Name: Google 的 Tulsee Doshi：Gemini 3、Deep Think、Nano Banana Pro 的分層編輯 // AI Inside 110 (Google's Tulsee Doshi: Gemini 3, Deep Think, Layer-Based Editing in Nano Banana Pro // AI Inside 110)
Uploaded: 2026-01-20T01:15:39.000Z
Duration: 33 min 55 s

This episode of the AI Inside podcast is sponsored by your360AI get 10% off through January 2026.

AI Inside Podcast 本集由 your360AI 贊助，使用優惠碼可享 9 折優惠，優惠至 2026 年 1 月。

to 不定詞

By using code inside Google's head of Gemini product, Tulsi Doshi, reveals how user Demand is pushing nanobananapro toward Photoshop style layer editing capabilities, why DeepThink's most surprising application actually turned out to be creative writing rather than academic research, and how Gemini3's reasoning powers are reshaping everything from Geni email to Google search.

透過使用，Google Gemini 產品負責人 Tulsi Doshi 揭露了使用者需求如何推動 nanobananapro 朝向類似 Photoshop 的圖層編輯功能發展，DeepThink 最令人驚訝的應用實際上是創意寫作而非學術研究，以及 Gemini 3 的推理能力如何重塑從 Geni 電子郵件到 Google 搜尋的一切。

現在進行式

That's coming up on the AI Inside Podcast next.

更多精彩內容，就在下一集的 AI Inside Podcast。

Well, hello and welcome to another episode of the AI Inside Podcast, the show where we take a look at the AI that is layered throughout the world of technology.

嗨，歡迎來到另一集的 AI Inside Podcast，這個節目我們會深入探討科技世界中無所不在的 AI。

我是你們的主持人之一 Jason Howell。

As per usual with our interview episodes, he's going to join us for the interview.

就像我們平常的訪談節目一樣，他會加入我們進行訪談。

And speaking of the interview, just super excited to welcome Tulsi Doshi to the show.

說到訪談，我們非常興奮地歡迎 Tulsi Doshi 來到節目。

She is is senior director and head of Product for Gemini models at Google DeepMind, where she leads the roadmap and product strategy for Google's flagship AI foundation models.

她是 Google DeepMind Gemini 模型的高級總監兼產品負責人，負責 Google 旗艦 AI 基礎模型的路線圖和產品策略。

Now, Tulsi has been instrumental in bringing Gemini 3, also Nana Banana Pro and DeepThink, from research to production, and she's overseeing their integration across Google's ecosystem, from search to Gmail to NotebookLM, all over the map.

Tulsi 在將 Gemini 3、Nana Banana Pro 和 DeepThink 從研究推向實際應用方面功不可沒，她也負責將它們整合到 Google 的生態系統中，從搜尋、Gmail 到 NotebookLM，遍及各個角落。

Before leading the Gemini product team, she served as head of product for Google's responsible AI efforts, where she helped establish frameworks that now shape how Google builds inclusive AI systems at scale.

在領導 Gemini 產品團隊之前，她曾擔任 Google 負責任 AI 事務的產品負責人，協助建立了如今塑造 Google 大規模建構包容性 AI 系統的框架。

We cover a lot of ground in a short amount of time.

我們在短時間內涵蓋了許多內容。

Pleased to welcome Tulsi Doshi to the show and actually first time having Tulsi on this podcast.

很高興歡迎 Tulsi Doshi 來到節目，這也是 Tulsi 第一次上這個 Podcast。

But Jeff, you and Tulsi, you go way back to like the middle of last year.

不過 Jeff，你和 Tulsi 認識很久了，從去年年中就認識了。

我本來想說，是四個 Gemini 之前。

我的意思是，這絕對是一個重大的發布。

You know, obviously Gemini 3 released after that appearance.

其實 Gemini 3 是在那次露面之後才發表的。

That was just a couple of months ago and I had the opportunity.

那才幾個月前的事，我有機會。

Thank you, by the way, for sitting in a room for the friendsgiving.

謝謝你，順帶一提，你還來參加了 Friendsgiving。

我不知道你想怎麼稱呼那樣的聚會。

I feel like we don't get to do as many of those kinds of like smaller get togethers, which means you actually get just to have more conversation, which is good.

我覺得我們比較少有那種小型聚會了，這表示你真的可以有更多對話，這很好。

Well, and what I really appreciated about it is it was very human.

嗯，而且我真的很欣賞它的一點是，它非常有溫度。

It was a humanizing experience around this, like really this major Release moment.

這真是一次讓人感受到人情味、像是重大發布的時刻。

The Gemini team was right in the midst of a pretty crazy week.

Gemini 團隊正處於一個非常忙碌的時期。

The day before Friendsgiving was the release of Gemini 3.

在 Friendsgiving 前一天是 Gemini 3 的發布。

Now that we're on the other side, I mean, talking about Nano Banana Pro, because everybody I remember on the team was like, we're really excited, but we're cautiously optimistic about the release now.

既然我們已經度過了，我是說，聊聊 Nana Banana Pro，因為我記得團隊裡的每個人都說，我們真的很興奮，但對於這次發布，我們現在是謹慎樂觀的。

Were there any surprises along the way as far as that's concerned?

關於這點，過程中是否有任何令人驚喜的地方？

People are using these pretty strongly now.

人們現在已經很強勢地在使用這些了。

I. I think it's actually been really fun to see how much our team is even using nanobananapro for slide making and creation.

我。我認為看到我們的團隊甚至在製作幻燈片時，有多麼頻繁地使用 nanobananapro，這真的很有趣。

And I think actually it's been surprising how much we're seeing folks lean into that kind of creation format.

我認為，令人驚訝的是，我們看到大家有多麼傾向於那種創作形式。

So when you can actually get that level of text rendering quality, reasoning quality, visual quality, the kinds of use cases it unlocks, and just the comfort it unlocks, I think is really cool.

所以，當你真的能獲得那樣的文字渲染品質、推理品質、視覺品質，它所開啟的應用場景，以及它所帶來的便利性，我認為真的很酷。

I think when we met, we were 24 hours out from the launch.

我記得我們見面時，距離發布還有 24 小時。

I think we were still trying to make sure that everything was buttoned up and that we were actually going to be able to launch it across all of these surfaces.

我認為我們還在努力確保一切都已就緒，並且我們真的能夠在所有這些平臺上推出它。

And I think it's been really cool to see how it's playing out in slides, how it's playing out in the Gemini app Notebook lm.

而且我認為，看到它在製作幻燈片時的應用，以及在 Gemini app Notebook lm 中的應用，真的非常酷。

And I think we're learning a lot from what people are now asking for now that the model is out there.

而且我認為，我們從大家現在的需求中學到了很多，因為模型已經推出。

And so that feedback loop has also been, I think, really exciting too.

所以我認為，這個回饋循環也一直都非常令人興奮。

I'm super curious, like, what they're, you know, what they're getting out of this versus what they still want out of a Nano Banana Pro Pro or whatever it ends being called.

我非常好奇，像是他們從中獲得了什麼，又還想要從 Nana Banana Pro Pro 或任何它最終被稱為的產品中獲得什麼。

About as good as Google's branding as it's.

The things that we're hearing people love.

Are just like the fidelity of the visuals.

And I think what we're seeing a lot of people do is actually really reasoning heavy visuals.

我覺得我們看到很多人其實是在做很多需要深度思考的視覺內容。

If you can do infographics and you can also use these things for marketing materials and kind of more pro design.

如果你能做資訊圖表，也能用這些東西做行銷素材，還有比較專業的設計。

And I think where we're actually seeing a lot of demand is more and more leaning into that pro design set of use cases.

我覺得我們看到很多需求，是越來越傾向於專業設計這類的應用。

So, for example, how can you make some of these images more editable?

所以，舉例來說，要怎麼讓這些圖片更方便編輯呢？

Now you create an image and then you prompt it to iterate on it.

現在你建立一張圖片，然後再請它做迭代。

Are there ways that we can actually make it more layered and actually enable some of that more high fidelity editing, for example?

有沒有什麼方法可以讓它更有層次感，並且實際啟用一些更高品質的編輯，例如？

So I think those are some of the directions that we're now exploring and thinking about because we're hearing more and more of that kind of interest.

所以我覺得這些是我們現在正在探索和思考的方向，因為我們聽到了越來越多的這類興趣。

And then also of course there's always bugs with a model or you know, folks asking for better text rendering quality or better visual quality in certain ways that we're going to continue to iterate on too.

然後當然，模型總會有 bug，或者大家會要求更好的文字渲染品質，或是在某些方面更好的視覺品質，這些我們也會持續迭代。

One area we're also thinking about is just how do we continue to bring both Nano Banana and Nano Banana Pro, continue to improve those, the quality on both fronts.

我們也在思考的一個領域是，我們要如何持續將 Nano Banana 和 Nano Banana Pro，持續改進這兩者，在兩個方面的品質。

And so we're just going to keep, keep iterating on that, which will be good.

所以我們會持續、持續地迭代，這會是很好的。

Tulsi, I've seen obviously that you've been integrating especially Gemini 3 into other Google products.

Tulsi，我當然看到你特別將 Gemini 3 整合到其他 Google 產品中。

Gmail last week said we're putting Gemini 3 all around.

Gmail 上週說，我們將 Gemini 3 佈局在各處。

I'm curious what the frontiers are for integrating both Gemini 3 and Nano Banana into other Google products.

我很好奇將 Gemini 3 和 Nano Banana 整合到其他 Google 產品的邊界是什麼。

And then just to ask two more questions past that, do you think we get to a point where we radically change the UI of some of these products, that we, we interact with them very differently with search or Gmail or whatever and also whether we're going to see these integrations outside of Google in others products?

那再問兩個問題，你覺得我們會不會進入一個階段，讓這些產品的 UI 大幅改變，我們跟搜尋或 Gmail 等產品的互動方式很不一樣，另外，我們是不是也會看到 Gemini 整合到 Google 以外的其他產品裡？

Yeah, I mean I think more generally one of the big themes of Gemini 3 is Gemini everywhere and Gemini where you are meeting you, where you are improving your workflows, where you are.

是的，我認為更廣泛來說，Gemini 3 的一個重要主題就是「Gemini 無所不在」，Gemini 會在你身邊，在你工作的地方，在你需要的地方，來改善你的工作流程。

And so I think that's true both for Google products and for products outside Google.

我認為這對於 Google 產品和 Google 以外的產品來說都是一樣的。

So one of the things we were excited about when we launched Gemini 3 is other products also adopting Gemini.

所以，我們在推出 Gemini 3 時很興奮的一件事是，其他產品也能採用 Gemini。

Whether that is code and some of the IDEs offering Gemini for coding purposes or whether that is a bridgewater using Gemini for particular use cases, we want to make sure that we're actually supporting Gemini at a larger scale.

無論是在程式碼和一些 IDE 中提供 Gemini 來進行程式開發，或者像 Bridgewater 這樣使用 Gemini 來處理特定用途，我們都希望確保我們能真正大規模地支援 Gemini。

One is from a use case perspective, where can we bring Gemini to be able to do more?

一點是從使用案例的角度來看，我們可以在哪裡引入 Gemini 來做更多事情？

And then also where does that actually then change the experience?

然後，這又會如何改變使用者體驗呢？

And I think one thing that Gemini already has pushed on is this idea of richer ux.

我認為 Gemini 已經推動了一件事，那就是更豐富的使用者體驗。

So the traditional for example search experience is you type a query, you get text, and that is still an extremely helpful and useful use case.

所以，傳統的搜尋體驗，例如你輸入一個查詢，得到的是文字，這仍然是一個非常有幫助且實用的使用情境。

But with Gemini 3, what we can start doing is pushing towards more richer layouts, interactive layouts, layouts that give you content and info in the modality you want it.

但有了 Gemini 3，我們可以開始推動更豐富的版面配置、互動式版面配置，以及以你想要的模式提供內容和資訊的版面配置。

So maybe something is better explained in the context of an image.

所以，有些東西用圖片來解釋會比較好。

Maybe something is better explained in the context of an interactive widget.

有些東西用互動式小工具來解釋會比較好。

Maybe something is better explained in the context of a video.

有些東西用影片來解釋會比較好。

I see us moving towards this direction where you get the content in the way that it makes sense for the question that you're asking, or maybe even in the way that it makes sense for you personally.

我看到我們正朝著這個方向邁進，也就是你會以對你提出的問題有意義的方式，或者甚至是以對你個人有意義的方式來獲取內容。

There's a lot that we can start playing with now as these models are able to bridge that gap, which I think is really exciting.

隨著這些模型能夠彌合差距，我們現在有很多東西可以開始嘗試，我認為這真的很令人興奮。

Google 的 Tulsee Doshi：Gemini 3、Deep Think、Nano Banana Pro 的分層編輯 // AI Inside 110 (Google's Tulsee Doshi: Gemini 3, Deep Think, Layer-Based Editing in Nano Banana Pro // AI Inside 110)

字幕列表影片播放

Google 的 Tulsee Doshi：Gemini 3、Deep Think、Nano Banana Pro 的分層編輯 // AI Inside 110 (Google's Tulsee Doshi: Gemini 3, Deep Think, Layer-Based Editing in Nano Banana Pro // AI Inside 110)

perspective

episode

context

awesome