Skip to main content
Skip to content
Menu

Chatgpt Intermediate 4 parts Updated

Build a fact-check script for high-stakes outputs

You'll have a working script that flags potential hallucinations in AI-generated text for medicine, law, or finance work before you share it with clients or stakeholders.

What you'll make

Fact-Check Script

  1. extract_claims() — pulls verifiable statements from AI text: what to include
  2. flag_suspicious() — identifies high-risk claim patterns: what to include
  3. route_verification() — assigns appropriate verification method: what to include
  4. generate_report() — outputs flagged claims with verification tasks: what to include

Paste your Python script with example output from running it on real AI text from your domain

The 4 parts

Each part ends with a check you can see.

  1. Map Your Domain's Weak Spots

    Find where AI consistently gets facts wrong in your specific field by testing it on known ground truth

    Done when

    You have 10+ examples of AI hallucinations in your domain, categorized by claim type (statistics, dates, procedures, regulations, etc.)

  2. Build the Claim Detection Engine

    Create a function that pulls verifiable statements from AI text and flags them by risk level

    Done when

    You have a working extract_claims() and flag_suspicious() function that processes real AI text and identifies high-risk statements by category

  3. Route Claims to Verification

    Build the workflow logic that assigns different verification methods based on claim type and risk level

    Done when

    You have documented verification steps for each claim category and a route_verification() function that assigns the right verification method to each flagged claim

  4. Ship the Complete Fact-Check Script

    Integrate all components into a working script and test it on real high-stakes AI output from your domain

    Done when

    You have a complete script that processes AI text end-to-end and generates verification reports you'd actually use before sharing outputs with clients

    Capstone: you put together your Fact-Check Script.

Where this helps

An elective: a single lesson that sits outside rtwo's job certificates and doesn't count toward one.

About this lesson

What you'll be able to do

  • Run a working fact-check script on real AI-generated text from your domain
  • Identify the 3-5 specific claim types where AI hallucinations cluster in your work
  • Have documented flag-and-verify workflows for high-stakes outputs before client delivery
  • Know when to trust AI output vs when to require human verification based on claim categories

Who it's for

  • Professionals in medicine, law, finance, or consulting who use AI to draft client-facing materials
  • Anyone who needs to verify AI claims before sharing outputs with external stakeholders
  • People comfortable with basic scripting or willing to copy-paste and modify code

What it won't cover

  • Building production-grade fact-checking infrastructure
  • Legal liability assessment for AI-generated content
  • Training custom models for domain-specific verification
  • Replacing human expert review — this augments verification, doesn't replace it

Skills you'll practise

  • Claim extraction from AI text
  • Systematic hallucination pattern recognition
  • Risk-based verification workflow design
  • AI output reliability calibration