# DeepMind’s New AIs: The Future is Here!

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=lgsD_wSZ0hI
- **Дата:** 13.03.2025
- **Длительность:** 6:07
- **Просмотры:** 154,092
- **Источник:** https://ekstraktznaniy.ru/video/12546

## Описание

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambdalabs.com/papers

Guide for using DeepSeek on Lambda:
https://docs.lambdalabs.com/education/large-language-models/deepseek-r1-ollama/?utm_source=two-minute-papers&utm_campaign=relevant-videos&utm_medium=video

📝 The Gemma 3 paper and the rest are available here:
https://blog.google/technology/developers/gemma-3/
https://developers.googleblog.com/en/experiment-with-gemini-20-flash-native-image-generation/
https://deepmind.google/technologies/gemini-robotics/
https://aistudio.google.com/

Sources:
https://x.com/thepushkarp/status/1899874826669744425/photo/1
https://x.com/Angaisb_/status/1899852603107721388
https://x.com/alexanderchen/status/1900013570575794414

📝 My paper on simulations that look almost like reality is available for free here:
https://rdcu.be/cWPfD 

Or this is the orig. Nature Physics link with clickable citations:
https://www.nature.com/articles/s41567-022-01788-5

🙏 We would like to thank our gener

## Транскрипт

### Segment 1 (00:00 - 05:00) []

I am really surprised today, because Google  DeepMind just released their new Gemma 3 AI,   but Gemma 2 before was okay, but nothing  spectacular. Now check this out. Wow. It seems   nearly as good as the full-size DeepSeek, also  anbut look. Running the full DeepSeek typically   requires a bunch of graphics cards, but this one,  only one. And an open model. This is unbelievable. We are also going to look at 2 other stunning  things they released lately, including generated   images that are just fantastic, and a  new robot that can pack lunch for you. Now, Gemma can also look at images,  for instance, if you have a huge check,   but you would only like to pay for your part, with  a nice 18% California tip, it can calculate it for   you. It is an absolute killer at creative  writing, getting second place globally. Now just to demonstrate how much of a stunner  this is. This was Gemma 2, small, not super smart,   then came the big guns like Llama and Deepseek,  super smart but really big, and now, Gemma 3 gives   you nearly that kind of quality, but it is 20  times smaller. And it can still keep up with them. This is an amazing gift for all of us. This gift  comes in 4 sizes, most speak 140 foreign languages   and can have a look at images. So if you have a  Japanese remote controller, and you wish to put   it in heating mode, it will be able to tell you  how. So cool. For the smallest model, no images   and only English, but in return, it almost runs  on a toaster if you got a good one. Where +1,   ShieldGemma can evaluate the safety of text  and images if you define a set of policies. You can try them in Google AI studio,  the link is in the description,   or if you wish to run it yourself like I do,  it is super easy to do in Lambda GPU Cloud. But it gets better. They also  showcased conversational image   generation. Regular image generation is  not that interesting, anybody can do that,   however, here you chuck in an input image, and  ask that it adds flowers to that table. Now,   with previous techniques, the problem  was that we got the flowers, but the   scene changed a lot. So what I wanna see here  is…oh my, that is exactly what I wanted to see.    Same scene. And you can keep iterating on it, if  you wish to see tulips instead, not a problem. Now there are some game changers that you can  build on top of this. For instance, recipes. You   can ask for a recipe for chocolate chip cookies,  but after each step, you get an image of what you   should be seeing. And all this is generated on  the fly. Incredible. You Fellow Scholars are also   using this to great effect. Adding some chocolate  drizzle here, yummy. Not a problem. And it really   is the same image. This is not amazing, this is  beyond amazing. I’ve never seen anything like it. Now I would also like to see some images  created with text in them. Something that   works kind of okay with a word or a couple of  words with other systems, so let’s see…wow,   that is superb. And some of you Fellow  Scholars are already using this here. Now pair it up with their Imagen 3 AI  image generator, which is one of the best   around and rivals many other top tier paid  models. I mean, just look at these images,   huge text prompts are okay, and it even has a  hint of creativity. I am kind of blown away. So really, what happened at Google  DeepMind? Imagen 2 was good, not the best,   but Imagen 3 is absolutely fantastic. Gemma 2 was okay, not the best,   but Gemma 3 is incredible. I mean, look at this  one from the paper. 3 is better in every way,   and not by a little. And the list goes  on. Something must have happened at   Google DeepMind lately, because they  are coming out with these bangers. And now, the new robot. I had the honor of having  a look at this before release. I will note that   we have no business relationship with Google  DeepMind, and of course, I always say that I   only do it if I can say whatever I want about  it. The Papers are beholden to no one. Ha! You see, last time I was there, they  had these robots playing football,   and they won the world championship,  but its a world championship for robots,   they are not Ronaldo, not even close. It was a  very impressive project, until you see this one. Now, the best features in my opinion. It runs  in real time and reacts to the world changing   around it. Or in other words, you can troll  it all you want. Super fun. It is also good   at high-dexterity tasks, which is pretty stunning.   I think of robots folding laundry as a thing that

### Segment 2 (05:00 - 06:00) [5:00]

is perhaps a decade away, and when I see this…wow.   And finally, it generalizes to new tasks. That is   what we want from an intelligent being. It  was asked to slam dunk this ball here, well,   it’s not Michael Jordan, but it gets  the job done. Absolutely incredible. And yes, it packs lunch too. Dear Fellow  Scholars, this is Two Minute Papers with   Dr. Károly Zsolnai-Fehér. So, what do you  think? Let me know in the comments below.