Google added a new agentic feature to Gemini on Wednesday. Dubbed Deep Research, the feature was introduced alongside the new Gemini 2.0 family of AI models. It can create multi-step research plans, run web searches, and prepare a detailed report on complex topics. The Mountain View-based tech giant says the feature can be useful to researchers and students who have to write research papers. The new artificial intelligence (AI) feature is available to the Gemini Advanced subscribers on the web version of the chatbot.
Gemini Gets New Deep Research Feature
Advanced reasoning has become a topic of interest for AI firms who are trying to increase the intelligence and processing power of their AI models. While upgrading the analytical capability of large language models (LLMs) will require reimagining the network architecture and learning algorithms, researchers have developed some shortcuts to somewhat achieve this.
One popular method, which is used by OpenAI’s o1 models and the recently released models by Alibaba increases the compute time, allowing the AI to spend more time on the same question. This allows the AI to verify its answers, consider other alternatives, and eventually generate more detailed responses. Another way to do it is via AI agents, which Google did with the Gemini AI model.
The Deep Research feature is essentially an agentic feature. Once a user adds a query about a complex topic, the AI agent begins creating a multi-step research plan. A typical plan includes steps such as breaking down the topic into smaller parts, finding relevant research papers and articles on the topic, looking into trends, future impacts, and related areas.
Gemini’s Deep Research agentic feature
Photo Credit: Google
Once the research plan is ready, users can intervene and add or remove steps or change them to better target the topic. After that the AI begins executing the plan and starts researching all the information. The agentic feature can also run several new web searches based on its learnings to understand additional topics.
After that, it analyses the gathered data, creates a detailed report and shows the output to the user. These responses will generally be more in-depth and detailed than a typical response from Gemini. Google says this is ideal for market researchers, academicians, research students, and even entrepreneurs.
Currently, Deep Research is available in Gemini Advanced on the web version of the chatbot in the English language. Google highlighted that the feature will be made available in the mobile app in early 2025. Deep Research can be found in the model selector drop-down menu and is listed as “Gemini 1.5 Pro with Deep Research”.