The large language model (LLM) APIs from OpenAI, Anthropic, and Google offer acceptable responses most of the time. But they all fail miserably when it comes to responses about recent happenings. Since these LLMs are trained on a certain dataset, they have a knowledge cut-off date. To avoid this problem, Google today announced a new feature in Google AI Studio and the Gemini API that will allow users to ground their responses with Google Search.
The new Grounding with Google Search feature will enable developers to get more accurate and fresh responses from the Gemini LLMs. The best part of this new feature is that the model even provides grounding sources (in-line supporting links) and Search Suggestions corresponding to the grounded response.
This new Grounding with Google Search feature is supported with all generally available versions of Gemini 1.5 models. However, it is quite costly. Developers need to pay $35 per 1,000 grounded queries. To enable this feature, developers can go to the "Tools" section in Google AI Studio or enable the "google_search_retrieval" tool in the API. As always, this new Grounding feature can be tested for free in Google AI Studio.
Google recommends developers use this new feature in the following cases:
- Reduced hallucinations: Grounding helps ensure that AI applications provide users with more factual information.
- More up-to-date information: With grounding, models can access real-time information, making AI applications relevant and applicable to a wider range of scenarios.
- Enhanced trustworthiness and traffic to publishers: By providing supporting links, grounding brings transparency to AI applications, making them more trustworthy and encouraging users to click on the underlying sources to find out more.
- Richer information: By drawing information from Google Search to enhance the model response, grounding can provide richer color on many queries.
When this new feature is enabled, and a user query is sent to the Gemini model API, the API will use Google"s search engine to find up-to-date information relevant to the query and send it to the Gemini model. The model will then respond with more accurate and up-to-date information.