How I'd Learn AI From Scratch in 2026 (skip the useless 80%)
About this video
Jeff lays out how he would learn AI from scratch, focusing on the practical 20% across three levels. Level one: pick one model and go deep. Citing an Artificial Analysis chart, he argues top models have clustered in capability so differences are negligible for average users, and features converge across companies, so skills transfer. He narrows the field to ChatGPT, Claude, and Gemini (dismissing xAI as no longer competitive and Perplexity as a search tool without its own frontier model, in his assessment). Choose by three principles: prioritize paid tiers (paid Gemini from work beats free ChatGPT); match the tool to your work — ChatGPT for research and web search, Claude for writing, design, and coding, Gemini for native mixed media and heavy Google Workspace use; and vibes, since enjoyment drives usage. Also change your defaults, because companies default you to the weakest, cheapest model. He shares two sponsor-segment Gemini tips: typing @aisummary in Google Docs adds a refreshable AI summary block, and =AI in Sheets gives plain-English instructions without formulas. Level two: context beats prompting. His only framework is OC — outcome plus context — illustrated by pasting a push-pull-legs article and asking for a tailored 4-day, 45-minute plan rather than writing a long prompt. Get context by naming a known framework (e.g. the pyramid principle), pasting real examples of good work (the last two to three approved status updates), and connecting tools like email, Drive, Slack, and Notion. Save recurring context in Projects (ChatGPT/Claude) or Gemini Gems, made of instructions, knowledge files, and memory — preferring .md markdown files over PDFs. Level three: AI systems that pull context across projects and learn from feedback, via Gemini Spark (beginner-friendly, least control), Claude Cowork (more control, some setup), or Claude Code and OpenAI Codex (fully customizable, needs coding comfort). His health example: Cowork cross-referenced checkups, workouts, and supplements to flag missing cardio given borderline high cholesterol.
Key takeaways
- Pick one of ChatGPT, Claude, or Gemini and go deep — capability has converged, so skills transfer; prioritize a paid tier and always switch off the weakest default model
- Context beats prompt craft: use the OC framework (outcome + context), e.g. paste a push-pull-legs article instead of writing a long descriptive prompt
- Source context three ways: name a known framework like the pyramid principle, paste real examples of approved work, and connect tools such as Drive, Slack, and Notion
- Store recurring context in Projects or Gems (instructions, knowledge files, memory), and prefer markdown .md files over PDFs because AI reads them more easily and cheaply
- Graduate to an AI system (Gemini Spark, Claude Cowork, or Claude Code/Codex) when projects need to cross-reference each other and compound learnings from your feedback
Episode details: published June 9, 2026 · 13:17 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are Jeff Su’s own claims, shared for education — they are not typical or guaranteed.
← Previous: Claude Cowork: Top 5 Tips for Productivity · Next: Gemini Spark: Google’s Most Powerful Personal Agent →
All Jeff Su videos · Jeff Su profile · Jeff Su (official site)
