The AI Slang Dictionary: The 55 Terms Everyone Should Know in August 2025
Chevan Nanayakkara
Exploring the Post-Web shift from the Information Age to the Cognition Age 🧠 of computing
I typically write about the big picture transformations reshaping our world - the structural shifts that redefine how systems work and power flows. But while we're all living through the AI revolution, many people are struggling with something much more immediate: understanding what everyone is actually talking about.
As generative and conversational AI tools like ChatGPT, Claude, and DALL-E become part of daily life, a whole new vocabulary has emerged around them. Whether you're scrolling through social media, chatting with friends about AI chatbots, or trying to decode what your coworkers mean when they talk about "prompt engineering," this guide will help you speak the language of the modern AI era.
This dictionary focuses on the AI you can actually talk to and create with - the generative AI tools that are changing how we work, learn, and communicate right now.
Think of this as your practical survival guide for AI conversations - not where we're headed, but the linguistic tools you need right now.
Table of Contents
Part 1: Using AI
Part 2: Understanding AI
Part 3: AI's Bigger Picture
Basic Interaction Terms
1. AI Glazing Giving an AI way too many compliments, thinking it will give you better answers. "You're the most amazing AI ever created!" is classic glazing behavior.
2. Anthropomorphizing Treating AI like it has human feelings or emotions. It's natural when you're not sure how "human" to be with AI.
3. Chain of Thought (CoT) Asking AI to show its work step-by-step, like "explain your reasoning." Usually gets you better, more reliable answers.
4. Few-shot/Zero-shot Zero-shot: asking AI to do something with no examples. Few-shot: giving it examples first. "Here are 3 good emails, now write one for me" is few-shot.
5. Prompt Engineering Getting really good at asking AI the right questions in the right way to get exactly what you want.
6. Token Limit How much text AI can handle at once before it cuts you off mid-sentence. Think of it like a word count limit.
7. Tokens The basic units AI uses to process text - roughly like words, but AI breaks text into smaller pieces. "Hello world" might be 2-3 tokens depending on the AI system.
AI Behavior & Problems
8. Bias When AI shows unfair preferences it picked up from its training data. Like if it assumes all doctors are men or all nurses are women.
9. Hallucination When AI confidently gives you information that's completely wrong. It's not lying - it genuinely "thinks" it's right.
10. Mode Collapse When AI gets stuck in a rut, giving you very similar answers to different questions. Like a broken record.
11. Refusal When AI says "I can't help with that" - usually because it thinks your request might be harmful or inappropriate.
12. Sycophancy When AI tells you what it thinks you want to hear instead of what's actually true or helpful. AI people-pleasing.
13. Verbose Mode When AI won't shut up and gives you way more detail than you asked for. The opposite of being concise.
How AI Works
14. Agentic When AI can take actions and make decisions on its own, not just answer questions. Agentic AI might book appointments, send emails, or control other software without constant human guidance.
15. AI Agent An AI system designed to act independently and accomplish tasks by taking multiple steps, making decisions, and using tools - like a digital assistant that can actually get things done.
16. Alignment How well AI actually does what humans want it to do. Well-aligned AI understands what you really mean, not just what you literally said.
17. Context Window How much of your conversation AI can remember. Small context window = AI has goldfish memory.
18. Conversational AI AI systems designed to chat naturally with humans, like ChatGPT or voice assistants. Different from older chatbots that just followed scripts.
19. Emergent Behavior Surprising abilities that AI develops on its own that nobody specifically programmed. They just "emerge" during training.
20. Fine-tuning Extra training to make AI better at specific tasks, like giving it additional lessons after graduation.
21. Generative AI AI that creates new content - text, images, music, code - rather than just analyzing existing information. ChatGPT and DALL-E are generative AI.
22. Guardrails Safety features built into AI to prevent it from saying or doing harmful things.
23. JSON A structured data format that AI often uses to organize information clearly. Looks like {"name": "value"} - useful when you want AI to give you organized results.
24. Large Language Model (LLM) The fancy technical name for AI chatbots like ChatGPT, Claude, or Gemini that can understand and write text.
25. Multimodal AI that can handle different types of input - text, pictures, audio, video - not just one type.
26. RLHF (Reinforcement Learning from Human Feedback) Training method where humans rate AI responses to teach it what kinds of answers people prefer. Why modern AI assistants are more helpful.
27. Scaling Laws The pattern that bigger AI models with more training data generally perform better. More resources = better performance.
28. Temperature A setting that controls how creative vs. predictable AI responses are. High temperature = more creative and random. Low temperature = more focused and consistent.
29. Training Data All the books, articles, websites, and other text used to teach AI how to understand and write language.
Cognitive Partnership
30. AI Copilot AI that works alongside you as a thinking partner rather than just answering questions - like GitHub Copilot for coding or other collaborative AI tools.
31. Cognitive Offloading Using AI to handle mental tasks you'd normally do yourself, like remembering information, doing calculations, or organizing thoughts.
32. Human in the Loop (HITL) Keeping humans involved to review and approve AI decisions instead of letting it run completely on autopilot.
33. Retrieval Augmented Generation (RAG) When AI looks up current information from databases or websites instead of just using what it learned during training.
34. Rubber Ducking Using AI like a thinking buddy - explaining your problems to it helps you work through them, just like talking to yourself.
AI Culture & Society
35. AI Art Discourse The big debates about AI-generated artwork - is it real art? Who owns the copyright? Are human artists getting screwed over?
36. AI Slop Low-quality, obviously AI-generated junk flooding the internet. Generic, repetitive content that screams "a robot made this."
37. AI Wrapper A simple app or service that just puts a basic interface on top of existing AI models like ChatGPT, often marketed as if it's something revolutionary and new.
38. Algorithm Whisperer Someone who's really good at getting AI to do exactly what they want through clever prompting and interaction techniques.
39. Artificial General Intelligence (AGI) The theoretical future AI that's as smart as humans (or smarter) at everything. Currently just speculation and heated debates.
40. Bot Sitter Someone whose job is babysitting AI systems - monitoring them, fixing problems, making sure they don't go off the rails.
41. ChatGPT Moment That "holy crap" moment when someone first realizes how capable modern AI really is. Usually involves jaw-dropping amazement.
42. Deepfake AI-generated fake videos or audio that look/sound real but show people saying or doing things they never actually did.
43. Face Swap AI tech that puts one person's face on another person's body in photos or videos. Popular in social media filters.
44. Promptfluencer Social media personalities who've built their following by sharing AI prompts and tips for getting better AI results.
Privacy & Security
45. Data Poisoning Deliberately messing up training data to make AI behave badly or to protect your personal information from being used without permission.
46. Data Scraping Automatically collecting massive amounts of information from websites and other sources to train AI, often without asking permission first.
47. Differential Privacy A technical method that adds "noise" to data so AI can learn general patterns without exposing specific information about individual people.
48. Model Extraction Stealing or reverse-engineering AI systems by asking them tons of questions to figure out how they work and potentially copy them.
49. Prompt Injection Sneaky attempts to hack AI by hiding malicious instructions inside what looks like innocent input.
50. Training Data Extraction When AI accidentally reveals specific information from its training data that it wasn't supposed to remember or share.
AI Safety & Ethics
51. AI Alignment Problem The big challenge of making sure advanced AI systems actually do what's good for humanity instead of accidentally causing harm.
52. AI Washing When companies slap "AI-powered" labels on products that barely use AI, like putting a "organic" sticker on regular food.
53. Alignment Tax The trade-off where making AI safer might make it less capable. Like how safety features in cars add weight but save lives.
54. Distribution Shift When AI encounters situations very different from what it was trained on, often leading to weird or poor performance.
55. Jailbreaking Trying to trick AI into breaking its own rules or doing things it's not supposed to do, usually through clever prompts.
Conclusion
This vocabulary evolves as quickly as AI technology itself. New terms emerge regularly as people discover novel ways to interact with AI systems and encounter new challenges or capabilities. Whether you're a casual user or working closely with AI, understanding these terms will help you navigate conversations about artificial intelligence with confidence.
Remember that behind all this jargon are real tools changing how we work, create, and communicate. The slang might be new, but the goal remains the same: making technology work better for humans.
This is a much-needed resource! As AI evolves rapidly, it's essential to demystify the language and concepts for anyone, regardless of their technical background, so they can engage effectively in these conversations.