⚡ IT Wisdom
Why did the DevOps engineer get divorced? He said "It's not you, it's the environment."
Skilled Practitioner
4249 XP
1751 to Expert
Manage Cards
Add manual cards or delete existing ones
All Cards
80 cards total
AITECH
Agentic AI
20 cards
·
20 basic
· 0 cloze
Basic
20 cards
| What is the key difference between Agentic AI and Generative AI? | claude | |
| What makes a system "autonomous" rather than just a prompt-response tool? | claude | |
| Why is agency better described as a spectrum than a binary label? | claude | |
| Why might forcing autonomy onto a task that doesn't need it (like a scheduled summary email) be... | claude | |
| What is "orchestration" in a multi-agent system? | claude | |
| Why should an agent's tools be narrowly scoped rather than broadly permissioned? | claude | |
| Why does an agent need a retry limit or escalation path instead of retrying forever? | claude | |
| What is the Model Context Protocol (MCP)? | claude | |
| What are the three main MCP primitives? | claude | |
| Why should tool/resource content from an MCP server be treated as untrusted input? | claude | |
| Why might exposing one broad tool (like "run_sql_query") be riskier than several narrow tools? | claude | |
| What is "human-in-the-loop" (HITL)? | claude | |
| What factor should determine whether an agent action needs HITL review? | claude | |
| What is "rubber-stamping" in a HITL context, and why is it a problem? | claude | |
| Why is confidence-based HITL routing only as reliable as the model's confidence calibration? | claude | |
| What is "data mapping" in an AI agent pipeline? | claude | |
| Why is silently dropping unrecognized fields during a mapping step risky? | claude | |
| Why should schema/type validation happen after a data transformation step? | claude | |
| Why is logging both the original and transformed value useful in a mapping pipeline? | claude | |
| Why might isolating failing records (dead-letter handling) be better than failing an entire batch? | claude |
Data Research and Analysis
10 cards
·
10 basic
· 0 cloze
Basic
10 cards
| What is exploratory data analysis (EDA)? | claude | |
| How does generative AI assist with EDA? | claude | |
| Why must an analyst still verify AI-generated EDA conclusions? | claude | |
| What are common automated data preparation tasks? | claude | |
| What is the risk of silently imputing missing values (e.g., filling with the column average)? | claude | |
| What is re-identification risk? | claude | |
| What control reduces sensitive data exposure when using AI for analysis? | claude | |
| Why does "correlation does not imply causation" matter for AI-assisted data analysis? | claude | |
| How can AI assist with research ideation? | claude | |
| What is "hallucination" in the context of AI-assisted research drafting? | claude |
Development and Workflow Automation
20 cards
·
20 basic
· 0 cloze
Basic
20 cards
| How does AI assist across the SDLC? | claude | |
| Why should AI-generated code always be reviewed before merging? | claude | |
| What is the main value of AI-assisted rapid prototyping? | claude | |
| Why is a polished AI-generated prototype risky if shipped directly to production? | claude | |
| What should you validate between steps in a multi-step AI workflow? | claude | |
| Why does an AI-powered workflow need ongoing monitoring, not just a pre-launch test? | claude | |
| How can a workflow balance human-in-the-loop safety with speed? | claude | |
| Why does sending unnecessary context in every AI prompt matter during prototyping? | claude | |
| What happens if a request exceeds a model's context window? | claude | |
| What is the risk of an AI assistant adding overly broad error handling (e.g., a bare except: pass)? | claude | |
| How can AI help with code documentation, and what's the catch? | claude | |
| Why is acceptance count (suggestions accepted) a flawed metric for AI-assisted code quality? | claude | |
| Why might an AI coding assistant suggest deprecated or outdated code patterns? | claude | |
| What is a key auditability practice for using AI throughout a regulated SDLC? | claude | |
| Why is reviewing AI auto-fixes (like lint fixes) important even for "minor" changes? | claude | |
| What false-positive risk comes with AI-augmented static analysis tools? | claude | |
| When does building a retrieval step (instead of sending full documents) become worth the added... | claude | |
| What is the tradeoff when trimming context to reduce token cost? | claude | |
| Why is using only velocity metrics (lines of code, PRs/week) risky when measuring AI's impact on... | claude | |
| What systemic risk arises when AI generates requirements, code, and tests with minimal human checkpoints? | claude |
Ethics and Security
10 cards
·
10 basic
· 0 cloze
Basic
10 cards
| What are the core responsible AI principles? | claude | |
| What is bias mitigation in AI? | claude | |
| What does accountability mean in responsible AI? | claude | |
| How can an organization protect data privacy when using a third-party AI model? | claude | |
| What is direct prompt injection? | claude | |
| What is indirect prompt injection? | claude | |
| Why is AI-generated misinformation considered especially dangerous? | claude | |
| What is AI governance? | claude | |
| What is an AI risk assessment? | claude | |
| Why does AI risk management need to be continuous, not just a pre-deployment step? | claude |
Generative AI Models
10 cards
·
10 basic
· 0 cloze
Basic
10 cards
| What is a Large Language Model (LLM)? | claude | |
| What is a diffusion model? | claude | |
| What is the main trade-off between cloud-hosted and locally-hosted AI models? | claude | |
| What is a context window? | claude | |
| What is a token in generative AI? | claude | |
| What should you check on a model hub before selecting a model for a task? | claude | |
| What does "multimodal" mean for a generative AI model? | claude | |
| What is Retrieval Augmented Generation (RAG)? | claude | |
| What is an embedding? | claude | |
| Why are vector databases used with RAG? | claude |
Prompt Engineering
10 cards
·
10 basic
· 0 cloze
Basic
10 cards
| What are the three core elements of a well-structured prompt? | claude | |
| What is few-shot prompting? | claude | |
| What is chained (sequential/iterative) prompting? | claude | |
| What is a prompt injection attack? | claude | |
| Direct vs. indirect prompt injection — what's the difference? | claude | |
| Why are agentic AI systems more exposed to prompt injection risk than a simple chatbot? | claude | |
| What is a hallucination in generative AI? | claude | |
| Name a basic defense against prompt injection from untrusted content. | claude | |
| How does RAG help mitigate hallucinations? | claude | |
| Why combine input/instruction separation with human-in-the-loop approval for an AI agent? | claude |
Add Card