Google’s Gemini gets new Gems assistants, Imagen 3

Google has updated its Gemini large language model (LLM) with a new feature, dubbed Gems, that allows users to train Gemini on any topic of their choice and use it as a customized AI assistant for various use cases.

Gems, previewed at Google I/O this year, is currently available for Gemini Advanced, Business, and Enterprise users.

Gemini Advanced is a paid premium subscription of Gemini with the first month free presently. Alternatively, Business and Enterprise are two other modes to subscribe to or use Gemini.

Advanced, Business, and Enterprise subscribers of Gemini can create a new Gem or customized Gemini model by logging into their Gemini page and starting the process by locating the Gem Manager button.

Once a user has created a new Gem, they have to write instructions for it to follow, Dave Citron, senior director of product management at Gemini Experiences, wrote in a blog post.

The users need to detail goals, desired behaviors, and preferred format to help the Gem personalize Gemini’s responses, very similar to the practice of writing good prompts, Citron explained, adding that the main areas to be considered are persona, task, context, and format.

The instructions for Gem, according to the company, can be written with the help of Gemini itself.

“When you create your Gem, you can use Gemini to help re-write and expand on your instructions,” the company wrote in another blog post.

Further, Citron said Google has also launched a set of pre-customized and ready-to-use Gems that it thinks will aid various use cases. These include Brainstormer, Career Guide, Writing Editor, and Coding Partner.

While the Learning Coach Gem is aimed at helping users break down complex topics for easy understanding, Brainstormer is expected to provide users with inspiration, such as fresh ideas for a theme party, or a gift idea for a birthday.

Career Guide, as the name suggests, will help users unlock career potential with detailed plans to refine skills and achieve career goals, Citron said, adding that the Writing editor and Coding partner Gems will help elevate writing and level up coding skills of users respectively.

The Gem feature, according to the company, is currently being rolled out to desktop and mobile devices in more than 150 countries in most languages.

Gems can also be used to handle repetitive or tedious tasks, the company added.

The release of Gems comes at a time when rival LLM providers, such as Microsoft, Meta, AWS, and Anthropic among others, are not only offering similar services, such as Gems but are also racing to bring out their next-generation models to get a larger share of the chatbot market.

This week, The Information reported that ChatGPT-maker OpenAI may soon launch its next-generation LLM, codenamed Strawberry, in order to outpace rivals and boost its revenue from its chatbot business.

The rationale here is simple: better-performing models or models that accurately respond to queries will be used more and will be adopted faster than models from rival providers.  

This may help the Microsoft-backed startup ensure that it repeats its success with ChatGPT, which had helped the then GPT-3.5 and later GPT-4 models become popular.

OpenAI also introduced its GPT Store in January to provide nearly the same feature that Google’s Gemini Gems aims to provide. Users of the GPT Store can share custom versions of ChatGPT.

In May, Microsoft introduced a low-code toolkit, dubbed Copilot Studio, that would allow enterprise users to build AI agents tailored for specific use cases.

Other rival offerings of Gemini Gems include Meta’s AI Studio and to some extent Salesforce’s Einstein-powered agents. 

Gems is not the only feature that the internet giant is taking to boost its chatbot business. Google’s Logan Kilpatrick took to X this week to announce that the internet giant was dropping a few experimental models for enterprise users to try via the Google AI Studio. 

These models include — the smaller Gemini 1.5 Flash-8B, a stronger Gemini 1.5 Pro model, and a significantly improved Gemini 1.5 Flash model. 

The Gemini 1.5 Pro model, according to Kilpatrick, is better at coding and complex prompts.

Imagen 3 to boost Gemini’s capabilities further

In addition to Gems, Google said that it was adding Imagen 3 to Gemini Apps and expanding its availability across all languages in order to bolster Gemini’s capabilities and make it friendly for a variety of use cases.

Imagen 3, which is Google’s updated image generation model, according to Citron, comes with new safeguards.

Earlier in February, Google faced backlash over the AI system’s poor results in creating images of people.

“We don’t support the generation of photorealistic, identifiable individuals, depictions of minors or excessively gory, violent or sexual scenes,” Citron wrote in the blog post. However, he said that there were chances that Gemini may get some images wrong and the company would continue to listen to feedback and improve its image generation capabilities.