Research to Reality with Google DeepMind — Paige Bailey, Google DeepMind

AI Engineer · 10 min · 84 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

  1. 00:12trickle in from other locations and then we'll get started pretty shortly.
  2. 00:16Um, ideally there will be some time for lots and lots of live demos.
  3. 00:20Um, so you'll ideally learn some of the things about our new models, how to use them as part of your projects.
  4. 00:27Um, and then there also should be time for questions, but it also seems like we'll have a small enough audience where if you ask questions throughout the duration of the presentation, we can also do that.
  5. 00:37Um, so end goal is to make this as useful as possible for all of y'all.
  6. 00:42Um, my name is Paige.
  7. 00:43I work at DeepMind and excited to show you all what we've been working on too.
  8. 00:54And as mentioned, we'll get started in just another couple of minutes.
  9. 00:57It looks like people are still getting scanned at the door.
  10. 02:38All right, let's get started.
  11. 02:41Cool.
  12. 02:41So, greetings everyone.
  13. 02:42As mentioned, my name is Paige.
  14. 02:44If you have questions throughout the duration of the presentation, please feel free to raise your hand and shout them out.
  15. 02:49Um, ideally, we'll be able to take some uh take some as I kind of carine along doing live demos.
  16. 02:56Um, I also want to give a kind of the requisite caveat at the very beginning that um, DeepMind's entire mission is to build AI responsibly and for the benefit of humanity.
  17. 03:08So trying to cast as much light as we can um in this uh in this world that increasingly has uh increasingly has shade.
  18. 03:18Um this translates to things like alpha folds um much of the work that we do with our open models including medge gemma um and a lot of our robotics and science use cases
  19. 03:29as well as things like AI for math AI for frontier research.
  20. 03:33Um, so in general, everything that I show you today, um, is being used for those projects that are deeply deeply embedded in AI for science.
  21. 03:42Um, and if any of y'all work in the AI for science space, um, please feel free to send questions and ask afterwards about how you can use AI to to kind of transform and accelerate that uh that good work.
  22. 03:55Um, I don't think it's a secret that Google has been a little bit busy over the course of the last few months.
  23. 04:00um feels like we've been releasing new models, new features, new products every single week.
  24. 04:06Um most notably, uh just recently we released a computer use API which we'll be talking about a little bit.
  25. 04:12um something called managed agents which gives you the ability to take um kind of a higher order task uh describe it in a natural language and have a fleet of agents execute on it in a Linux workstation
  26. 04:24like a sandbox environment where you can add skills you can add um kind of dependencies that get pulled in along the way and a whole bunch more.
  27. 04:33Um and then also our speechtoech translation API which we'll take a look at in a second.
  28. 04:38And one of the things that I really really love about Google is that not only are we shipping Frontier models.
  29. 04:45So you might have heard about Gemini 3.5 Flash which got released at IO.
  30. 04:50Um but we've also been focused pretty significantly on our open model releases.
  31. 04:55Um so how many folks in the room have heard of Gemma 4?
  32. 04:59Um quite a few hands.
  33. 05:01That's excellent.
  34. 05:02Uh Gemma 4. Oh yes, absolutely.
  35. 05:05Go for it.
  36. 05:06Um, Gemma 4 is our open model family.
  37. 05:09It's the latest iteration.
  38. 05:11Um, we have many different sizes available.
  39. 05:14So, there's a two billion parameter, a 4 billion.
  40. 05:16We just released a 12 billion parameter.
  41. 05:19Um, and then we also have a 26 billion and a 31 billion um, mixture of experts and dense model um, respectively.
  42. 05:27Um, they're useful for a lot of things and they're also Apache 2 licensed which means you can download them, use them as part of your company, fine-tune them, uh, and kind of, uh, expand on them however you feel like would be most useful.
  43. 05:41Um, and this Gemma universe is actually pretty cool to see, um, in terms of people building.
  44. 05:48I I hate slides, so we're going to see how few slides I can get through today.
  45. 05:54Um but uh this is an example of something that someone from the community has built using Gemma 4 um and Fable 5 before it was taken off the market to rewrite some of the kernels.
  46. 06:04Um you can load the model directly within the browser.
  47. 06:08So this is loading Gemma directly in the browser using it via web assembly.
  48. 06:12It's sandboxed.
  49. 06:14Um and then it can do things that that feel pretty magical, right?
  50. 06:18So this model is uh this model is currently running completely locally.
  51. 06:23Um and if I say something to the effect of create a table with emoji comparing and contrasting all of the uh Harry Potter books um based on which are the funniest
  52. 06:42and most exciting.
  53. 06:44um make sure to give me recommendations on which to read and also um um maybe incorporate Harry Potter Harry Potter in the methods of rationality um or and then
  54. 07:06you get kind of a a response that feels almost instantaneous.
  55. 07:11Um, you have uh kind of the Deathly Hollows, Half Blood Prince, Sword or the Phoenix Goblet of Fire, etc. Um, I'm not sure if I agree with the ranking,
  56. 07:21but that's uh but that's a pretty interesting pretty interesting comparison.
  57. 07:26Um, and then also uh since it is local um to the machine, none of your data is getting sent elsewhere.
  58. 07:33This is not using an API.
  59. 07:35It's just something that's running locally in the browser with transformers.js um and the Gemma 4 model.
  60. 07:41Gemma 4 is also pretty cool in the sense that you can use um Google AI edge gallery in order to analyze um some of the model capabilities that we have on device.
  61. 07:52You can download it and use it with Android and with iOS.
  62. 07:56Um and it includes everything from kind of taking images and describing them to automatically transcribing audio in multiple languages.
  63. 08:04I believe Gemma supports over 140 different languages.
  64. 08:08Um, and then also testing it out with function calling directly on device.
  65. 08:13If you haven't had a chance to take a look at the Google AI edge gallery, um, we have a collection of skills as well.
  66. 08:20Um, so things like building games, doing haikus, um, asking queries about weather, being able to schedule events on your calendar for you, um, that are all available to use just with this AI edge gallery app completely for free and just with locally installed models.
  67. 08:36Um, so if you haven't downloaded it, definitely try it out.
  68. 08:40There's also a way to use um the the kind of accelerator on your local device in order to power the model and do all of the inference work as opposed to just the fall back to the CPU.
  69. 08:52Um, so if you do have something like a Pixel 10 or a higherend um a higherend mobile device, you're already able to use Gemma kind of locally um for for all of that work, which is quite cool.
  70. 09:07Gemma 4 just to to give a recap or to place how the model performance versus size shapes up.
  71. 09:15Um this is the two of the two kind of largest versions of Gemma that I had mentioned before the 31B and the 26B.
  72. 09:23Um they're quite small in comparison to some other models that are on the market but they're performing way above what you might expect.
  73. 09:32um so more than models that are in order of magnitude or larger than their size.
  74. 09:37This is great because it kind of translates into a cost savings perspective.
  75. 09:42You don't have to worry about distributed inference for local models.
  76. 09:46Um you don't have to worry about as large of a GPU footprint.
  77. 09:49Um and it's really intended to run super super well on a single commodity GPU as opposed to needing something a little bit more fully featured.
  78. 09:57You can also run Gemma, some of the variants on even things like Jets and Nanos.
  79. 10:02Um, and for 12b and below, you can run it locally on your laptop.
  80. 10:06Um, 2B can even fit handily on mobile devices.
  81. 10:10We've also released quantized versions.
  82. 10:12So, the the model that we saw at the very beginning that was running so so blazingly fast um was using some of the QAT checkpoints that we have for Gemma.
  83. 10:21Um and the QAT checkpoints um are quite small when you uh when you take a look even less than a single gigabyte in size for the the uh kind of two billion parameter version.
  84. 10:49[music]