Microsoft is introducing a new feature called “correction” aimed at enhancing the accuracy of AI outputs. Customers utilizing Microsoft Azure for their AI systems can now automatically detect and rewrite incorrect content. This capability is currently in preview as part of the Azure AI Studio, a comprehensive suite of safety tools designed to identify vulnerabilities, detect “hallucinations,” and prevent malicious prompts.
Once activated, the correction system analyzes AI outputs by comparing them against a customer’s source material. It highlights inaccuracies, explains why they are incorrect, and rewrites the content before the user sees the mistake. While this feature offers a promising solution for addressing inaccuracies in AI responses, it may not be foolproof.
Similarly, Vertex AI, Google’s platform for AI development, includes a feature that verifies AI outputs against Google Search, client data, and soon, third-party datasets. A Microsoft spokesperson explained that the “correction” system employs both small and large language models to align outputs with source documents. However, they cautioned that this approach does not guarantee complete accuracy, as it primarily focuses on aligning generative AI outputs with grounding documents.