Learn · Fundamentals
AI Fundamentals
This page covers what AI is and what an LLM is—the same ideas as step ①–② on the beginner path (/learn/learn-path). After reading, you will know whether to explore Meet AI concepts, the tool library, or agent demos next.
What is AI?
Artificial intelligence (AI) is software that learns patterns from data to predict, classify, generate, or decide—often at scale where hand-written rules break down. Classic ML powers recommendations, fraud scores, and speech recognition. Generative AI adds fluent text, images, and code; today that usually means large language models or multimodal models behind a product UI.
What is an LLM?
A large language model (LLM) is trained on huge amounts of text (and often code or captions). It reads, summarizes, reasons, writes, and translates by predicting likely next tokens—not by looking up a fixed database. ChatGPT, Claude, and Gemini conversations are LLMs plus product rules, safety filters, and optional tools.
Because output is probabilistic, LLMs can sound confident while being wrong (hallucination). Treat factual claims as drafts until verified.
- Input: your prompt plus optional files or images
- Output: generated text (sometimes tool calls)
- Context window: max text the model can consider at once
- Settings like temperature: more creative vs more conservative
Training vs using
Training happens in vendor data centers—expensive and not something end users do. Inference is what runs when you send a message: the model computes a reply on the fly. The same base model can feel different across apps because of system prompts, browsing, file upload, and tool policies.
Chat, copilot, and agent
Chat mostly delivers answers in turns. Copilots inline in editors or office apps suggest edits. Agents pursue a goal across steps—plan, pick tools, act, check, repeat until done or until they need your approval.
- Chat: brainstorming, explanations, short copy
- Agents: research reports, multi-file code changes, cited search
- MEAGENT’s try demo shows a **task loop**, not a single reply
Models, tools, and agents
Think in layers: models provide capability; tools are products you pick (ChatGPT, Cursor, …); agents orchestrate multi-step work on top. On Discover (AI Universe), those paths are parallel—you can start with a tool or an agent, but knowing LLM limits saves frustration.
Limits and responsible use
- Hallucinations: verify important facts, numbers, and citations
- Knowledge cutoff unless the product browses or uses RAG
- Privacy: avoid secrets in untrusted chat windows
- Compliance: generated content rules vary by region and use case
- Human in the loop: AI drafts; you decide and publish
Four prompt habits
Use a simple scaffold: role + context + task + format + constraints.
- Audience: executive summary vs engineer detail
- Materials: outline, data scope, or counterexamples
- Format: bullets, table, Markdown headings
- Bounds: length, tone, “do not invent sources”
Where to go next
- Concept overviews → Meet AI (/learn/concepts)
- Step 3 on the path → What is an AI agent? (/learn/concepts/what-is-ai-agent)
- Pick a product → AI tool library (/tools)
- Try hands-on → Home demo (/#try-agent) or Agent store (/agents)
Continue beginner path
Use the interactive demo on MEAGENT or open a product guide—start with a small, low-risk task.
Continue beginner path