How to Use Gen AI
Maximize your AI
New to generative AI? So is everyone else. Use it to get ahead, not to replace your own judgment. This guide covers the fundamentals, how to personalize your AI, what to avoid, and how to think about which model to use.
Generative AI Fundamentals
What is Gen AI?
Generative AI describes systems that produce new content — text, images, code, audio — by predicting likely continuations from patterns learned during training. A large language model does not “look up” answers; it generates a plausible response one piece at a time. That makes it powerful for drafting, summarizing, and brainstorming, and also means its output should be treated as a confident first draft, not a verified fact.
Who uses Gen AI?
People in nearly every field: engineers use it to draft and review code, writers to tighten prose, analysts to summarize documents, and students to explain hard concepts. You do not need a technical background — the main skill is describing what you want clearly and checking the result.
Is it going to take our jobs?
The more useful framing is that AI changes tasks more than it eliminates whole roles. It handles repetitive drafting and lookup work, which frees you to focus on judgment, context, and decisions. Treat it as a capable assistant that still needs a human in the loop — you own the final result.
Personalize Your AI
Give it context
The quality of an answer tracks the quality of the context you provide. Tell the model who you are, what you are trying to achieve, the audience, the format you want, and any constraints. Many tools also offer persistent workspaces — such as Projects or Spaces — where you can store background material and instructions so you do not have to repeat yourself every time.
Iterate on your prompt
Rarely is the first prompt the best one. If a response misses, do not start over — tell the model what was wrong and ask it to adjust. Ask it to be more concise, change the tone, show its reasoning, or give alternatives. Prompting is a short conversation, not a single command.
Manage the AI’s memory
Some assistants remember details across chats. Look for a memory or personalization setting in your tool to review, edit, or clear what it has stored about you. Clearing memory is useful when you switch topics, or when you want a clean, unbiased response.
Don’ts With AI
What not to put into AI
Do not paste personal or sensitive information you would not want stored or seen by others — no personally identifiable information (PII), no protected health information (HIPAA-covered data), no passwords, no client secrets, and no proprietary or classified material. Once information is submitted, you often cannot take it back.
Too much or too little instruction
Vague prompts produce vague answers; over-stuffed prompts bury the actual request. Aim for a clear goal, the essential context, and the output format — then refine. If the model is guessing at what you mean, that is a signal to add one specific detail, not ten.
Are they using my information?
Check the privacy settings in the app you are using. Many services let you turn off training or history. As a general rule, assume that text you input and receive may be retained by the provider — often not tied to your account, but retained nonetheless. When in doubt, leave sensitive details out.
Which Model Should I Use?
Different models are better at different things — some excel at images, some at long-form text, some at data analysis or coding. Rather than name specific products (which change constantly), use the table below as a template: fill in the current models you have access to and note their strengths, trade-offs, cost, and where to find them. Try the same prompt in two tools and compare — that quickly tells you which one fits a given task.
| Model name | Strengths | Trade-offs | Cost | Best for |
|---|---|---|---|---|
| OpenAI - ChatGPT | Strong general reasoning, coding, writing, tool use, broad ecosystem | Can be expensive at higher tiers, and output quality varies by mode/version | About $5 / $30 per 1M tokens for GPT-5.6 Sol in Vellum’s comparison | Everyday chat, coding, business writing, agents |
| Anthropic - Claude | Excellent reasoning, long-context work, strong coding and analysis | Often pricier than mid-tier alternatives; some variants are optimized for non-reasoning use | About $5 / $25 per 1M tokens for Claude Opus 4.8 | Long documents, coding, careful analysis, agentic workflows |
| Google - Gemini | Strong multimodal ability, broad utility, good browser and productivity use | Can be less consistent than the very best reasoning-first models on some tasks | About $2 / $12 per 1M tokens for Gemini 3.1 Pro | Multimodal work, research, productivity, general assistant use |
| Chinese Hedge Fund - DeepSeek | Very strong value, good reasoning and coding, often lower cost | Smaller ecosystem than OpenAI/Anthropic/Google; deployment and support vary by provider | About $0.435 / $0.87 per 1M tokens for DeepSeek V4 Pro | Budget-conscious coding, reasoning, high-volume use |
| Meta - Llama | Open-weight flexibility, self-hosting, privacy control, lower cost to operate | Usually requires more engineering effort and hosting infrastructure; quality depends on deployment | About $0.01 / $0.34 per 1M tokens for Llama 4 Scout in one leaderboard comparison | Self-hosted apps, private deployments, customization |
Ready to practice? Browse the prompt library and try one in your tool of choice.