In order to avoid "hallucinations" in the operation process of artificial intelligence, Google announced that it would add a hybrid comparison of Google search and third-party data to its Vertex AI service in Google Cloud.Basic comparison function, to prevent artificial intelligence from conveying wrong information.
In fact, at the Google NEXT'24 event this year, Google proposedSearch Enhancement Generation The retrieval augmented generation (RAG) function allows services that integrate the Gemini artificial intelligence model to output interactive results by comparing data results and old and new data through Google searches, thereby reducing the common "hallucination" phenomenon of artificial intelligence.
This update strengthens the verification of generated data through mixed comparison of Google search and third-party data, as well as expanding vertical comparison search, so that the data automatically generated by artificial intelligence is consistent with the truth.
This approach will simply reduce the "hallucinations" produced by artificial intelligence, but it will also increase the computing cost of artificial intelligence operation. Therefore, in subsequent practices, Gemini will dynamically decide whether to search and compare data through Google. However, the prerequisite is that Gemini must be based on sufficient training. Only when the answers generated are beyond its knowledge scope will they need to be verified through Google search and third-party data comparison, so as to achieve a balance between response speed and computing cost.
Google will also use high-fidelity matching to generate data based on content within specific conditions. For example, this could generate authentic content for strictly regulated financial, medical, or insurance businesses, or even generate more accurate content by referencing specific data. For specific application scenarios, Google will expand vertical search methods to enhance data matching results, enabling AI to be used for services such as specific content recommendations and advertising.
The new version of Vertex AI is now open for public beta testing, and we plan to collect more feedback from developers and corporate testers for subsequent adjustments.
