To prevent "nonsense", Apple's Apple Intelligence smart reply function adds "anti-hallucination" instructions

Recently, some netizens found in the macOS 15.1 beta versionappleCompany Apple Intelligence The Smart Reply feature of the .NET Framework is a set of internal instructions that are set in the form of JSON files, which specify the workflow of the feature in detail. The most notable of these is:Apple explicitly requires the system to "not fabricate facts, not create hallucinations".

To prevent "nonsense", Apple's Apple Intelligence smart reply function adds "anti-hallucination" instructions

It is understood that Smart Reply is a feature of Apple's Mail app that aims to automatically generate possible reply options by analyzing the content of the email. Behind this feature is Apple's self-developed intelligent technology. However, like other generative AI, Smart Reply also faces the problem of "hallucination", that is, the system may generate false or misleading information.

To prevent "nonsense", Apple's Apple Intelligence smart reply function adds "anti-hallucination" instructions

To prevent this from happening, Apple has set strict instructions within the system. According to the leaked instructions, the smart reply function is required to only extract questions explicitly asked in the email and provide corresponding answer options. This practice helps reduce AI errors when generating replies.

Although Apple explicitly requires AI not to fabricate information, industry insiders believe that it is still challenging to completely eliminate the "hallucination" problem because the generative AI model does not understand the content of the text and cannot accurately judge the authenticity of the information.

statement:The content is collected from various media platforms such as public websites. If the included content infringes on your rights, please contact us by email and we will deal with it as soon as possible.
ActivityInformation

OpenAI announces the 2nd DevDay Developer Conference in October: no new AI models, focus on APIs and development tools

2024-8-7 9:17:45

Information

Several U.S. states sent letters asking Musk's Grok chatbot to stop spreading false election information

2024-8-7 9:19:44

Search