ChatGPT Basics

Understanding ChatGPT versions, limitations, and how to get better results

For anyone using ChatGPT — here's what you need to know about different versions, their limitations, and how to get the most reliable answers.

ChatGPT Version Comparison

ChatGPT-3.5 (Free)

Trained on data before Jan 2022

📅 Limited to early 2022

ChatGPT-4 (Paid)

More recent training + web access

🌐 Current Info

Key Takeaway

ChatGPT-3.5 is trained only on data before January 2022 and will "guess" about current events. For up-to-date information, you need a paid OpenAI account to access GPT-4 with web browsing capabilities.

🧠 What "Trained on Data" Actually Means

Understanding how Large Language Models (LLMs) work helps you use them more effectively:

  • Pattern recognition, not knowledge
    (The AI learns patterns from text but doesn't truly "know" facts like humans do.)
  • Statistical predictions
    (It predicts what words should come next based on patterns it has seen before.)
  • No real-time updates
    (Once trained, the model is "frozen" and cannot learn new information.)
  • Quality depends on training data
    (The accuracy and usefulness depend on the quality and recency of the data it was trained on.)
Important: Think of it like having read millions of books up to a certain date, then never being able to read anything new.

📅 ChatGPT-3.5 Training Cutoff: Before January 2022

Critical Limitation: ChatGPT-3.5 (the free version) was trained on data only before January 2022. This means:

  • No knowledge of events after early 2022
    (It doesn't know about recent news, current events, or developments since early 2022.)
  • Outdated information on many topics
    (Technology, politics, science, and other rapidly changing fields may have outdated information.)
  • Will "guess" about current topics
    (When asked about recent events, it may provide plausible-sounding but incorrect information.)
  • Limited awareness of new products/services
    (It won't know about new software, apps, or services released after early 2022.)
Important: Always verify information about current events, recent developments, or anything that might have changed since early 2022.

Microsoft Documentation and ChatGPT-3.5: Critical Limitations

Important Fact: ChatGPT-3.5 has NO access to official Microsoft Azure documentation. This is not a gap in knowledge - it's a complete absence of authoritative Microsoft documentation.

Why This Matters: Many users assume ChatGPT knows everything about Microsoft products, but ChatGPT-3.5 has zero knowledge of official Microsoft documentation due to licensing and partnership timing.

Think of it like asking someone who's never read the official instruction manual for a complex machine - they might have heard about it from others, but they don't have the authoritative information.

What ChatGPT-3.5 Was Trained On

  • Public websites (Wikipedia, StackOverflow, Reddit)
  • Publicly accessible programming documentation
  • Open-source documentation (not behind logins)
  • General discussions about technology
  • No proprietary or subscription-only content

What ChatGPT-3.5 Was NOT Trained On

  • Official Microsoft Azure documentation (learn.microsoft.com)
  • Microsoft Purview official guides
  • Any Microsoft documentation due to licensing restrictions
  • Gated or subscription-only content
  • Current Microsoft service updates
  • Proprietary Microsoft resources

What This Means for Microsoft Technologies

No Official Knowledge: ChatGPT-3.5 has zero access to official Microsoft documentation. Any knowledge comes from public discussions, not authoritative Microsoft sources.

Key Point: ChatGPT-3.5 will provide plausible-sounding but incorrect information about Microsoft services because it has no access to official documentation.

Risks When Using ChatGPT-3.5 for Microsoft Questions

Note: These limitations apply specifically to ChatGPT-3.5. ChatGPT-4 with web browsing can access current Microsoft documentation.

1. No Official Documentation Access

ChatGPT-3.5 has never seen official Microsoft documentation due to licensing restrictions

🚫 Complete absence
2. Hallucinated Information

Will make up plausible-sounding but incorrect Azure commands, features, and procedures

Fabricated details
3. Public Discussion Knowledge Only

Any knowledge comes from public forums, not authoritative Microsoft sources

💬 Unreliable sources

Bottom Line: Always verify Microsoft-specific answers against official documentation at learn.microsoft.com, especially for Azure, Purview, or other Microsoft services.

The Solution: Register for a Paid Account with Access to ChatGPT-4

For Access to Current Microsoft Information: The limitations above explain exactly why you need a registered OpenAI account to access current Microsoft documentation. However, having access doesn't automatically guarantee trustworthy responses - you still need to ask the right questions and critically interpret the results.

Key Point: ChatGPT-4 with web browsing can access current Microsoft documentation live, while ChatGPT-3.5 (free version) cannot. But remember: access to current information is just the first step - you still need to know how to ask the right questions and interpret the results.

1. Free ChatGPT-3.5

No access to current Microsoft docs, may provide outdated or incorrect information

Not trustworthy
2. Paid ChatGPT-4

Can access current Microsoft documentation through web browsing

🌐 Access to current docs
3. Registered Account Required

You must be logged in with a paid account to access ChatGPT-4 and web browsing

💳 £20/month investment

Bottom Line: If you need access to current Microsoft documentation (like the Purview example below), you must use a registered account with ChatGPT-4 and web browsing enabled. The free ChatGPT-3.5 version simply cannot access current Microsoft documentation. However, getting trustworthy answers still requires knowing how to ask the right questions and critically evaluating the responses.

Real Example: Why ChatGPT-3.5 Responses Can't Be Trusted

Problem: Here's an actual question asked to ChatGPT-3.5 (when logged out) and its response, demonstrating why the results cannot be trusted.

Question Asked: "In a Microsoft Purview search, if it says a file was accessed, what does that mean?"

ChatGPT-3.5's Response (Why It's Wrong)

Note: This response was generated by ChatGPT-3.5, which was trained before January 2022. Microsoft Purview was still relatively new at that time.

❌ ChatGPT-3.5's Incorrect Response:

"In Microsoft Purview, if a file is shown as 'accessed' in an audit or content search, it typically means that someone interacted with the file in a way that reads or opens it. Specifically, 'accessed' can refer to:

Common actions that count as 'access':
• Opening a file in SharePoint, OneDrive, or Teams (e.g., viewing a Word document or Excel file).
• Previewing a file in the browser or within Microsoft 365.
• Downloading the file, which triggers a read action.
• Syncing a file via OneDrive (automatic or manual that reads file contents).
• Using the file in an app (e.g., embedded file in a Teams message that is viewed).
• Sharing a link and then someone using that link to view the file.

Actions that do not necessarily count as "access":
• Merely seeing the file listed in a folder (no preview or open).
• Hovering over a link or file name.
• Being granted permissions (but not using them).
• Indexing by Microsoft search (sometimes excluded, but can be context-dependent).

Why This Response Cannot Be Trusted: ChatGPT-3.5 is providing information based on its pre-2022 training, which doesn't reflect current understanding of Microsoft Purview. It's essentially "guessing" about a topic it doesn't have current knowledge about.

1. Outdated Information

ChatGPT-3.5 was trained before Microsoft Purview was widely understood and documented

📅 Pre-2022 knowledge
2. Incorrect Assumptions

It assumes "accessed" in Purview reports means user interaction, when Purview often shows "accessed" for system processes like indexing or scanning

Wrong interpretation
3. Missing Context

Doesn't distinguish between user actions and automated system processes

⚠️ Incomplete picture

Understanding Hallucinations

Why ChatGPT-3.5 "guesses" and how this affects the reliability of its answers.

What Happens Why It Happens How to Spot It
ChatGPT makes up information It doesn't have current data, so it predicts what sounds plausible Check dates and verify facts
Provides outdated information Training cutoff means it only knows pre-January 2022 information Look for recent developments
Confidently states incorrect facts It doesn't know what it doesn't know Cross-reference with reliable sources
Mixes old and new information Patterns from old data combined with guessing Verify each claim separately

Key Point: ChatGPT-3.5 is essentially "guessing" when it doesn't have current information, which is why verification is crucial.

The Importance of Well-Structured Prompts

How to design better prompts and questions to get high-quality, trustworthy answers from ChatGPT.

💡 What is a "Prompt"?

A "prompt" is simply the question or instruction you give to ChatGPT. Think of it like giving directions to a very knowledgeable but literal person who only knows what you tell them.

Good prompts are: Specific, detailed, and include context about what you're trying to achieve.

Bad prompts are: Vague, too broad, or don't provide enough information for ChatGPT to give you a useful answer.

  • Be specific and detailed – Include context, requirements, and constraints
  • Ask for step-by-step explanations – Request structured responses with clear sections
  • Specify the format you want – Ask for lists, tables, code, or specific structures
  • Include examples – Show what kind of response you're looking for
  • Ask for verification – Request sources or ask it to explain its reasoning

Avoiding Poor Prompts

  • Vague questions – "Tell me about technology" is too broad
  • Leading questions – Don't suggest the answer you want
  • Complex multi-part questions – Break them down into smaller parts
  • Assuming current knowledge – Remember the training cutoff
  • Not specifying context – Provide relevant background information

Example: Good vs Poor Prompts

Poor: "How do I cook chicken?"

Better: "I'm a beginner cook with a gas stove and need to make dinner for 4 people. What's the safest way to cook chicken breasts so they're not dry? I have basic kitchen equipment and about 30 minutes. Please give me step-by-step instructions."

Poor: "What should I do about my car?"

Better: "My 2018 Honda Civic is making a rattling noise when I accelerate, and the check engine light came on yesterday. I'm not very car-savvy and need to know if this is urgent or if I can wait until next week to get it checked. What are the most likely causes and what should I do first?"

Best Practices for Reliable Results

Follow these guidelines to get the most trustworthy answers from ChatGPT.

Glossary of ChatGPT Terms

Key terms to understand when working with ChatGPT

Training Cutoff

The date when the AI model's training data ends. ChatGPT-3.5's cutoff is before January 2022, meaning it has no knowledge of events or information after that date.

Hallucination

When an AI generates information that sounds plausible but is actually incorrect or made up. Common when the AI doesn't have current information.

Web Browsing

A ChatGPT-4 feature that allows the AI to search the internet and access current information, overcoming the training cutoff limitation.

Prompt Engineering

The practice of designing effective prompts and questions to get better, more reliable responses from AI language models.

Large Language Model (LLM)

An AI system trained on massive amounts of text data to understand and generate human language. ChatGPT is an example of an LLM.