Introduction to AI & ML
Overview
Before you can use AI tools well, you need to understand how they actually work and where they reliably fail. LLMs predict the next token based on statistical patterns in training data. They do not reason, retrieve facts, or know when they are wrong. Hallucination is not a bug to be fixed - it is intrinsic to the architecture. Designers who understand this use AI as a powerful accelerant. Those who do not will ship flawed research, broken audit reports, and AI-generated copy that contradicts itself. This section gives you the critical foundation.
AI workflow
6 steps
- 01
Trace the AI Timeline
Self-directed reading
From Turing (1950) and Shannon through the perceptron, symbolic AI winter, deep learning revival (Hinton 2006), ImageNet (2012), Transformers (2017), GPT-3 (2020), to the current LLM era. Context matters for evaluating AI capability claims.
- 02
Understand LLM Architecture
Study + Claude Q&A
LLMs are next-token predictors. They generate text by calculating probability distributions over a vocabulary. Ask Claude to explain its own architecture - the self-awareness is imperfect but instructive.
- 03
Map the Hallucination Risk
Claude
Prompt: "List 5 scenarios where you would hallucinate confidently in a UX research context." Use this output to build a personal hallucination risk checklist for your workflow.
- 04
Learn GAN and Diffusion Models
Study
Generative image models (Midjourney, DALL-E, Stable Diffusion) are Diffusion-based, not GANs. Understanding this helps you evaluate synthetic user research images and AI-generated personas critically.
- 05
Test the Limits of Free Tiers
Claude / ChatGPT / Gemini
Spend one hour deliberately trying to break each major model. Ask for citations (note hallucinations), ask for recent events (note training cutoffs), ask for competitor pricing (note confabulation). Document your findings.
- 06
Build Your AI Evaluation Framework
Notion / Claude
Create a personal scoring rubric: accuracy, citation reliability, context retention, instruction following, hallucination tendency. Apply it every time you adopt a new tool in your workflow.
Tools
| Tool | Use | Free tier | Client-safe |
|---|---|---|---|
| Claude | Conceptual Q&A, architecture explanation, hallucination mapping | Free tier available | Yes |
| ChatGPT | Comparative testing, cross-model benchmarking | Free tier available | Yes |
| Gemini | Google ecosystem AI, strong with Google Workspace integration | Free tier available | Yes |
| Perplexity | AI with cited sources - use to validate LLM claims | Free tier available | Yes |
Constraints
3 constraints
- Honest Assessment: No AI tool is reliable for factual lookups without source verification. Always treat LLM output as a strong first draft, not a finished artefact.
- Hallucination risk is highest in: specific statistics, competitor data, academic citations, recent events, and niche domain expertise.
- Suitable for all project types and seniority levels. No client data involved.