digest
Short Summary
Main recommendation: Use a small set of AI tools matched to specific jobs rather than one tool for everything; for everyday work, ChatGPT is best for complex, multi-step instructions, Gemini for mixed media like video and audio, and Claude for first drafts, code, and polished copy. Specialist tools like Perplexity (fast, accurate fact-finding) and NotebookLM (answers only from your own sources) round out part of the toolkit, with two more categories covered in a planned part two video. The speaker says most people should stick with one paid chatbot and add others only when they solve a real problem.
Key Points
- The speaker uses around 10 AI tools, each excelling at one area, and this video covers the everyday and specialist categories, with a second part covering the other two.
- ChatGPT is described as the most "obedient" model: it follows long checklists and instructions to the letter, while Gemini was shown to quietly drop some rules.
- Gemini is the only one of the three that natively processes video, audio, images, and text, and has a 1 million token context window for consumers (2 million for enterprise) for large files.
- Claude is said to produce higher-quality first drafts, especially for working code and for matching a personal writing style.
- Perplexity is for fetching accurate, current facts quickly, while NotebookLM only answers from the sources you upload, reducing made-up answers.
- Perplexity and NotebookLM are described as built on fine-tuned existing models (Perplexity's Sonar model is a fine-tuned version of Meta's Llama, and NotebookLM uses a fine-tuned Gemini model) rather than being foundational models themselves.
Details Worth Noting
Concrete instructions:
-
To test ChatGPT's instruction-following, ask each model to optimize a rough prompt for itself, then run the optimized prompt through both ChatGPT and Gemini and compare whether every requirement is met.
-
For a complex task with many moving parts where one error breaks everything, start with ChatGPT.
-
For video, audio, or large mixed files, upload them to Gemini and ask it to summarize, pull out decisions, or create an SOP; for example, upload a meeting recording, a 20-slide deck, and a whiteboard photo together and ask for a summary, key decisions, and a follow-up email draft.
-
Use Claude for generating working code (including scripts) and for diagrams: ask Claude for Mermaid code and paste it into a tool like Excalidraw.
-
Use ChatGPT or Gemini for ideation, research, and outlining, then pass the rough output to Claude to finish the final draft.
-
Use Perplexity for quick, specific facts, and add Google-style operators such as site:reddit.com to narrow results to one source.
-
Use NotebookLM when accuracy matters more than creativity: upload your script and research, then ask it to flag any claims not supported by the sources.
-
Keep sources accurate for NotebookLM, since it will confidently repeat errors in its sources.
-
Adopt a tool only when it solves an actual problem you have, and most people should stick with one paid chatbot first.
-
The speaker does not use Grok, because its main strength is real-time access to Twitter/X, which they did not need.
-
The speaker mentions the other specialist tools they use less often: Gamma (presentations), ElevenLabs (voice cloning), Zapier and n8n (automation), and Excalidraw and Napkin AI (quick visuals).
-
The video notes that the latest Gemini beat the older version of Claude on every benchmark except coding.
Bottom Line
Build a small, deliberate AI toolkit: pick the tool whose strength matches the job, and only add subscriptions when they clearly solve a problem you have.
