AI Landscape Quiz: Do You Know the Models That Matter?

A company wants to connect an AI assistant to its private internal knowledge base so it can answer employee questions. Which technique is most relevant?

Switching to an open-source model Fine-tuning the foundation model from scratch Retrieval-Augmented Generation (RAG) Building a new foundation model

A company builds a prompt library in Notion with 30 useful templates. After 3 months, some templates produce worse results than they used to. What is the most likely cause and solution?

The team is using too many templates — reduce to fewer than 10 Notion is not a suitable storage tool for prompt templates All templates expire after 3 months due to API policy AI model updates changed behavior — review and update templates when models are upgraded, and record a version history

A healthcare company is evaluating an AI vendor's product for processing patient records. The vendor's product is built on a closed-source foundation model API. What is the primary data privacy consideration?

Patient data will be transmitted to the foundation model provider's servers Closed-source models automatically comply with HIPAA The foundation model will permanently memorize patient data The application layer encrypts data before it leaves the company

A healthcare startup wants to use AI to analyze sensitive patient records. Sending data to external cloud APIs is not permitted under their compliance policy. What type of deployment is most appropriate?

Google Gemini Pro — all Gemini data stays in healthcare-dedicated servers ChatGPT Enterprise — data privacy is guaranteed by contract Whichever model has the highest benchmark score Self-hosted open-source model (e.g., Meta Llama) on the organization's own infrastructure

What is the biggest risk when an AI produces a confident-sounding analysis of your business?

AI analysis is too long — it always produces 10+ pages AI doesn't know your actual organizational context — it analyzes the concept you described, not the specific realities of your situation AI may accidentally leak your business data to competitors The AI may refuse to complete an analysis if the business is in a regulated industry

Which Gemini feature most directly reduces hallucination on questions about recent events?

Gemini's fine-tuning on Google's proprietary dataset Gemini's large context window Gemini Flash's lower latency Google Search grounding — responses can be anchored to live Search results

A small physical therapy clinic wants to use an AI scheduling chatbot on their website. Patients will input their names, contact info, and appointment reasons (which may reference medical conditions). What is required before deploying the chatbot?

Chatbots don't count as HIPAA-covered technology — only EHR systems do A Business Associate Agreement (BAA) with the chatbot provider, since the chatbot may process Protected Health Information under HIPAA HIPAA only applies to clinics with more than 20 employees A standard terms of service acceptance by the clinic is sufficient

Why can many different AI products (e.g., a writing tool, a customer service bot, a code assistant) feel similar in capability?

They may all be using the same underlying foundation model via API They all copy ChatGPT's training data Foundation models are open-source and freely copyable AI models can only do one type of task well

A small business wants to use AI for drafting responses to customer complaints that include customer names and order details. What is the simplest way to reduce privacy risk while still using AI?

Ask the AI not to store the customer's name after the session Anonymize the data — replace the customer name with 'Customer' and describe the order generically — before sending to the AI tool Use AI only on Tuesdays when server load is lower and data is more secure Never use AI for any customer-related communications

A sales team writes a different prompt every time they need to generate a follow-up email after a meeting. What is the highest-leverage improvement?

Each salesperson should develop their own unique prompting style Switch to a more powerful AI model Create a reusable prompt template with clearly marked variables for meeting date, client name, key discussion points, and next steps Only the most senior team members should use AI — it causes quality variance