Aarush Selvan
@AarushSelvan
Followers
5K
Following
899
Media
33
Statuses
409
Product Manager, Gemini App. I started Deep Research a smattering of other things. From 🇬🇧 living in 🇺🇸. Stanford BS, MS. Views and opinions my own.
Joined August 2019
Well I wonder where they got that idea from?🤔.
Today we are launching our next agent capable of doing work for you independently—deep research. Give it a prompt and ChatGPT will find, analyze & synthesize hundreds of online sources to create a comprehensive report in tens of minutes vs what would take a human many hours.
110
46
2K
Today we’re announcing the next iteration of Deep Research. It’s now powered by Gemini 2.0 Flash Thinking, which makes it better at every part of its research. Also, we’re making it free to try for all Gemini users!.
Today, we’re making big upgrades to the performance and availability of our most popular Gemini features, and adding a new feature that will make Gemini even more personal and helpful: Let’s break down today’s updates 🧵⬇️
33
62
914
We just put out Deep Research powered by Gemini 2.5 pro. It's only available to Advanced users for now but is a step-change improvement in capabilities of this feature. Excited to see how people will use it.
📣 Deep Research is now powered by Gemini 2.5 Pro, our most intelligent AI model. ✨. This upgraded Deep Research is now even better at:. 🔍 Finding & synthesizing information.📊 Providing more insightful reports.🧠 Analytical reasoning. Gemini Advanced users can access the new
27
32
609
Update from the weekend: the upgraded Gemini Deep Research is now ramped to all iOS and Android phones. Also we doubled the usage limits for all free users 🙂.
17
18
453
Today we are announcing Canvas, a new way of collaborating with Gemini to write and refine documents or code. It's got a ton of nifty features (inspired by some of our other popular collaboration tools), is free and is available in all languages and countries where @GeminiApp is.
With Canvas in Gemini, you can:. ⌨️ Write, iterate & preview React/HTML code.📝 Draft & edit comprehensive documents.🎨 Build interactive prototypes, games & visualization.…and more. Simply select ‘Canvas’ in your prompt bar and you can write and edit documents or code, with
7
12
239
Favorite Deep Research query: "You are tasked with building out a business case for a new robotaxi business in the UK - a JV between waymo and uber. Build a discounted cash flow model for the value of the business after 3 years, assuming they start today". Check it out:.
16
8
213
We are also enabling you to create Audio Overviews from within the @Geminiapp, thanks to a little help from the @NotebookLM team. What this now means is that you can turn your Deep Research reports into an audio overview and listen on the go.
6
6
190
Some simple tips on how to use Deep Research from me and our eng lead Mukund
8
14
142
Happy IO from the Deep Research team (especially the heroes not in the photo making sure we stay up)
2
0
129
Don't know about you, but I love a good deal.
@GeminiApp is now amazing. Best Model + Deep Research . 20 Per Day!! Compared to @ChatGPTapp 10 per month. For the same price. Crazy value. I find myself using it more. Amazing work @AarushSelvan and @GoogleAI team.
2
3
124
Super excited about the mobile rollout! You can now do Deep Research while on the go and even use it cross platform (kick off tasks from your phone, read when you get to your computer). Thread below on how to use 🙂.
We're inspired by all of the ways that people have used Deep Research since we launched in December. Today, we're excited to expand availability to hundreds of millions of Android users worldwide and are working hard to expand to 100% of mobile users including iOS next week! 📲.
13
5
118
Been dogfooding Gemini Live with Video for a little while now, its a lot of fun. Favorite way of using it was walking around Barcelona, pointing it at cool buildings and having a conversation about them.
4
3
108
We built Deep Research to help you overcome information overload. Here, Product Manager Aarush Selvan breaks down how it plans, reasons, and adapts to complex research tasks. Visit and select Deep Research from the Gemini Advanced model drop-down to get
7
3
98
Whoa, who's this handsome fella?.
We built Deep Research to help you overcome information overload. Here, Product Manager Aarush Selvan breaks down how it plans, reasons, and adapts to complex research tasks. Visit and select Deep Research from the Gemini Advanced model drop-down to get
7
1
94
😊.
Deep Research with Gemini 2.5 has become very good. It spontaneously generates tables, scenarios, and compiles evidence. Haven’t spotted errors in spot checks.
5
1
79
@emollick (2) isn't quite correct. Deep Research today does reflect on findings before issuing new searches. It's why you see the website count in the UI increase in batches as it researchers. Certainly headroom to make it a lot better.
3
0
61
6
0
61
Little nugget in here: . "One early example is Deep Research. It uses agentic capabilities to explore complex topics on your behalf and give key findings, along with sources. It launched in Gemini Advanced in December and is rolling out to Android users all over the world".
Read the transcript of @sundarpichai's remarks from today's 2024 Q4 earnings call.
4
2
47
And starting today you can export Gemini code outputs in over a dozen programming languages directly to Replit. Thanks for the partnership @amasad!.
Excited to see what people make with Gemini on Bard x Replit.
4
8
46
Today we announced Bard Extensions - allowing you to collaborate in Bard with Google apps and services like Maps, Flights, YouTube and Workspace. It's been an absolute privilege leading this effort with so many incredible team members!
3
6
40
Great overview of our launches from the last week with our very own @OfficialLoganK.
In the latest episode of Release Notes, Gemini's Dave Citron joins @OfficialLoganK to deep dive into some of the latest Gemini updates. 🎙️ Learn more about Gemini with personalization, Canvas, Audio Overviews, Deep Research, and more:. Timestamps:.0:00 Introduction.0:59 Recent
0
1
41
This has been my favorite feature on Gemini for a little while (though I am a bit biased). No files needed!.
One of our favorite ways to use the Audio Overviews feature is to pair it with Deep Research. Now, with just a click, you can turn your Deep Research reports into a lively conversation for learning on-the-go. Your AI hosts will summarize the material, draw connections between
1
1
39
With AI Studio, Gemini 2.0 Flash/Flash Thinking APIs and now Gemma 3, we are dramatically lowering the cost of entry to get started and build with generative AI. Its going to be so exciting to see the products that will be dreamed on this new frontier
4
1
38
Thanks so much to @jason and @sundeep for reviewing Bard Extensions and showing off some really cool journeys on @twistartups. Appreciate the passing grade, but really hoping my Indian parents don’t see this C+ on my transcript. Working hard to get that A next semester!.
6
3
36
Thrilled to finally share what I've been working on!.
Soon we’ll launch extensions in Bard, tapping into services from Google and across the web to help you get more done and spark your creative explorations. These extensions will be enabled in a secure and private way. #GoogleIO
3
0
35
@OfficialLoganK @deedydas Gemini Deep Research is available globally on Android as of today, so hopefully we can get a green box there 🙂. This is a cool query, will share with the team.
3
0
33
Its really great at queries that require combining researching the open web with math/analytical reasoning.
This was actually very impressive. An assignment in my class is to determine the price & TAM of a product that can give the calorie count of food using a new form of spectroscopy, with $75 COGS. The new Gemini 2.5 Deep Research got all of that & much more. It used to take hours.
6
0
32
I love reading Gemini Deep Research's thoughts as it works in real time - make this super clear and are also really easy to follow.
One thing the Deep Research models make clear is letting AIs agentically use their own tools to do search & see the outcomes can result in much better outcomes than RAG. The ability for AI to understand context for answers, see what isn’t found, & follow “curiosity” all matter.
5
1
31
We heard you liked Deep Research, so we expanded access to a bunch more languages. Give it a try, let us know what you think 🙂.
Deep Research is now available in 45+ languages and more than 150 countries with Gemini Advanced → 🤓.
1
3
27
Using Bard Extensions for inspiration on what to do with that one free machine at the gym 🤓. Image in -> video out
0
4
24
@OfficialLoganK @jwblackwell @OpenAI All Advanced users globally should have access on their personal gmail account. Enterprise accounts are coming soon!.
5
1
24
We’re also launching an experimental capability called Personalization, which uses your Search history to give you more tailored responses. Give it a try, we’re excited to see creative ways people might use it. It’s also free!.
1
2
22
Gemini Deep Research FTW.
With the new reasoning models, there is increasing value in coming up with good questions. That reminded me of an old short scifi story about a future where coming up with new PhD ideas in a world of AI was hard. I couldn't remember the name. Only Gemini Deep Research found it.
3
0
22
You can try Deep Research in over 45 languages and all over the world. Gemini Advanced users should have access to the new Deep Research on already. We are currently ramping up to the rest of Gemini. The mobile app will take a little longer.
2
1
20
Dropped this small improvement to the YouTube extension on Bard a couple days ago. Glad its getting some traction, hope folks find it helpful 🙂. Lots more work to do!
1
1
21
It's been an absolute privilege working with the @Replit team to build an integration into Bard! Earlier today our very own @DynamicWebPaige sat down with @pirroh.and @CeciliaZin to talk about our partnership and the future of AI-powered software development. Check it out here:.
Bard x Replit AI Podcast Livestream
1
2
21
Lovely surprise seeing Gemini Deep Research discussed on this week’s episode of @theallinpod, thank you once again @Jason! Can confirm we didn't pay him 😆. You can check it out here (live demo also at the 1hr 7 minute mark).
1
3
18
@emollick We added a “Create” feature in Deep Research for you to do this kind of task more seamlessly (including building a game 😊). Lmk what you think
5
0
19
1
0
18
Love the creativity! Never imagined it could be used like this.
Only ONE free AI assistant I've found can pull info directly from YouTube. ✨ Google Bard. I asked for guided notes (with blanks) for a Crash Course video and just provided a link. No transcript. It used its YouTube extension to access the video and create the notes.
2
5
15
So excited for this! We started with being able to export Python code to Replit, but soon realized people want to do so much more. Starting today Bard can export 18 programing languages. And with Gemini Pro, Bard is getting even better at coding. I want to give a huge shoutout.
We’re expanding our integration with @Google Bard. You can now export code from Bard to Replit in 15+ programming languages (including JavaScript, Python, CSS, HTML, and TypeScript), across multiple files.
1
3
17
Finally got an A! Thank you @Jason and @sundeep for showcasing Deep Research on TWIST, looks like my GPA is trending up 😅. In all seriousness, the team has been working really hard on this, so thanks for giving it a go! Lots more improvements to come.
Thanks so much to @jason and @sundeep for reviewing Bard Extensions and showing off some really cool journeys on @twistartups. Appreciate the passing grade, but really hoping my Indian parents don’t see this C+ on my transcript. Working hard to get that A next semester!.
4
1
16
Really enjoyed talking to @swyx and @FanaHOVA about how we built Deep Research. Thanks so much for having us on!.
🆕 pod: The Inventors of Deep Research!. While everyone cloning Deep Research, we asked @AarushSelvan and Mukund Sridhar, the original PM and Tech Lead who created the newest killer use case of AI Agents now copied by @openai, @xai, @perplexity_ai and a.
2
3
15
Completely agree! Good ideas should be reproduced and built upon. And kudos to the @OpenAI team for shipping it, the results do look impressive.
@AarushSelvan Was my first thought when I saw this. And it's named the same too. But to be fair , ai industry copy each other's ideas all the time. It's a case of best implementation wins. Congratulations on leading the industry in this crucial idea.
4
0
14
🇯🇵🇰🇷 Yesterday we also launched Bard Extensions in Japanese and Korean. Now users in these countries can use Bard to pull together information from apps like Maps, YouTube, Gmail and Docs in their native language. Incredibly grateful for the teams of engineers and linguists that
3
2
13
@garrytan ngl, this wasn't a use case we imagined when we designed Deep Research. Pretty neat productivity hack.
1
0
13
@joshwoodward @phill__1 @GeminiApp Working on a few infra adjustments to accommodate these cases - mainly stemming from the fact the model is more compute intensive and has longer outputs. Should have a fix shortly. Apologies!.
1
0
13
Been an absolute blast building this with the team! Rolling out today for Gemini Advanced users.
We’re also introducing a new agentic feature called Deep Research in Gemini Advanced, a research assistant that can dig into complex topics and create reports for you with links to the relevant sources.
1
3
12
So incredibly proud of @triswarkentin and the Gemma team for this launch! Congrats friend.
Today, we launch Gemma 3 in Paris! Excited to share the strongest model in the world that can run on a single GPU -- outperforming Llama-405B, DeepSeek-v3, and even o3-mini. Multilingual in over 140 languages with a 128 context window and multimodality! 🔥.
0
0
12
This was one of the most requested features when we first launched Deep Research back in December and enables you to go from research idea → podcast-style conversation all in one place. Just look for the dropdown next to the “Export to Docs” button to get started.
1
1
12
Seeing Deep Research make this list is the sincerest form of flattery 😍.
common themes:. AGI.agents.much better 4o upgrade.much better memory.longer context.“grown up mode”.deep research feature.better sora.more personalization. (interestingly, many great updates we have coming were mentioned not at all or very little!).
2
1
12
@ai_for_success @OfficialLoganK I just watched a Gemini engineer grab 4 Monster cans from the fridge just now. Sleep is overrated.
3
0
11
@infoxiao @NotebookLM @GeminiApp You can create an audio overview directly from within Deep Research. Just look for the "Audio Overviews" button next to "Export to Docs".
3
0
10
@DynamicWebPaige @MahawarYas27492 @OfficialLoganK @joshwoodward @GeminiApp Yes it's powering Deep Research as well.
1
0
10
The fact that the OpenAI team are still shipping new features amongst all the flux is nothing short of incredible. Kudos 🫡.
0
2
10
Enjoy!.
📣 Gemini is now powered by 2.0 Flash, our latest model that is designed for the agentic era, delivering fast responses and stronger performance across a number of key benchmarks for everyday help with tasks like brainstorming, learning, or writing. Rolling out to all users
1
0
9
This is one of the hardest jobs in AI and certainly the hardest job for PMs in AI. Worth a read.
// i lead model behavior at openai, and wanted to share some thoughts & nuance that went into setting policy for 4o image generation. features capital letters (!) bc i published it as a blog post:. --.This week, we launched native image generation in ChatGPT through 4o. It was.
0
0
9
Thank you to everyone who’s tried Extensions out already! Excited to keep improving this experience to make Bard more helpful for everyone. This is day 1.
We’re adding extensions to Bard so you can connect it to your favorite Google apps including Gmail, Drive + Docs for even deeper collaboration. We’re also updating how we validate the claims in Bard’s responses with an improved “Google It” button + more.
2
1
9
This is epic.
I've been reading Bethany Hughes book on the 7 Wonders & the myths of the Colossus of Rhodes (eg it did not span the harbor). As an experiment, I asked Google Deep Research to come make a prompt to describe the historical Colossus for veo 3. Nice, down to the location & riveting
0
0
9
@AravSrinivas @amasad @rauchg It's a culture that permeates Replit. Everyone I was lucky enough to partner with there, from marketing to legal to eng team was super knowledgeable about the product.
0
0
8
@deedydas It's only worth something if you act on it. I remember meeting the Robinhood team at a career fair and thinking they would never make it 😅.
0
0
7
1
0
8
Proof I'm still smarter than an AI.
1
0
8
As a wordcell by trade, there’s a few features that make Gemini Canvas pretty handy when writing docs 🧵.
1
1
8
This feature is available today on desktop and currently ramping up on mobile.
0
0
8
@omartosca Its only available if you are an Advanced user. You go to the 2.5 Pro model and click the "Deep Research" model in the input bar.
1
0
7
Love this! But my favorite feature was missed off the list:. 11. The "Google it" button - Bard can fact check itself.
Let's be honest. Bard has been lagging behind ChatGPT, Claude, and other LLM's for awhile now. But not anymore. With the Gemini update, I'm actually in awe of Bard's new capabilities. Here's 10 incredible new things you can do with Bard (for free):. —. 1. Craft Content with
1
1
6
@omartosca The model will try to ground as much as possible on the content it finds on the web.
1
0
5
Seeking technical cofounder: superconductor-based AI chips.
1
1
6
@MahawarYas27492 This is a cool query. Deep Research is really just focused on web research right now. For pure math and physics problems, Flash Thinking will be a better model to try. Your query is a bit of a combo - physics + web research. Will pass onto the eng team, hope to make better soon.
1
0
6
And Here. We. Go. Today Bard officially becomes Gemini (farewell Bard 🫡). We also launched our app for iOS and Android so you can use Gemini on the go. What’s more, we unveiled Gemini Advanced, our premium offering that gives you access to Ultra 1.0, our largest and most capable.
0
0
4
@GokulPrdeep @joshwoodward @phill__1 @GeminiApp Always working to reduce hallucinations. Feel free to DM me any examples btw and I'll pass on to the eng team.
0
0
4
@Presidentlin @OfficialLoganK @GeminiApp Not quite 🙂. It's folded into the main model, has access to the same apps and a bunch more (like Calendar, Notes, Flights) with the latest update.
1
0
4
So excited for this! The team is absolutely brilliant, and the product is profound. This is what "its time to build" actually looks like 🚀.
So we (@ImpulseLabs_ ) made the highest performing . stove. - 3x the performance of induction and ~5x gas (equiv. of ~72,000 BTU/h!). - A nontrivial advance in temperature sensing, transforming cooking. - And it ships this year (taking deposits starting . today).
0
0
5
@test_tm7873 @MahawarYas27492 @Jay_sharings @ai_for_success @SaiNemani1 @techikansh @slow_developer @CodeByPoonam @itsPaulAi @Angaisb_ @OfficialLoganK @joshwoodward Limits might change but currently at 20 a day. It’s more to prevent abuse.
1
0
4