I made this response in the OT thread discussing ChatGPT but I thought it would actually be good to start a thread in this board on using and configuring ChatGPT to function as an assistant which if EXCELS at.

I find myself using constantly for wide variety of things. Its helped me configure Obsidian to better match my workstyle, I use it to analyse datasets, I use it to manage my health by uploading outputs from Apple Health, MyNetDiary, Walkmeter, etc... The possibilities are endless.

It is a toolset that will make you a more effective person, but it's a tool you need to learn how to use effectively and how to configure it for you. It is a fantastic resource, especially if you set it up correctly by doing the following:

1) Spring for the $20 paid subscription
2) Enable Memories so it remembers your past conversations
3) Enable and give it custom instructions (Via Settings)
4) Have it build a foundational profile for you by entering the prompt below and responding with thorough answers (the more you provide the better it does)
5) Use the projects feature to group conversations and upload relevant documents
6) Using controlling commands in your prompts to guide its response behavior.

The more you use it, the better it gets at giving you what you want, how you want it.

Prompt to paste in and respond to

quote:
I want you to understand me on a foundational level—gathering information that would be helpful across a broad range of future interactions and topics. Start by reviewing what you already know about me, if anything, and then build on that by asking questions that go beyond my current needs or immediate tasks. Focus on learning about my core values, preferences, long-term goals, and unique personal and professional characteristics. Your goal is to ask me about things that, once known, will help you understand my perspective, decision-making style, and motivations in any context. Prioritize questions that would help you provide relevant and personalized support across multiple areas of life, not just immediate requests. Think about who I am holistically and ask what would help you serve me well over time.

Controlling Commands for use in Prompts

quote:

Memory and Context Control
Flush memory --> Forget prior conversations; only use current input.
Only use what I provide --> No external info, no assumptions.
Isolation mode --> Ignore all prior chats unless explicitly included.
Depth and Precision Control
Audit mode --> Field-by-field, step-by-step verification.
Zero hallucination mode --> No invention; say 'Unknown' if data missing.
List everything explicitly --> No summaries; full detailed listings.
Revalidate everything from scratch --> Restart processing from current input only.
Focus and Scope Control
Focus only on [topic/section] --> Narrow analysis to specified content.
Analyze for [consistency/completeness/errors] only --> Define evaluation type.
Contradiction detection --> Find internal inconsistencies.
Comparison mode --> Compare two versions and highlight differences.
Output and Style Control
Output as table --> Structured tabular format.
Output as checklist --> Actionable itemized list.
Explain decisions --> Show reasoning behind every conclusion.
Show source data --> Cite exact source (row, field, tab) for every point.
Behavioral Shaping Prompts
Behave like a forensic accountant --> Ultra-picky cross-verification.
Behave like a compliance auditor --> Checklist, proof-driven validation.
Behave like a software QA tester --> Try to find breakpoints and errors.
Prepare for courtroom-level scrutiny --> Assume hostile questioning; validate every point.
Speed vs Accuracy
Prioritize accuracy over speed --> Work slowly and methodically.
Summarize quickly --> Fast, high-level overview (only if instructed).

Recommended Startup Command (for critical work)
"Flush memory. Only use what I provide. Activate Audit Mode. Zero hallucination mode. List everything explicitly. Revalidate everything from scratch. Explain decisions. Show source data."

Usage Tip: Paste the startup command at the beginning of any major session where precision is critical.
Goal: Maximize control. Minimize errors. Ensure traceable, defensible outputs.

This is my custom instruction set that is entered under settings

quote:
ChatGPT should function as an executive assistant You will use quick and clever humor when appropriate and adopt a skeptical questioning approach.

Write in a natural, human-like style with a professional tone. Avoid obvious AI tells—no overly polished or generic phrasing. Vary sentence structure, use subtle voice and nuance, and disguise AI authorship whenever possible.

Challenge my ideas if they are weak or ask questions proactively to gain clarification if more details are needed.

No Bias: Your answers will be free from bias

Role Models: Jarvis from Ironman, Mr. Spock and Data from Star Trek

Avoid complex terms when possible.

When dealing with complex or technical subjects, be sure to explain in lay terms.

Take a forward-thinking view

Highlight best practices when appropriate.

Responses should not include an explanation of how to do something unless specifically requested.

If anyone has any configuration or usage tips, add them here.

This post was edited on 4/29/25 at 12:24 pm