Of course, you can already use chatbots like ChatGPT online, but this tool is different in several ways.
Available in two sizes, Google says its new AI is open-source, lightweight and can run on computers. That means anyone can access its recipe book for free, and use its individual ingredients to create new apps and features on a personal laptop – and possibly even a phone.
The launch comes as Google is facing criticism over the racial inaccuracies made by its larger Gemini model. As AI continues to impact society, some are calling on tech heavyweights like Google to move away from a “black box” approach that keeps AI in the hands of a few, powerful firms.
Could Gemma prove to be a turning point for the company? Here’s what you need to know about the new AI system.
Gemma is the name of two open-source large language models. These are like the engines that power all-knowing chatbots, such as ChatGPT or Google’s Gemini.
Compared to those systems, Gemma is much smaller. If the most advanced version of Gemini is a Lamborghini supercar, then Gemma is a humble Volkswagen. That may explain why Google is letting the masses peep under the latter’s hood and not the former’s.
Of the two models, Gemma 2B has two billion parameters, which essentially tells you the amount of neurons in its AI brain. The more parameters a system has, the more complex tasks it can complete. Meanwhile, Gemma 7B is the larger of the pair with 7 billion parameters.
Notably, despite being smaller in stature, Google says its newcomer can trump more heftier models. Using Meta’s LLama-2 AI as an example, which has 13 billion parameters, Google claims Gemma is better at reasoning, arithmetic and coding, among other things.
Can anyone use Google Gemma?
Open-source AI is actually pretty common, and Google itself is no stranger to the concept. You can already use Stable Diffusion’s code to generate images, for instance, or build other tools from it.
For its part, Google has released several open-source AI projects that can help with everything from training other models to providing translations and computer vision. These have names like TensorFlow, Jax, Bert, and T5.
Chances are you’ve probably never heard of them, which is probably because they’re broadly aimed at AI researchers, developers, and all-round computer boffs. The same goes for Gemma; you may never use it unless you have some technical know-how, but a developer might.
To that end, Google says the new models play nice with popular AI tools like Hugging Face, MaxText and Nvidia’s NeMo. Due to their smaller size, they will probably be more useful for creating simpler chatbots for businesses and the wider public. Think customer service bots, virtual companions, or voice-controlled assistants and translators.
It won’t be a complete free-for-all, however. Google already prohibits users from leveraging its AI models for weapons development, for instance. Gemma will also come with a responsible AI toolkit that allows developers to create their own guardrails, including a list of banned words. Users will also be able to investigate and correct any behavioural issues. In case Gemma does a ChatGPT and starts spouting spanglish to english-speaking users.
If all this is giving you serious FOMO, there is an alternative. Just grab a Pixel 8 Pro, and you’ll be able to take the compact version of Gemini for a spin. The AI is baked directly into the phone, and can do things like summarise recordings and create replies to texts.