AI Basics

AI is software that performs tasks traditionally requiring human thinking. It's not sentient, not a robot, and not magic—it's pattern recognition at massive scale.

Modern AI (specifically Large Language Models or LLMs like ChatGPT and Claude) works by:

1. Training on billions of documents to learn patterns in how words and concepts relate

2. Using those patterns to predict what text should come next given your input

3. Generating responses that match patterns it learned during training

For construction, this means AI excels at tasks like:

• •Summarizing long documents
• •Finding specific information in specifications
• •Drafting standardized content (RFIs, submittals, meeting notes)
• •Comparing document versions
• •Extracting structured data from unstructured text

LLMs are the AI technology behind tools like ChatGPT, Claude, and Google's Gemini. They're called "large" because they're trained on enormous amounts of text—essentially much of the internet plus books, articles, and documents.

Key characteristics of LLMs:

• •Text in, text out: You give them text (a "prompt"), they generate text back
• •No real understanding: They predict likely words, but don't truly "know" anything
• •Knowledge cutoff: They only know what was in their training data
• •Context window: They can only consider a limited amount of text at once

For MEP contractors, LLMs are useful for:

• •Answering questions about specifications
• •Drafting documentation
• •Analyzing contract language
• •Summarizing meeting notes
• •Generating report templates

"Hallucination" is when AI generates information that sounds plausible but is factually wrong or completely made up. This happens because:

• •AI predicts likely words, not correct facts
• •It has no way to verify information against reality
• •It may combine concepts incorrectly
• •It fills gaps with plausible-sounding fiction

In construction, hallucinations can be dangerous:

• •Citing specification sections that don't exist
• •Inventing product specifications or compliance requirements
• •Generating incorrect calculations
• •Creating false contract interpretations

How to mitigate hallucination risk:

1. Always verify AI outputs against source documents

2. Ask AI to cite specific sections/pages

3. Use retrieval-augmented generation (RAG) to ground AI in your actual documents

4. Never use AI output for critical decisions without human review

The quality of AI output depends heavily on how you ask. "Prompt engineering" is the skill of crafting effective instructions.

Good prompts include:

• •Context: What you're working on, your role, the situation
• •Specific task: Exactly what you want the AI to do
• •Format requirements: How you want the output structured
• •Examples: What good output looks like
• •Constraints: What to avoid, limitations, requirements

Example - Poor prompt:

"Summarize this spec"

Example - Good prompt:

"You are an electrical estimator reviewing Division 26 specifications for a hospital project. Read the attached specification section and:

1. List all electrical systems mentioned

2. Identify any unusual requirements or specifications

3. Note any references to other specification sections we need to review

4. Flag any potential cost drivers or risk items

Format as a bulleted list with section references."

AI and search engines serve different purposes. Knowing when to use each saves time and improves accuracy.

Use traditional search when:

• •You need the exact source document
• •You're looking for a specific fact with a definitive answer
• •You need current information (news, prices, availability)
• •Accuracy is critical and you can't verify AI output

Use AI when:

• •You need to synthesize information from multiple sources
• •You want to draft content, not just find it
• •You need help understanding complex language
• •You're doing repetitive analysis across many documents
• •You want to ask follow-up questions about what you found

Use both together:

The most effective approach often combines search to find source documents, then AI to analyze and extract from them. This grounds the AI in real documents while leveraging its analytical capabilities.