Learn AI Right: Skip the 80% That Doesn't Matter

Master AI in three levels: pick one powerful model and go deep, learn to provide the right context instead of perfect prompts, then connect projects into a system that compounds over time. Most people stay at level one; level three is where AI becomes truly useful.

Level 1: Pick Your Model and Go Deep

All top models are now equally powerful

Modern AI models have converged in capability—they're clustered in the top right of capability charts, meaning the difference between ChatGPT, Claude, and Gemini is negligible for most users. Skills learned on one transfer directly to the others.

Three viable choices: ChatGPT, Claude, Gemini

xAI is not competitive, Perplexity is a search tool not a frontier model, and Chinese open-source models lag Western counterparts. Your real choice is between these three based on your work and access.

Prioritize paid tiers over free versions

The gap between free and paid models is night and day. If your job provides paid access to any model, use that one instead of free alternatives, since the capability difference is dramatic.

Match the model to your work and personality

Each AI has different strengths and personality. The more you enjoy using it, the more you'll use it and improve. Switching between models is easy via memory import features.

Always default to the most powerful model available

Companies default you to weaker models because they're cheaper to run. For real work, manually select the most capable model you have access to—it will break down requests, map steps, and catch nuances you missed.

Level 2: Context Beats Perfect Prompts

Prompting frameworks are now obsolete

Modern models are so powerful they infer role, format, and tone automatically. You only need to provide two things: a clear outcome and the right context. Memorizing prompting frameworks wastes time.

The OC Framework: Outcome plus Context

Replace complex prompts with this single framework. Clearly state your desired outcome and provide relevant context (examples, frameworks, reference documents). The AI will infer everything else and produce better results than detailed prompts.

Three ways to find the right context

Explicitly name relevant frameworks (e.g., pyramid principle), provide real examples of what good looks like (past approvals, templates), or connect your tools so AI pulls context directly from email, Drive, Slack, or Notion without manual uploads.

Projects: permanent homes for recurring work

ChatGPT and Claude have 'Projects,' Gemini has 'Gems.' Each contains project instructions (rules and goals), knowledge files (reference docs and examples), and memory (auto-updated by AI). This eliminates repeating yourself for recurring tasks.

Use markdown files instead of PDFs

Markdown files are easier for AI to read and cheaper to process than PDFs. You can ask the AI to convert PDFs to markdown format automatically.

Level 3: AI Systems That Compound Over Time

Projects are silos; AI systems connect them

Individual projects can't see each other's data. An AI system pulls context across multiple projects, spots patterns, and surfaces insights no single project could find. It also learns from your feedback and improves automatically.

Three AI system options for different skill levels

Gemini Spark is most beginner-friendly with minimal setup but less control. Claude Cowork is for non-technical users with moderate control and setup. Claude Code and OpenAI Codex are fully customizable but require coding comfort.

AI systems cross-reference data across projects

Example: migrating separate health checkup, supplement, and workout projects into one system lets the AI flag that you lack cardio despite borderline high cholesterol, recommend adding cardio to rest days, and note your fish oil supplement is sufficient. No single project could connect these dots.

Reconcile feedback to teach the system your style

Share your edited version of AI output back to the system and ask it to reconcile. The AI dissects every change, extracts rules about your preferences, and remembers them for next time. The more feedback, the smarter it gets and the fewer instructions you need.

Practical Tips and Shortcuts

Google Gemini productivity shortcuts

In Google Docs, type @aisummary to add an AI summary block at the top of documents—teammates can click refresh to get the gist without rereading. In Google Sheets, type =AI and give plain English instructions like 'append video number and topic to video title' without writing formulas.

Import memory between models

Switching AI models is easy. In Gemini, go to Settings > Import Memory and follow on-screen instructions to transfer your conversation history and preferences.

Notable quotes

The gap between using AI and using AI well has always been invisible. — Jeff Su
The right context will always beat the perfect prompt. — Jeff Su
Once you go deep on one, that skill carries straight over to the rest. — Jeff Su

Action items

  • Pick one of the three models (ChatGPT, Claude, or Gemini) based on your work type and access level, then commit to going deep on it.
  • Upgrade to the paid tier if your job or budget allows—the capability gap between free and paid is substantial.
  • For your next recurring task, create a Project (or Gem in Gemini) with project instructions, knowledge files, and memory to eliminate repeating yourself.
  • Identify three pieces of context (frameworks, examples, or connected tools) for your most common work and test the OC framework instead of writing long prompts.
  • Convert your PDFs to markdown format and store them as knowledge files in your projects for cheaper, faster AI processing.
  • If you're doing cross-functional work, explore an AI system (Gemini Spark, Claude Cowork, or Claude Code) to connect multiple projects and spot patterns.
  • Try the reconcile feedback technique: share your edited AI output back and ask the system to extract rules about your style for future use.
Jeff Su
13 min video
3 min read
Learn AI Right: Skip the 80% That Doesn't Matter
You just saved 10 min.
The big takeaway
Master AI in three levels: pick one powerful model and go deep, learn to provide the right context instead of perfect prompts, then connect projects into a system that compounds over time. Most people stay at level one; level three is where AI becomes truly useful.
Level 1: Pick Your Model and Go Deep
All top models are now equally powerful
Modern AI models have converged in capability—they're clustered in the top right of capability charts, meaning the difference between ChatGPT, Claude, and Gemini is negligible for most users. Skills learned on one transfer directly to the others.
Three viable choices: ChatGPT, Claude, Gemini
xAI is not competitive, Perplexity is a search tool not a frontier model, and Chinese open-source models lag Western counterparts. Your real choice is between these three based on your work and access.
1
ChatGPT
Best for research, web search, most tutorials
2
Claude
Best for writing, design, coding
3
Gemini
Best for mixed media, Google Workspace users
Model strengths by use case
Prioritize paid tiers over free versions
The gap between free and paid models is night and day. If your job provides paid access to any model, use that one instead of free alternatives, since the capability difference is dramatic.
Match the model to your work and personality
Each AI has different strengths and personality. The more you enjoy using it, the more you'll use it and improve. Switching between models is easy via memory import features.
Always default to the most powerful model available
Companies default you to weaker models because they're cheaper to run. For real work, manually select the most capable model you have access to—it will break down requests, map steps, and catch nuances you missed.
Level 2: Context Beats Perfect Prompts
Prompting frameworks are now obsolete
Modern models are so powerful they infer role, format, and tone automatically. You only need to provide two things: a clear outcome and the right context. Memorizing prompting frameworks wastes time.
The OC Framework: Outcome plus Context
Replace complex prompts with this single framework. Clearly state your desired outcome and provide relevant context (examples, frameworks, reference documents). The AI will infer everything else and produce better results than detailed prompts.
1
Define clear outcome (what you want)
2
Provide right context (examples, frameworks, docs)
3
AI infers role, format, tone automatically
4
Output quality exceeds detailed prompts
The Outcome + Context framework
Three ways to find the right context
Explicitly name relevant frameworks (e.g., pyramid principle), provide real examples of what good looks like (past approvals, templates), or connect your tools so AI pulls context directly from email, Drive, Slack, or Notion without manual uploads.
1
Name frameworks explicitly
Two words carry more context than paragraphs
2
Provide real examples
Examples contain implicit expectations and preferences
3
Connect your tools
AI pulls context from email, Drive, Slack, Notion
Best sources of context for AI
Projects: permanent homes for recurring work
ChatGPT and Claude have 'Projects,' Gemini has 'Gems.' Each contains project instructions (rules and goals), knowledge files (reference docs and examples), and memory (auto-updated by AI). This eliminates repeating yourself for recurring tasks.
1
Project instructions: goals and constraints
2
Knowledge files: source docs, examples, frameworks
3
Memory: AI auto-tracks updates and milestones
4
Reuse across conversations without re-explaining
Three components of an AI project
Use markdown files instead of PDFs
Markdown files are easier for AI to read and cheaper to process than PDFs. You can ask the AI to convert PDFs to markdown format automatically.
Level 3: AI Systems That Compound Over Time
Projects are silos; AI systems connect them
Individual projects can't see each other's data. An AI system pulls context across multiple projects, spots patterns, and surfaces insights no single project could find. It also learns from your feedback and improves automatically.
Three AI system options for different skill levels
Gemini Spark is most beginner-friendly with minimal setup but less control. Claude Cowork is for non-technical users with moderate control and setup. Claude Code and OpenAI Codex are fully customizable but require coding comfort.
1
Gemini Spark
Beginner-friendly, minimal setup, less control
2
Claude Cowork
Non-technical, moderate control, some setup
3
Claude Code / Codex
Power users, fully customizable, requires coding
AI system options by complexity and control
AI systems cross-reference data across projects
Example: migrating separate health checkup, supplement, and workout projects into one system lets the AI flag that you lack cardio despite borderline high cholesterol, recommend adding cardio to rest days, and note your fish oil supplement is sufficient. No single project could connect these dots.
Reconcile feedback to teach the system your style
Share your edited version of AI output back to the system and ask it to reconcile. The AI dissects every change, extracts rules about your preferences, and remembers them for next time. The more feedback, the smarter it gets and the fewer instructions you need.
Practical Tips and Shortcuts
Google Gemini productivity shortcuts
In Google Docs, type @aisummary to add an AI summary block at the top of documents—teammates can click refresh to get the gist without rereading. In Google Sheets, type =AI and give plain English instructions like 'append video number and topic to video title' without writing formulas.
Import memory between models
Switching AI models is easy. In Gemini, go to Settings > Import Memory and follow on-screen instructions to transfer your conversation history and preferences.
Worth quoting
"The gap between using AI and using AI well has always been invisible."
— Jeff Su, at [13:02]
"The right context will always beat the perfect prompt."
— Jeff Su, at [4:34]
"Once you go deep on one, that skill carries straight over to the rest."
— Jeff Su, at [0:31]
Try this
Pick one of the three models (ChatGPT, Claude, or Gemini) based on your work type and access level, then commit to going deep on it.
Upgrade to the paid tier if your job or budget allows—the capability gap between free and paid is substantial.
For your next recurring task, create a Project (or Gem in Gemini) with project instructions, knowledge files, and memory to eliminate repeating yourself.
Identify three pieces of context (frameworks, examples, or connected tools) for your most common work and test the OC framework instead of writing long prompts.
Convert your PDFs to markdown format and store them as knowledge files in your projects for cheaper, faster AI processing.
If you're doing cross-functional work, explore an AI system (Gemini Spark, Claude Cowork, or Claude Code) to connect multiple projects and spot patterns.
Try the reconcile feedback technique: share your edited AI output back and ask the system to extract rules about your style for future use.
Made with Glimpse by Wozart
glimpse.wozart.com/v/eqr7m4gt
Share this infographic

More like this