Everyday AI Podcast – EP 425:
Google Gemini Deep Research – The Best New AI Tool You’re Not Using Yet
Host: Jordan Wilson
Date: December 18, 2024

Overview

In this episode, host Jordan Wilson dives deep into Google Gemini’s new Deep Research tool, a cutting-edge AI-powered research assistant that surprisingly hasn’t made many headlines despite its game-changing potential. Jordan positions Deep Research as a “sleeper” tool—one that can outperform popular solutions like Perplexity and ChatGPT search for complex, multi-source web research. He not only explains the tool’s capabilities and impact on content publishing and internet usage but also compares it live to other AI tools, demonstrates use cases from the livestream audience, and reflects on the broader implications for the internet’s content ecosystem.

Key Discussion Points and Insights

1. AI News Roundup (01:50–08:10)

• Nvidia Jetson Orin Nano Super Developer Kit: New $249 AI computer, 67 trillion operations/sec, aimed at AI/robotics enthusiasts and developers.
• Google Gemini 2.0 Releases:
  • CEO Sundar Pichai announced an experimental Gemini 2.0 EXP 1206 model available for Gemini Advanced subscribers.
  • Promises better performance for coding, math, reasoning, detailed instructions.
  • Access via Gemini model dropdown, paid plan required.
• Salesforce Hiring 2,000 Sales Reps for AI:
  • Despite AI intended to “sell better than humans,” Salesforce plans to hire a significant sales force for its AI agent offering.

Notable Quote:
"If they're trying to sell an AI that is supposed to sell better than humans, why are they hiring humans to sell the AI? That, I don't know. Confusing to me." — Jordan (07:45)

2. The State of Google Gemini & New “Deep Research” Feature (08:10–31:50)

• Historical Frustration with Gemini:
  • Jordan openly admits the Gemini front end has been “absolutely terrible” for users, lagging behind more technical tools like AI Studio or Vertex AI.
  • Front end models would often be months out of date and hard to access meaningfully.
• Recent Updates:
  • In the last week, three new features/modes/models on Gemini’s front end:
    • 2.0 Flash
    • 2.0 Experimental (EXP 1206)
    • Deep Research
  • Deep Research is available only on Gemini Advanced (paid), desktop only for now; mobile release planned in 2025.
• What Makes Deep Research Special:
  • Designed for complex, multi-source web research.
  • Think of it as Perplexity—“but on steroids.” First use, it summarized 169 websites.
  • Can handle tasks that would take a human a day or more, in just a few minutes.
  • Provides comprehensive, cited, and structured reports—can export to Google Docs or Sheets.
• Changing the Culture of Internet Search:
  • LLMs are fundamentally changing how we look for information.
  • Traditional content publishers are losing traffic, making sites “absolutely terrible” to navigate due to ad overload.
  • LLMs “steal” web info and aggregate it; Deep Research does this at a much larger scale than competitors.
• Comparison to Other Tools:
  • Perplexity: Good for real-time, accurate, cited info, but less iterative and more biased towards product recommendations since new features.
  • ChatGPT Search: Good for broad, multi-step, or conversational queries.
  • Deep Research: Superior for in-depth, multi-step, cross-domain research; not conversationally iterative but highly thorough.

Notable Quotes:
“Deep Research is just something else. It’s a different kind of animal. It is literally like having a human sit down and research a very complex task that in general would require a human to read online all day, many hours.” — Jordan (24:30)

3. How Deep Research Works + Pros & Cons (31:50–42:40)

Workflow:

• Prompt Gemini Deep Research with a complex question.
• It presents a “plan” and seeks confirmation.
• Scans and analyzes tens to hundreds of web sources.
• Produces a detailed, source-cited structured report, often with charts and dropdowns for more details.

Strengths:

• Able to tackle complex, multi-faceted, research-heavy questions with speed and depth.
• Cites all sources; “human in the loop” recommended for error-checking.
• Streamlines tasks like competitor analysis, data gathering, marketing research, and more.

Limitations:

• May run into content moderation errors, sometimes arbitrarily.
• Not always clear on what’s permitted; lacks transparency on edge cases.
• Not good at iterative prompting (“make it funnier,” “add more,” etc. after the fact).
• Potential for bias—based on biases present in source material on the web.
• Only for desktop (mobile planned 2025); only for paid Gemini Advanced personal Google accounts.

Notable Quotes:
“Google Deep Research can take on complex, multi-step researching tasks that a human would do. And a human might take more than a day and it can do it in a couple of minutes.” — Jordan (38:40)

4. Live Demonstration: Real Use Cases (42:40–1:00:00)

Example 1: Major AI News Summary

• Jordan prompts Deep Research:
  “OpenAI and Google have collectively released more than a dozen noteworthy new AI products and updates over the past two weeks. Please research in detail every AI release from OpenAI and Google over the past two weeks. For each, tell me what the update or release is, what it does, who it’s for, when it’s available, and the three main competitors for each product and service.”

• Deep Research output:

  • Searched 39+ sites (live), completed in ~3 mins, produced structured, detailed report with sources, summaries, charts (some incomplete but impressive).

• Comparison to ChatGPT/Perplexity

  • ChatGPT and Perplexity were faster (5–20 seconds) but both only gave partial/incomplete answers and covered fewer releases.
  • Deep Research correctly covered most, offered richer formatting and better citation.

Example 2: Audience-Suggested Research

• Prompt: “Please deeply research tiny house communities in all 50 states in the U.S. How many are in each state and where are they?”
• Deep Research generated a comprehensive overview, searched 92 sites, provided detailed breakdowns by state.
  • Hit a token limit—Jordan attempts to prompt Gemini to “continue” (results in newsletter tease).

Notable Quote:
“Imagine going individually to these 92 websites…This could have taken you weeks. The internet is a very distracting, terrible place. So now, Google Gemini Deep Research went to those 92 houses...gave me a comprehensive guide.” — Jordan (52:40)

Key Observations from Demonstrations:

• Deep Research is not always fully iterative; may hit response or token limits.
• For truly exhaustive or extremely broad queries, content may need to be chunked and requested piece by piece.

5. What This Means for the Internet and Content Creators (56:00–End)

• LLMs are Undercutting Traditional Content Models:
  • As more people turn to LLMs/answer engines, traffic to content creators declines.
  • If tools like Deep Research, ChatGPT Search, and Perplexity capture user research phases, who pays for content creation?
  • Possibility of future ad/revenue-share models between LLM aggregators and source content publishers (akin to Spotify or YouTube monetization).

Notable Quote:
"What happens when these answer engines...start taking that [content]?...I'm a human creating content for other humans. I love AI. I have nothing against this. But what happens? Who's gonna pay these humans creating this great content for us to enjoy? I'm not sure, but I think it's a radical shift." — Jordan (56:50)

• Efficiency Gains for Knowledge Workers:

  • Deep Research can cut hours—potentially days—of research to minutes.
  • Either reduce research time by 95% or go 10x deeper in the same time.

• Final Take:

  • Deep Research is not yet a total replacement for human research but is a transformative tool.
  • Always check for errors/bias (“human in the loop”), but expect significant productivity gains.
  • Implications for the web ecosystem are profound—will content creators be compensated in the LLM-dominated future?

Notable Quotes & Memorable Moments (with Timestamps)

• “Even though I know...there’s people at these companies...sometimes I feel bad if I say a product stinks...I don’t care.” — Jordan (10:25)
• "The Internet sucks. It is an absolute terrible place...The browsing the Internet is a complete wasteland." — Jordan (20:10)
• “I think Deep Research is a sleeper. The first time I used it, I said, ‘Perplexity’s cooked.’” — Jordan (18:55)
• “On the back end, Google AI Studio, top of the class; Vertex AI outstanding. Google Gemini on the front end...don’t touch it...You can touch it now. The stove is no longer hot.” — Jordan (50:55)
• “If you want to know how this turns out...go to our website, your everydayai.com. Sign up for the free daily newsletter.” — Jordan (54:35)
• Customer Testimonial: "Jordan Wilson from Your Everyday AI is my cheat code...this is my secret." — (56:25)

Important Segment Timestamps

• 00:16 — Intro to episode’s main topic
• 01:50 — AI news roundup (Nvidia, Google Gemini 2.0, Salesforce)
• 08:10 — Transition to Gemini Deep Research; Jordan’s honest take on Gemini
• 18:55 — Why Deep Research is a “sleeper” tool; comparison to Perplexity
• 24:30 — What Deep Research is (capabilities, access requirements)
• 31:50 — How it works, pros and cons
• 42:40 — Live demo: Deep Research on major AI releases
• 48:55 — Audience examples; deep dive on “tiny house communities” research
• 54:35 — Token/response limits; newsletter tease
• 56:00 — The future for web content, monetization, and research work
• 57:17 — Closing remarks

Tone & Style

Jordan is candid, energetic, occasionally irreverent—but always focused on practical value. He balances technical details with relatable analogies (“faster than a human reading all day,” “the stove is no longer hot”), critiques both the fandom and flaws of big AI companies, and remains transparent about both capabilities and limitations—always returning to “what does this mean for real people using AI in work and life?”

Summary

Jordan Wilson positions Google Gemini Deep Research as a leap forward in AI-powered web research, outshining competitors like Perplexity and ChatGPT search for multi-source, comprehensive tasks. He demonstrates both its power and its quirks, offering honest, practical guidance to knowledge workers and content creators. As answer engines like this reshape the landscape of internet content and compensation, Jordan calls on listeners to use these tools wisely, keep their “human in the loop”—and remain engaged in the evolving future of work and information online.