Apple’s intelligence sparks discussion: AI makes big mistakes in summarizing news

While Chinese stock investors are hotly discussing Apple’s negotiations with certain technology companies on AI model cooperation, overseas investors are discussing another topic: the AI ​​of the world’s most valuable listed company seems to be unreliable in key areas.

As the hottest case in recent days, the BBC has published reports in the past week, accusing Apple of misattributing the media’s news push to its “Apple Smart” app, which was launched in the UK last week.

The screenshot shows that Apple’s AI push summary mistakenly stated that Luigi Mancione, who recently attracted global attention for shooting the CEO of UnitedHealthcare, “shot himself” – in fact, this did not happen. What is unbearable for media organizations is that when Apple AI pushes the wrong summary, it looks like the information pushed by the BBC news terminal.

In fact, this is not an isolated case. On social media, Apple AI summarized the news push of the New York Times and came up with a false result that “Netanyahu was arrested.”

A BBC spokesman said it had raised its concerns with Apple but had not received a response for a full week.

Similar reports are emerging one after another

At the center of this incident is the “Apple Intelligence” notification summary feature. One potential possibility that caused these problems may be the model Apple uses to summarize content – the “small” model with 3 billion parameters installed on the device is not capable enough – and this is an Apple-developed model, so it can’t blame anyone else.

Research published on Apple’s official website also shows that the satisfaction with the summary ability of the end-side model is roughly around 70%. Although the results are slightly better than the mini models of competing products, they are still far from “no major errors.”

In fact, there is no shortage of jokes on social media about “Apple intelligence,” such as Apple summarizing notifications from smart home cameras and notifying users at 1 a.m. that there are a lot of people at your front door, back door, and driveway.

For now, the notification summary of “Apple Smart” is still a tool with both advantages and disadvantages, and can occasionally create some laughs. However, in the field of current affairs news, errors in key information may lead to serious cognitive biases (imagine if you place an order and find that the news summarized by the phone is wrong).

Artificial intelligence is currently still a software that “generates correct content with a certain probability”, but whether information can be accurately disseminated should not be determined by probability.