Last June, Apple lifted the veil on Apple Intelligence, its new vision of AI. This revolution should notably benefit Siri, the iOS voice assistant, with the integration of ChatGPT for more relevant responses. Strong expectations, fueled by the tempting demonstrations at WWDC 2024.
Last August, however, you could already discover the first impressions of the new Siri with Apple Intelligence. The beta version of iOS 18.1 finally allowed developers to test this major development presented during WWDC 2024. At the time, 9to5mac journalist Fernando Silva already shareda mixed observation on the limited performance of the voice assistant.
This first analysis was unfortunately only a foretaste of the disappointments to come. More in-depth tests carried out in recent days reveal an even more worrying situation.The new Siri indeed seems to have regressedon certain crucial aspects of its operation.
Catastrophic results
American programmer and blogger John Gruber, creator of the Daring Fireball site and co-creator of the Markdown language, with a friend, have just put Apple's voice assistant to the test. The test on the results of the Super Bowl is final!
On58 questionssimple questions asked to Siri,only 20 answers were correct.Ce66% error rateis all the more worrying since other assistants like ChatGPT or DuckDuckGo display a perfect score on the same test.
Even more problematic, the new Siri sometimes invents totally fanciful responses. For example, the assistant attributed 33 Super Bowl victories to the Philadelphia Eagles, while the team has only one in its entire history. A particularly worrying tendency towards fabrication.
An incomprehensible regression
John Gruber claims that the old Siri, lacking Apple Intelligence, paradoxically performed better on specific questions. Faced with a complex question about local sports results, he simply admitted not knowing the answer and offered you relevant links.
The most surprising thing remains that ChatGPT, although integrated into Siri via Apple Intelligence, systematically provides the right answers when used directly. This degradation of performance casts doubt on the quality of the integration between the two systems.
These failures of Siri are also reminiscent of the setbacks ofanother feature of Apple Intelligence, which for its part is already controversial.
i-nfo.fr - Official iPhon.fr app
By : Keleops AG
