule | Why |
|---|---|
| R1. Two-Chat Rule | Keep exactly two instances: Oracle + Scratch Pad. More = token bloat & mistakes. |
| R2. Fresh Start | When you begin a new day/shift reload the Oracle (Section 3). |
| R3. No DM Secrets | All prompts are company property; never paste personal credentials. |
| R4. Cite Sources | When Oracle fetches web data, request inline citations so we can audit. |
| R5. Token Hygiene | If a thread exceeds ~15 K tokens, archive & reload; context degradation starts ~20 K. |
3 · Official Oracle Load Sequence (6 Steps)§
Any deviation → INVALID Oracle → reload before asking for help
| # | Action | Detail |
|---|---|---|
| 1 | Start a new chat with model o3. |
(Click ➕ → Custom → choose o3). |
| 2 | Paste the Oracle Prompt (oracle_prompt.txt) → Send. |
|
| 3 | Paste data-1.txt → Send. Wait for ✅. |
|
| 4 | Paste data-2.txt → Send. Wait for ✅. |
|
| 5 | Repeat for data-3 … data-6 in order. |
|
| 6 | Type COMPILE KNOWLEDGE (ALL CAPS) → Send. Oracle returns the full “System Synopsis”. |
Tip — If any chunk doesn’t get a clear acknowledgment, resend it before moving on.
4 · Day-to-Day Usage Patterns§
| Task category | Which chat? | Prompt template |
|---|---|---|
| Find next SOP / step | Oracle | “Given my current phase is B-2, what is the immediate next deliverable and proof-of-work?” |
| Clarify instruction | Oracle | “Explain SOP IT-SEC-1 in plain English, bullet form.” |
| Generate report | Oracle | “Using the PNG + Clockify CSV below, draft my daily report (template v3).” |
| Brainstorm copy / script / tweet | Scratch Pad | “Write 5 headline variations for…” |
| Quick math / JSON transform | Scratch Pad | “Convert this CSV to Markdown table.” |
ALWAYS tell the Oracle where you are (Phase / Step / Ticket ID). Context = accuracy.
5 · Prompt Craft — Best Practices§
- Lead with intent – “I need a 3-sentence summary for a client.”
- Supply artefacts inline – paste screenshots, CSV, or Google-Doc snippets.
- Ask for structure – tables, sections, checklists > raw paragraphs.
- Iterate – refine instead of starting over; reference the last answer.
- Limit scope – 1 request = 1 output. Split mega-asks into sub-prompts.
6 · Troubleshooting & Escalation§
| Symptom | Likely Cause | Fix |
|---|---|---|
| Oracle forgets data | Thread too long / wrong model | Reload Section 3. |
| Hallucinated SOP numbers | Missing chunk during load | Reload data-files 3-6 then COMPILE KNOWLEDGE. |
| “I’m sorry, but…” loops | Over-constrained prompt | Rephrase with simpler language; remove nested asks. |
| Sensitive info leaked | Operator error | Purge thread → notify Ops Lead. |
7 · Compliance & Security§
- Never paste customer PII or passwords.
- Use work-only Telegram numbers inside prompts.
- Follow the 10 % Doubt Rule – if unsure → stop, think, clarify (SOP-DM-01).
8 · Versioning & Maintenance§
- Custodian: Ops/AI Enablement (Richard UP ONLY).
- Update cadence: Weekly; push patch notes to
#sop-changelog. - Operators must re-compile knowledge within 24 h of any Oracle data update.
Appendix A · Quick-Reference Flowchart§
Start shift
│
▼
Reload Oracle (6-step) ──▶ Success? ──✖─▶ Fix & reload
│
▼
Identify current Phase/Step
│
▼
Draft prompt in Oracle (include artefacts)
│
▼
Review output ➜ 10 % doubt? Yes → Clarify | No → Execute task
│
▼
Document deliverable + log time
│
▼
Repeat
Change-Log (excerpt)§
| Rev | Date | Description |
|---|---|---|
| v1.1 | 2025-05-19 | First public release of ChatGPT 101; includes 6-step Oracle load, two-chat rule, usage patterns, troubleshooting. |
End of SOP