Artificial intelligence, as is well known, tends to say what people want to hear and can often lie to do so. This situation is called “hallucination”. Artificial intelligence often cites non-existent academic studies and tries to convince users. In this situation serious problems can cause.

Microsoft believes it will fix this problemCorrection” He published his tool called. Of course, it is good that a step has been taken to solve this problem, but experts argue that the skeptical approach should continue. If we look into the details of the function, we see that the experts may be right.

“Error-prone” verification tool

Microsoft’s Azure AI Content Security APIThe first problem of the vehicle that came with it “can make mistakes” happened. This tool obtains the information in artificial intelligence from sources such as statements and assertions, instead of querying it from primary sources. New feature, Meta llama and Open AI GPT-4o It can also be used with other models.

Make a statement on the subject Microsoft, “Correction is based on the process of matching output to underlying documents using small language models and large language models,” He added that they hope this new feature will support productive artificial intelligence users and developers, especially in fields like medicine.

Google also has a similar tool Artificial intelligence development platform Vertex AI He brought it with him. Experts said that preventing hallucinations was “like removing hydrogen from water” and suggested that hallucinations were mandatory for artificial intelligence technology.