How I'd Learn AI From Scratch in 2026 (skip the useless 80%)
Learning AI: Practical Steps for Beginners
Introduction to AI Learning
- The speaker expresses feeling overwhelmed by the vast amount of outdated or theoretical AI information available today.
- Focus is on the 20% of practical AI knowledge that will remain relevant over the next decade, organized into three levels.
Step One: Choose a Model and Go Deep
- Emphasizes selecting one AI model to master due to the negligible differences in capabilities among top models.
- Highlights that major AI companies are adopting similar core features, making skills transferable across models.
- Suggests three main models: ChatGPT, Claude, and Google Gemini; notes xAI's lack of competitiveness and Perplexity's limitations.
Step Two: Selecting Based on Principles
- Recommends prioritizing paid tiers for better performance; free versions often lack essential features.
- Advises choosing an AI model that aligns with your work needs—ChatGPT excels in web search, Claude in writing/design/coding, and Gemini in mixed media processing.
- Mentions the importance of personal preference ("vibes") when selecting an AI tool for better engagement.
Step Three: Change Your Defaults
- Urges users to select more powerful models as defaults since companies typically default to weaker options for cost reasons.
- Recaps key points from Level One: use paid versions, choose based on work relevance, and default to capable models.
Importance of Context Over Prompting
- Introduces a partnership with HubSpot providing a Google Gemini cheat sheet with productivity tips.
- Illustrates how context (e.g., past restaurant preferences for recommendations) is more effective than detailed prompts.
Framework for Effective Prompts
- Proposes using "Outcome + Context" (OC framework), emphasizing clear outcomes paired with relevant context rather than complex prompts.
Finding Relevant Context
Strategies for Providing Context:
- Name Framework - Use specific terms like "pyramid principle" instead of lengthy explanations.
- Provide Examples - Share real examples of desired outputs instead of detailing formats or tones.
- Connect Tools - Integrate tools like email or Google Drive so the AI can access necessary files directly.
Managing Recurring Work with Projects
Project Components:
- Project Instructions - Define overarching goals and constraints applicable throughout the project.
- Knowledge Files - Include reference materials such as source documents or frameworks used by the AI during tasks.
- Memory Updates - Allow the AI to track updates and milestones automatically within projects.
Advancing Beyond Individual Projects
Understanding an AI System:
- An AI system connects multiple projects, identifying patterns across them that individual projects may miss.
- It learns from user feedback over time, improving its responses based on previous interactions and edits made by users.
Options for Implementing an AI System:
- Gemini Spark - Beginner-friendly but offers less control; integrates easily with existing Google tools like Gmail and Calendar.
- Claude Cowork - More customizable but requires some setup; designed specifically for non-tech users seeking control over their workflows.[(622)]
- Advanced Options (Claude Code & Codex) - Highly customizable but require coding knowledge; suited for power users looking for extensive functionality.[(645)]
Real-Life Applications of an AI System
Example Scenarios:
- Migrating health checkup data into Cowork allows cross-referencing insights about fitness plans tailored to health conditions.[(675)]
- Using reconciliation features helps refine drafts by analyzing changes made between initial outputs and final edits.[(694)]
Conclusion Recap
- Level One focuses on mastering one chatbot model deeply while utilizing its strengths effectively.[(755)]
- Level Two emphasizes understanding context's role in enhancing output quality through saved projects.[(778)]
- Level Three involves integrating these projects into a cohesive system that improves efficiency over time without rushing progress.[(778)]