Google’s Customizable AI Gems Are Coming. Here’s What You Need to Know [CNET]

Google’s Gems, a feature that lets you customize the company’s Gemini chatbot as an AI expert in any topic you choose, will soon be available for Gemini Advanced subscribers.

Google made the announcement in a blog post on Wednesday, which also said its latest image generation model, Imagen 3, is coming soon for all Gemini users.

These features were previewed at Google’s I/O 2024 developers’ conference in May where AI was the dominant theme. Google declared this “the Gemini era” as it showcased updates designed tol make its Gemini chatbot more conversational and smarter in general. But, just as Google trailed behind ChatGPT maker OpenAI in the launch of a generative AI chatbot, it is once again following OpenAI’s lead in rolling out the ability to make custom versions of a bot.

Gems

Gemini Advanced subscribers, who pay $20 a month for access to the Gemini 1.5 Pro model, will be able to access Gems “over the coming days,” Dave Citron, senior director of product management at Gemini Experiences, wrote in the post.

Gems are meant to serve as experts on topics or as cheerleaders for goals like thinking through a project, brainstorming ideas for an event, writing captions for social media posts or serving as a gym buddy.

To create a Gem, you give it instructions, name it and chat with it. The Gem will remember the instructions, so you don’t have to start from square one each time you interact as you would with a traditional chatbot, Citron said.

Google is also releasing premade Gems, which include a learning coach, a brainstormer, a career guide, a writing editor and a coding partner.

Gems will be available in 150 countries in “most languages,” according to the post. They will also be available to business and enterprise users.

OpenAI announced its own version of a customized chatbot, GPTs, in November.

As with  Gems, with OpenAI’s counterparts you can create GPTs for specific purposes, like learning the rules to any board game, teaching your kids math or designing stickers. To start, you give it instructions and pick what it can do and you’re ready to go. OpenAI’s GPT store now has millions of GPTs, including a Canva design tool, a fitness trainer and a haiku writer.

Imagen 3

Google also announced that its Imagen 3 image generation model will be available to all Gemini users over the coming days.

At I/O, Google said this new model can create more photorealistic images, including details like sunlight or whiskers on animals. It also remembers to include specifics like wildflowers or birds in longer prompts. Plus, Imagen 3 can generate styles “like photorealistic landscapes, textured oil paintings or whimsical claymation scenes,” the post said.

Citron pointed out that Google has made “significant progress” in generating images of people with Imagen 3. However, the model will not generate photorealistic images of identifiable individuals, minors or excessively gory, violent or sexual scenes.

Imagen 3 uses the SynthID tool for watermarking AI-generated images.

Gemini Advanced

In addition to access to Gems and Gemini 1.5 Pro, Gemini Advanced subscribers have a 1 million token window, which CEO Sundar Pichai in May said will increase to 2 million tokens later this year.

Large language models like Gemini break words down into tokens in order to make sense of our queries. The tokens in the context window help the model remember things. The more tokens it has, the better its memory will be. According to Google’s figures, the 1 million token context window is capable of ingesting 1,500 pages of text.

However, in his review of Gemini Advanced, CNET’s Imad Khan said the model was plagued by errors and not a worthy competitor to OpenAI’s ChatGPT Plus subscription, which is also $20 per month.

Other Services & Software