Learning Library

← Back to Library

Gemini Launches Deep Research and Mariner

Key Points

  • Google Gemini’s “Deep Research” feature appears to dramatically reduce citation hallucinations, effectively automating accurate scholarly sourcing.
  • This breakthrough sparks a broader education debate: which research skills should still be taught manually versus delegated to AI, and how to prevent critical‑thinking atrophy.
  • The tool blurs the line between legitimate research assistance and plagiarism, prompting higher‑education institutions to reconsider policies on AI‑generated work.
  • Gemini’s “Mariner” is an experimental, agentic browser extension for Chrome that lets AI act autonomously on web pages, raising both strategic (despite antitrust pressures on Chrome) and privacy considerations.

Full Transcript

# Gemini Launches Deep Research and Mariner **Source:** [https://www.youtube.com/watch?v=Qqlfu5ml41U](https://www.youtube.com/watch?v=Qqlfu5ml41U) **Duration:** 00:06:00 ## Summary - Google Gemini’s “Deep Research” feature appears to dramatically reduce citation hallucinations, effectively automating accurate scholarly sourcing. - This breakthrough sparks a broader education debate: which research skills should still be taught manually versus delegated to AI, and how to prevent critical‑thinking atrophy. - The tool blurs the line between legitimate research assistance and plagiarism, prompting higher‑education institutions to reconsider policies on AI‑generated work. - Gemini’s “Mariner” is an experimental, agentic browser extension for Chrome that lets AI act autonomously on web pages, raising both strategic (despite antitrust pressures on Chrome) and privacy considerations. ## Sections - [00:00:00](https://www.youtube.com/watch?v=Qqlfu5ml41U&t=0s) **Gemini Deep Research & Mariner** - The speaker explains how Google Gemini’s new Deep Research tool appears to solve citation hallucinations and introduces the Mariner agentic browser, prompting discussion on how AI will reshape research practices and the teaching of scholarly skills. ## Full Transcript
0:00Google Gemini was such a big release 0:03that I didn't get to talk about two of 0:05the most important pieces that dropped 0:06yesterday one is deep research and one 0:09is what they're calling Mariner which is 0:11their agentic web browser I want to talk 0:13about both here for deep research I 0:16think the big takeaway is that we are 0:18learning that the hallucinations problem 0:21that has plagg in particular research 0:24citations through AI may be gone or may 0:27be going away or mostly gone we have to 0:29do work on deep research I'd love you to 0:31try it out and let me know what you find 0:34but playing around with it it just looks 0:36like the ability to site correctly after 0:40doing research on a topic is mostly 0:42solved by this 0:44product and if that's the case it begs 0:46the 0:48question how do we teach the skill of 0:50research to people who are coming up in 0:52the 0:53field so as an experienced researcher 0:56super easy you can go really fast with 0:58it I get the application great product 1:01I'm sure it's going to get adoption but 1:03students are also going to use it 1:05they're going to use it right away and I 1:07think this is a nice doorway into one of 1:09the larger meta questions around AI 1:11which is that it's this enormous leveler 1:13it's teaching us to do things faster 1:16than we've ever done them before but it 1:18raises a question of what are useful 1:21skills that we still want to know how to 1:23do versus what are skills that we are 1:26okay letting 1:27atrophy so one of my examples that I 1:30like to go back to is when I was a kid 1:32we had calculators uh we could do all of 1:35our math on calculators but the teachers 1:37would never let us use the calculators 1:39because they wanted us to learn to think 1:42logically on our own and solve math 1:44problems similarly I think one of the 1:47larger meta questions in education right 1:49now is to what extent are you keeping 1:53students from learning skills they need 1:55in the future by keeping them from 1:57learning with AI versus to what extent 1:59do you actually need them to learn 2:00without AI in order to gain the critical 2:02thinking skills they will need to use it 2:04properly in the 2:05future deep research is one of those 2:07that is aimed squarely at that debate 2:09and we're going to see in the next few 2:12weeks next few months how schools start 2:13to navigate that particularly higher 2:15education like if you're a junior or 2:16senior in high school if you're in 2:18college if you're in advanced degree 2:20studies deep research is for you and 2:22we're going to start to Grapple with 2:24that what does that mean it's not 2:26plagiarism researching is not plagiarism 2:28there's something else going on on here 2:30and it's going to be an interesting 2:31conversation number two Mariner uh no 2:35this is not the Seattle Mariners I'm 2:36from Seattle we're never going to win a 2:38World Series uh Mariner is the code name 2:41that Google gave to their agent based 2:45browser extension so this is an 2:47experimental extension in Google Chrome 2:50it is a different conversation as to why 2:52they're launching this in Google Chrome 2:54when Chrome is something that the 2:56justice department is seeking to make 2:58them sell 3:00I think the take there is probably that 3:02Google doesn't think the sale will 3:04actually happen and so it thinks it's 3:06safe to keep investing here it probably 3:08also feels like it can't miss on agent 3:12based AI in 2025 and it has to take the 3:14risk 3:15regardless anyway Mariner is out there 3:18experimental mode it asks your 3:20permission before taking sensitive 3:22actions like completing a purchase but 3:25it is able to independently navigate the 3:28web and take actions on its own to 3:31accomplish a larger task the demo video 3:33I saw was using the browser to go and 3:36plan a vacation now I will say as 3:38someone who saw the demo video for Sora 3:40and has now been playing with Sora from 3:42open AI demo videos are deliberately 3:45cherry-picked right like they're picked 3:47to show the most effective use case the 3:49best example the highest and best use 3:51the thing that makes it work the best I 3:54have played with Sora and gotten a few 3:55good videos I've gotten a few eh videos 3:57and I've gotten some really bad ones 4:00I think you'll see similar results 4:01playing with Mariner but that doesn't 4:04mean it's not coming in 2025 I think one 4:06of the big decisions we all will need to 4:08make in 2025 is to what extent are you 4:11comfortable with an agent using the 4:14browser on your 4:16laptop we'll have that opportunity next 4:19year and it's going to require a 4:21different kind of modality for us we're 4:23used to a browser method of interaction 4:25where the active tab is where we are and 4:28there's other passive tabs that we can 4:29go and access when we need them but with 4:32an agent it's really that the agent has 4:35a journey they're taking maybe they're 4:37opening multiple tabs they're going 4:39through their web search they're 4:40accomplishing their task for you and I 4:43think ideally you would also want the 4:45option to be on the laptop at the same 4:47time looking at something 4:49else we don't really have a modality for 4:53how that works we don't know what that 4:54user experience looks like yet it's 4:56going to be one of the big design 4:57challenges of 2025 and I don't think 5:00that Google got it right with this one 5:02so at the end of the day if you are 5:04installing Mariner and using it in test 5:06mode it simply uses the active tab right 5:10in front of you to complete its task 5:12which means you really can't go do 5:14something else unless you're going to 5:15leave your laptop like you just would 5:17sit there and stare at it that's not 5:19saving anybody time so we're going to 5:22have to come up with new ways for agents 5:24to use the computer while humans also 5:25use the computer and I think that's 5:27going to be one of the interesting 5:28design challenges for for the coming 5:30year so there you go two things that you 5:32may not have heard about that also 5:34launched with Gemini it's like Gemini 5:36heard that open AI did their 12 days of 5:39open Ai and said forget that we're going 5:41to do it all in one day we're going to 5:42drop like a dozen things have a Quantum 5:44chip you know have have agent-based web 5:48browsing have a brand new large language 5:50model all within like 24 hours it's 5:53crazy all right we will talk soon I'm 5:56sure the AI Gods will give us more news 5:58shortly