Essential Terms

AI Glossary

Plain-English definitions for the most common AI terms you'll encounter

LLM
Large Language Model
A type of AI trained on vast amounts of text to understand and generate human-like language. Think of it as a very well-read assistant that can write, summarise, and answer questions.
TKN
Tokens (Input & Output)
The "units" AI reads and writes in — roughly one word or part of a word. Input tokens are what you send; output tokens are what you get back. AI pricing is based on these.
CTX
Context Window
The maximum amount of text an AI can consider at once — its working memory. A larger context window means it can process longer documents in a single go.
1M tokens ≈ 750k words ≈ Harry Potter: Goblet of Fire + Order of the Phoenix + 2–3 earlier books. Or 1.3 copies of War and Peace, or roughly 1 Bible.
PRO
Prompt
The instruction or question you give to an AI. A well-written prompt leads to better answers — just like giving clear directions to a colleague.
Advanced Terms

Beyond the Basics

Key concepts you'll hear when evaluating AI solutions for your business

HAL
Hallucination
When AI confidently generates information that sounds plausible but is factually wrong. Like a colleague who guesses an answer rather than saying "I don't know."
RAG
RAG (Retrieval-Augmented Generation)
A technique that lets AI look up real information before answering, rather than relying only on its training. Reduces hallucinations and keeps answers current. Core component of Aiske
API
API (Application Programming Interface)
The connection point that lets your software talk to an AI service automatically. Instead of using a chatbot manually, your app sends requests and gets responses.
FIT
Fine-Tuning
Customising a general AI model with your own data so it becomes an expert in your specific domain — like training a new hire on your company's processes. Primary use of Aiske.ai Claude Skills
✨ Key Insight: RAG + Fine-Tuning are the two main ways to make a general AI model work specifically for Aiske.ai.
AI Governance

AI Governance Rules

Policies and guidelines for responsible AI use at Aiske.ai

D
Data & Confidentiality
  • Never input customer, personal, or financial data into unapproved AI tools
  • Never upload source code, architecture diagrams, or credentials to unapproved tools
  • Assume prompts are logged unless explicitly stated otherwise
T
Tool Approval & Usage
  • This is a work in progress — tools we are currently trialing may not be approved
  • Only use AI tools approved by Aiske.ai for work-related tasks
  • Browser extensions and AI plugins require approval before use
  • Free tiers are prohibited for company work unless explicitly approved
S
Security & Access Control
  • Never share API keys, tokens, or passwords with AI tools
  • Never connect AI tools to production systems without security review
  • Use least-privilege access when integrating AI with internal systems
A
Accuracy, Risk & Decision-Making
  • AI outputs must not be the sole basis for customer, financial, or legal decisions
  • AI content must be human-reviewed before external use
  • Treat AI as a junior assistant, not an authority
  • You are 100% responsible for every piece of AI-generated content you use
⚠️ Aiske.ai AI Governance Framework — These rules apply to all team members. When in doubt, ask before using an AI tool for company work.
AI Course

Ethics, Transparency & Core AI Principles

Anthropic's AI Fluency Framework — required viewing for all Aiske.ai team members

Required Course

Anthropic AI Fluency Framework

Foundations — Ethics, transparency, vendor awareness, and core AI principles

Open Course ↗
anthropic.skilljar.com/ai-fluency-framework-foundations
💡 This course covers ethics, transparency, vendor awareness, and core AI principles. All Aiske.ai team members should complete it as part of AI onboarding.
⚠️ Important

AI Slop Is the New Spam

You are 100% responsible for every piece of AI-generated content you use

❓
What is AI Slop?
Low-quality, unreviewed AI output — generic text, hallucinated facts, and filler that erodes trust.
🏢
Why It Matters at Aiske.ai
  • You own it — if you publish or send AI content, it's yours. No blaming the AI.
  • Customers trust us — sloppy AI content damages our reputation with farmers who rely on us.
👁️
People Can Tell
  • It's obvious — colleagues and managers can easily spot unedited AI content from its generic tone and filler phrases.
  • Errors stand out — wrong facts, made-up details, and inconsistencies will be noticed and reflect poorly on you.
✅
Your Responsibility
📖
Read every word
Never copy-paste AI output without reviewing it thoroughly
🔍
Verify facts
Check claims, links, and data against real sources
✍️
Make it yours
Rewrite in your voice, add genuine insight, and remove generic filler
Model Landscape

LLM Models — The Leading AI Models

The most popular and capable large language models powering AI in 2026

Proprietary Models
G
GPT-5 / o4 Series Approved
OpenAI
Most consistent all-purpose model; excels in coding, reasoning, and long-context tasks
C
Claude 4 Sonnet / Opus Approved
Anthropic
Deep analysis and multi-step reasoning; strong for long documents and agentic coding
G
Gemini 3 Pro / 2.5 Flash Approved
Google DeepMind
#1 on LM Arena; advanced reasoning with Deep Think mode and multimodal capabilities
X
Grok 4.1 Not Approved
xAI
Top reasoning and lowest hallucination rate (~4%); real-time data integration via X
Open-Source / Open-Weight Models
M
Llama 4 Not Approved
Meta
Most popular open LLM family; Scout offers a massive 10M token context window; multimodal MoE architecture
D
DeepSeek R1 / V3 Not Approved
Chinese AI company
State-of-the-art reasoning at ultra-low cost ($0.07/M tokens); chain-of-thought and self-verification
Q
Qwen 3 (235B MoE) Not Approved
Alibaba
Strong reasoning, code, and multilingual support; 131K context with YaRN; Apache 2.0 license
✨ Key Trends in 2026: Massive context windows (up to 10M tokens), Mixture-of-Experts architectures for efficiency, and built-in reasoning and agentic capabilities.
Claude Lineup — Feb 2026

Claude Models — Which One, Where & Why

Anthropic's three model tiers: match the right Claude to your task for optimal performance & cost

O
Opus 4.6
Flagship
Most intelligent model; 1M context window; agent teams; adaptive thinking; best for coding & deep reasoning
Price: $5 / $25 per MTok
Use When: Complex multi-step reasoning, large codebase analysis, financial & legal workflows, agentic automation, architecture decisions
S
Sonnet 4.5
Balanced
Best coding model globally (77.2% SWE-bench); ideal daily driver for 90% of tasks; 200K context (1M beta)
Price: $3 / $15 per MTok
Use When: Day-to-day coding, content generation, customer-facing apps, document analysis, enterprise copilots
H
Haiku 4.5
Fast & Lean
2x Sonnet speed at 1/3 cost; near-frontier intelligence; first Haiku with extended thinking; high-volume ready
Price: $1 / $5 per MTok
Use When: High-volume triage, real-time chat, quick edits, routing & classification, cost-sensitive pipelines
💡 Why Claude for Aiske.ai? Constitutional AI (industry-leading safety & alignment), up to 1M token context for entire codebases, and cascading model tiers to optimise cost & quality.
✨ Pro tip: Route simple queries to Haiku, everyday work to Sonnet, and only escalate to Opus for complex reasoning — saves up to 95% on costs.
GPT-5 Lineup — Feb 2026

ChatGPT Models — Which One, Where & Why

OpenAI's GPT-5 family: match the right model to your task for optimal performance & cost

G5
GPT-5
Flagship
Most capable model; 400K context; adaptive reasoning; agentic coding with Codex
Price: $1.25 / $10 per MTok
Use When: Complex multi-step reasoning, agentic coding workflows, large codebase analysis, production-grade apps
5m
GPT-5 Mini
Balanced
Cost-efficient daily driver; great for well-defined tasks & precise prompts; ideal for chatbots
Price: $0.25 / $2 per MTok
Use When: Day-to-day coding, content generation, customer-facing chatbots, document analysis, enterprise copilots
5n
GPT-5 Nano
Fast & Lean
Fastest & cheapest GPT-5 variant; 400K context; great for summarisation & classification
Price: $0.05 / $0.40 per MTok
Use When: High-volume triage, real-time chat, routing & classification, summarisation pipelines, cost-sensitive workloads
💡 Why ChatGPT for Aiske.ai? Richest AI ecosystem — Codex, search, image gen & function calling. Cascading model tiers with 400K context to optimise cost & quality.
✨ Pro tip: Route simple queries to Nano, everyday work to Mini, and only escalate to GPT-5 for complex reasoning — saves up to 96% on costs.
Trialing

Claude Desktop — Chat, Cowork & Code

Three modes for every type of AI work — from quick questions to autonomous task completion

C
Chat
Conversational AI
Quick questions, writing, analysis, brainstorming. Web search, image understanding, memory across chats.
W
Cowork
Agentic Knowledge Work
Multi-step tasks on local files: organise folders, draft reports, process receipts. Runs in isolated VM with folder-level access control.
{/}
Code
Developer Terminal in Desktop
Full Claude Code agentic coding in a GUI. Edit files, run terminal commands, manage git — no CLI needed.
How It Compares to ChatGPT Desktop
1
Agentic File Access
Cowork reads, edits & creates local files autonomously. ChatGPT can upload files for analysis but cannot modify files on your computer.
2
Autonomous Task Execution
Claude plans, breaks into subtasks & executes while you step away. ChatGPT requires turn-by-turn prompting with Canvas for collaborative editing.
3
Built-In Coding Agent
Code tab provides full terminal access inside the desktop app. ChatGPT offers Codex as a separate cloud tool, not a native desktop mode.
💡 Three tools in one app — Pro, Max, Team & Enterprise — macOS & Windows — MCP connectors & Chrome integration.
Your Permissions, Your Boundaries

Connector Access Levels

Claude can only access what you can access — your existing application permissions carry through every connector

You → Your credentials
Claude via MCP Connectors → Inherits your access
IntercomSame access as your Intercom account
JiraSame access as your Jira account
SlackSame access as your Slack account
GitHubSame access as your GitHub account
=
Permission Passthrough
Claude authenticates as you. If you cannot see a file, neither can Claude.
P
No Privilege Escalation
Connectors cannot grant broader access than your application role allows.
A
Audit-Friendly
All actions logged under your account with full traceability and zero shadow access.
💡 Same user, same permissions — OAuth 2.0 auth — Enterprise SSO — Per-user audit trail. No elevated privileges, no admin backdoors, no data leakage.
AI Inside Your Tools

Claude Plugins for Desktop Apps

Claude integrates directly into Microsoft Office and other desktop applications via add-in connectors

📊
PowerPoint
Build slides, format decks, create charts
📈
Excel
Analyze data, write formulas, create reports
📝
Word
Draft documents, edit text, format content
🔧
More Apps
Expanding to more desktop tools over time
How It Works
1
Install the Add-in
Enable the Claude connector from your Office add-ins panel
2
Connect Your Account
Authenticate with your Claude account via secure OAuth
3
Start Working
Claude assists directly inside your document in real time
💡
Why You Sometimes Need a New Conversation

Each conversation has a context window — a limit on how much information Claude can hold at once. As you work through edits and changes, the conversation fills up with prior tool calls and results.

When the window is full, Claude may lose track of earlier details or become less accurate. Starting fresh gives Claude a clean slate.

Start a new conversation when:

  • Claude starts repeating or forgetting edits
  • You are switching to a different task or topic
  • Responses slow down or become less precise
  • You have made many edits in a long session
💡 Works inside your apps — no data leaves your environment — new conversation = fresh context.
Getting Started

Claude Code — Using Code Mode in Desktop

From opening a project to autonomous coding — a step-by-step guide to Claude Code in Desktop

1
Open Your Project
Select the Code tab, choose your project folder, and pick your environment (local, SSH, or cloud)
2
Set Permissions
Choose a permission mode: Approve All (safest), Auto-Accept Reads, or Full Auto for trusted projects
3
Give Instructions
Describe your task in plain English. Claude reads the codebase, plans the approach, and starts coding
Key Capabilities
D
Visual Diff Review
Inline Changes
See every change as inline diffs with comments. Accept, reject, or request modifications before committing.
P
Parallel Sessions
Git Isolation
Run multiple sessions with automatic Git worktree isolation. Each session works on its own branch.
C
Connectors
Integrate Tools
Integrate GitHub, Slack, Linear, and more. Create PRs, respond to issues, and update project tools.
💡 Pro, Max, Team & Enterprise — Local, SSH & Cloud — Visual diffs & Git isolation — GitHub, Slack & Linear connectors.
Building Blocks

Claude Code — Core Concepts

The key building blocks that make Claude Code a powerful agentic development environment

M
CLAUDE.md — Project Memory
The default project reference file at the repo root. Contains architecture docs, conventions, build commands, and rules that Claude always knows.
S
Sub-agents — Parallel Workers
Specialized agents spawned via the Task tool to work in parallel with isolated contexts. Built-in types: Explore, Plan, and custom agents you define.
K
Skills — Auto-Activated Knowledge
Markdown files (SKILL.md) with YAML frontmatter that auto-activate when their description matches the current task. Teach Claude how to do specific things.
H
Hooks — Lifecycle Automation
Shell scripts tied to lifecycle events: PreToolUse, PostToolUse, Stop, SubagentStop. Enforce rules and automate actions deterministically.
C
Slash Commands
Saved prompts triggered by /command in the terminal. Project-level or global scope. Great for repeatable workflows.
P
Plugins — Shareable Bundles
Bundles of commands, agents, skills, and hooks distributed via GitHub repos. Install from marketplaces or share your setup.
A
MCP — External Connections
Model Context Protocol connects Claude Code to external tools: GitHub, Jira, Figma, databases, and APIs via standardized server connections.
✨ How They Fit Together: CLAUDE.md sets context → Skills teach behavior → Hooks enforce rules → Sub-agents execute in parallel → Plugins package it all for sharing.
Trialing

Claude.ai Artifacts — Overview

Interactive, shareable content generated in a dedicated window alongside your conversation

A
What Are Artifacts?
Standalone, interactive outputs generated from your Claude conversations. Displayed in a dedicated panel separate from the chat for easy editing, iteration, and reuse.
T
Artifact Types
Six core types: code snippets, Markdown documents, single-page HTML websites, SVG vector graphics, Mermaid diagrams (flowcharts, ERDs), and interactive React components.
C
When Claude Creates One
Content is self-contained and substantial (typically over 15 lines), likely to be edited or reused, and stands alone without needing extra conversation context.
I
Iterative Workflow
Ask Claude to modify artifacts with follow-up prompts. Each update creates a new version you can toggle between. Highlight specific elements for targeted edits.
S
Sharing & Publishing
Publish artifacts to generate a shareable link. Others can view, interact, or remix your creation. Sharing is free; usage counts against each viewer's own limits.
P
AI-Powered Artifacts
Build artifacts with Claude's intelligence embedded. Create chatbots, tutors, analysis tools, and interactive apps — no API keys or deployment needed.
M
MCP Integration
Artifacts connect to external services via Model Context Protocol. Access tools like Asana, Google Calendar, and Slack directly from within an artifact.
✨ Key Concept: Artifacts turn Claude from a chat assistant into a collaborative workspace — describe what you need and get shareable, interactive apps without writing code.