Gemma Uncovered: Google's Open AI Endeavor

0:00What can 160 years of experience teach you about the future? When it comes to protecting what matters, Pacific Life provides life insurance, retirement income, and employee benefits for people and businesses building a more confident tomorrow. Strategies rooted in strength and backed by experience. Ask a financial professional how Pacific Life can help you today. Pacific Life Insurance Company, Omaha, Nebraska, and in New York. Pacific Life and Annuity, Phoenix, Arizona. Google has announced a bunch of new lightweight AI models that are open source. This is super exciting, something that I've been saying they should do for quite a while.

0:40So today on the podcast, we're gonna be diving into their new Gemma models. They have two of them that they've recently announced. And where I think this is going in the future, there's a lot that is going on in this open source space. And I think that it is important. There's a bunch of things that I think people are definitely not paying attention to. Before we get into that, I want to say there are only two days left in the crowdfunding campaign for AI Box, my very own AI app builder and marketplace. If you're interested in how much money we have raised, or if you're an investor, even if you're not, go over to Republic and check out our crowdfunding page.

1:15See what we're currently at. Drop us a comment over there for some encouragement as we're finishing off this campaign and getting really excited to launch our new product, which is a no-code AI app builder to help you automate virtually everything you do with AI tools. Now let's get into the podcast. So Google right now has unveiled a new set of AI models, right? So they have, they're both called Gemma, but they have Gemma 2B and Gemma 7B. There's the two versions. This is kind of classic what we see with these open source models where there's, you know, a larger and a smaller one. So this is all just a couple weeks after the launch of their latest Gemini models.

1:57And these new models are described as kind of lightweight or openweight models that get a lot of their inspiration from Gemini, but they're available for both commercial and research purposes, which I think is really cool. And I think important, you know, if these things are going to be viable, that they are available to be used commercially. This is, you know, something. I'm excited to put both of these models onto the AI box platform so people can build AI tools with them, try them out, use them. And I think there's a lot of really awesome things about these kind of open source models where essentially you're able to lower costs, you get them for free, and some of them do have special capabilities that are really interesting.

2:38So unlike the rivals Meta and Mistral, Google has not yet given exact detailed performance comparisons, but they are labeling the Gemma models as quote unquote state of the art. I know that could just be kind of a buzzword, but the models are built on a dense decoder only architecture. So this is similar to both Gemini and also early Palm models. They have performance benchmarks that are expected to be shared on Hugging Face's leaderboard soon. But for developers that are interested in getting into Gemma. Google right now is giving a ready-to-use resources such as Colab and Kaggle notebooks alongside integrations with platforms like Hugging Faces, MaxText, and NVIDIA's Nemo.

3:27And all of this is to really just make it really easy to access and implement across a bunch of different environments, testing and using these. So despite highlighting the quote-unquote open nature of these models, Google clarifies that Gemini or that Gemma is not open source. So I know I've been calling it open source. You know, that's not entirely accurate. It's not actually open source. They're just saying that it is an open, it has an open nature, right? Whatever that means. So Jeannie Banks, who is from Google, kind of emphasizes the distinction between open models and open source. And this is how Jeannie is kind of clarifying or defining this.

4:08She said, quote, open models have become pretty pervasive now in the industry. And it often refers to open models or open weights models, where there is wide access for developers and researchers to customize and fine tune models. But at the same time, the term, the terms of use, things like redistribution, as well as ownership of those variants that are developed vary based on the model's own specific terms of use. And so we see some differences between what we would traditionally refer to as open source, and we decided that it made the most sense to refer to our Gemma models as open models. So Google says that the Gemma models sizes are, you know, versatile enough to, you know, work for a huge range of use cases.

4:57There was Tris Warnkins of Google DeepMind who kind of was highlighting some of the improvements in generation quality saying, quote, the generation quality has gone significantly up in the last year. Things that previously would have been remit of extremely large models are now possible with state of the art smaller models. This unlocks completely new ways of developing AI applications that we're pretty excited about, including being able to run inference and do tuning on your local developer desktop or laptop with your RTX GPU, or on a single host in GP or GCP with cloud TPUs as well. So I think all of this is super, super interesting.

5:41Right now is I think we see kind of the AI industry continuing to evolve. There's a lot of real world kind of performance of Google's Gemini models that has yet to be seen. I think this is especially in comparing it with, you know, ChatGPT and Claude and Mistral and like a lot of the main competitors that we're seeing. So I think Google's Gemini needs to really be put through some benchmarks, some benchmarking tests and compared so we really know kind of where it stands. But I think right now this is definitely kind of a move to bring in some interesting advancements in AI development. I think Google is also releasing a quote-unquote responsible generative AI toolkit that's coming out with the Gemma models.

6:31And the toolkit is designed to essentially give some guidance and some essential tools for creating more secure AI applications. and I think this is making some big steps towards addressing some of the ethical considerations in AI deployment and development or at least that's what Google is telling us at this time and saying what they think is important. So a phenomenal story, super excited to see where this goes and don't forget to check out the Republic crowdfunding campaign to see how much we have raised with only two days left to go. We're excited to finish strong and launch the AI Box platform.

7:04Thanks so much for tuning into the podcast and have a fantastic rest of your day.