Half the reason AI feels over your head is nobody defined the words in plain English. Here they are. Don't memorize this — recognize it. Keep it open while you work, and the illusion of complexity collapses.
| Word | What it actually means | Why you care |
|---|---|---|
| Model | The "brain" — Opus, Sonnet, or Haiku. Each is a different version of Claude. | Picking the right one saves money and time. |
| Token | The chunks of words AI reads and writes. Roughly 1 word ≈ 1.3 tokens. | Tokens = your usage. Long prompts burn credits faster. |
| Context Window | How much Claude can "see" at once — its short-term memory in one chat. | Hit the limit and it forgets the beginning. Start a fresh chat. |
| Knowledge Cutoff | The date the AI stopped learning. It doesn't know what happened after. | For "today's" info, use search or Perplexity. |
| System Prompt | The hidden instructions that shape how Claude behaves before you say a word. | This is what skills and projects edit — the "personality layer." |
| Multi-Modal | The AI can handle text, images, voice, and PDFs — not just text. | Drop in a screenshot, a PDF, a voice note. Use it. |
| Word | What it means | The fix |
|---|---|---|
| Hallucination | When AI confidently makes something up that isn't true — fake stats, fake citations. | Tell it: "Don't guess. Say 'I don't know' and flag with [VERIFY]." |
| Sycophancy | When AI agrees with you when it shouldn't. The yes-man trap. | Tell it: "Don't tell me what I want to hear — tell me what a smart skeptic would say." |
| Bias | When AI gives a skewed answer based on its training data. | Tell it: "Argue against yourself. Make the strongest case for the opposite." |
| Word | What it means | Why you care |
|---|---|---|
| Memory | What Claude remembers about YOU across chats. (Different from context window.) | Curate it. Bad memories = bad outputs. |
| Artifact | A clean output Claude builds in a side panel — doc, code, dashboard, design. | Can be shared, edited, re-used. |
| Project | A workspace with its own files, instructions, and chat history. | Rung 2 of the ladder. Standing context. |
| Connector | A bridge between Claude and another app — Gmail, Drive, Notion, QuickBooks. | Rung 3. Claude gets hands. |
| Skill | A custom reusable behavior you trigger by name. | Rung 4. Reusable procedures. |
| Cowork | Claude with agentic capability — multi-step work autonomously. | Rung 5. The top of the ladder. |
| Thinking | When Claude shows its reasoning before answering. Slower but smarter. | Toggle on for hard questions. |
| Deep Research | Claude searches the web for 10–30 min and returns a report. | When you'd otherwise pay a research analyst. |
| Word | What it means |
|---|---|
| Agent vs Chatbot | A chatbot drafts the email. An agent drafts it, sends it, and follows up in 3 days. The difference is whether AI can take action. |
| Vibe Coding | Building software by describing what you want in plain English. (Lovable, Replit, Bolt.) |
| The 4 Ds | Anthropic's AI-fluency framework: Delegation, Description, Discernment, Diligence. |
| The 3 Modes | Automation (AI does it), Augmentation (AI thinks with you), Agency (AI runs on its own). Live in Augmentation. |