Microsoft has introduced its new service “Correction,” designed to automatically review and correct AI-generated text that contains errors. The tool is available as part of Microsoft’s Azure AI Content Safety API. Although currently in a preliminary version, it is compatible with any text-generating model, such as OpenAI’s GPT-4 or Meta’s Llama.
Subscribe to the Softonic newsletter and get the latest in tech, gaming, entertainment and deals right in your inbox.
Subscribe (it's FREE) ►According to TechCrunch, among the errors that Correction would fix are misattributed quotes, which it would verify by comparing them with reliable sources. “Correction is powered by a new process of using small and large language models to align outputs with base documents,” explained a Microsoft spokesperson. According to the company, this tool can be especially useful “in fields like medicine,” where accuracy is crucial.
However, some experts, such as doctoral candidate Os Keyes, warn that this approach does not eliminate the so-called “hallucinations” of generative AI models. Keyes explains that these hallucinations are inevitable, as the models do not “know” anything, but rather predict words based on patterns.
The Microsoft tool is based on two metamodels: one that identifies possible hallucinations in the text and another that attempts to correct them based on reliable documents. Despite the progress it represents, Keyes expresses serious doubts about the success of Correction, as it “could also hallucinate.” Mike Cook, a researcher at Queen Mary University, agrees and adds that this tool “could give users a false sense of security.”
Although Correction has the potential to improve the reliability of AI-generated content for users and businesses that use generative models daily, its free use is limited to 5,000 records per month, after which 38 cents will be charged for every additional 1,000 records.
