GOOGLE ha CONECTADO sus 4 IAs (…da MIEDO)
A Spanish YouTuber presents a four-level AI productivity system using Google's tools: Notebook LM for source-verified knowledge, Gemini for deep analysis, Gems for specialized assistants, and Workspace integration for deliverable outputs. The system is demonstrated through a real example analyzing 10,000 customer service records and 200 conversations to build a quality auditor assistant. The presenter argues that using AI without a structured system reduces productivity and cognitive engagement.
Summary
The video opens by referencing a MIT MediaLab study claiming that using AI without a system makes users less productive and cognitively passive — letting the machine think for them while they accept lower-quality results. The presenter frames this as the core problem motivating his four-level system built around Google's AI ecosystem.
Level 1 uses Notebook LM as a knowledge foundation. Unlike general AI that can hallucinate convincingly, Notebook LM only works with user-provided sources — documents, PDFs, meeting notes, transcripts — and cites its sources with verifiable references. This grounds all AI outputs in real data rather than fabricated answers.
Level 2 uses Gemini for deep analytical reasoning. With a 2-million token context window, Gemini can ingest enormous volumes of structured and unstructured data simultaneously, find patterns across them, and layer its own reasoning on top of what Notebook LM has verified. This is particularly useful for analyzing large datasets that would take humans months to process manually.
Level 3 introduces Gems — persistent, specialized AI assistants within Gemini. Unlike standard chat sessions that reset each time, Gems retain their identity, the user's context, and their purpose. Users can build a financial advisor Gem, a proofreader Gem, or a domain expert Gem that is always ready without re-entering context.
Level 4 covers Gemini's integration into Google Workspace (Docs, Slides, Sheets), where AI-generated analysis can be directly exported into editable, shareable documents — eliminating the copy-paste workflow between AI chat and productivity tools.
The presenter demonstrates all four levels using a real dataset from a Derekchain research paper: a CSV with 10,000 classified customer service conversations and 200 transcripts, plus operational guides. In Notebook LM, he uploads the transcripts and guides, queries for the five most frequent complaint types (finding: technical failures, shipping delays, improper charges, product defects, price disputes), and generates an executive report exported to Google Docs. In Gemini, he uploads the 10,000-row CSV and asks for incident type analysis by conversation turns, membership level, and product — with Gemini intelligently combining two separate columns to calculate total conversation turns. He then uses Gemini to generate a system instruction for a Gem by combining the qualitative Notebook LM report with the quantitative CSV findings. The resulting Gem, named 'Customer Service Analyst,' answers targeted questions about quality problems and provides three data-backed recommendations: implementing hard authentication logs, automating refund status via self-service, and creating product-specialized FAQs. Finally, in Level 4, Gemini generates a Google Slides presentation of those recommendations, which is then refined using Workspace AI tools — redesigning slides with split vertical layouts, generating bar charts in Sheets, and rewriting paragraphs in Docs — all without leaving the Google ecosystem.
Key Insights
- The presenter cites an MIT MediaLab study claiming that using AI without a structured method causes users to stop thinking critically, remember less of what the AI shows them, and accept lower-quality outputs — effectively letting the machine generate 'fog instead of clearing the path.'
- Notebook LM is described as fundamentally different from general AI because it only draws from user-supplied sources and always provides citations — compared to general AI which the presenter likens to 'a brilliant lawyer who argues wonderfully while fabricating evidence.'
- Gemini's 2-million token context window is highlighted as capable of ingesting the entire Harry Potter and Lord of the Rings sagas simultaneously with space to spare, enabling pattern detection across massive document sets that would take humans months to analyze.
- When analyzing the 10,000-row CSV, Gemini autonomously identified that two separate columns — agent turns and customer turns — needed to be combined to accurately calculate total conversation turns per incident, rather than using either column alone, which the presenter notes would have been an analytical mistake.
- The Gem assistant, built from combined qualitative and quantitative findings, identified that refund status inquiries represent a category that could be fully automated via self-service, eliminating human intervention entirely from that complaint type — a recommendation it justified with specific data from the dataset.
Topics
Full transcript available for MurmurCast members
Sign Up to Access