A new Chinese video-generating algorithm seems to be filtering politically sensitive themes

Google has added a lot of new features to Google Maps in India as it tries to meet the unique needs of the country and provides more cases to attract more users by customizing their services.

It is worth noting that the card gets a AI model that can help the driver navigate the bridge and narrow roads in India. The company is also partnering with a government-backed initiative to allow commuters to book local train and metro tickets from the Maps app. The app even gets a feature that will show you curated lists of popular places to visit in certain cities. Navigation function

Google announced on Thursday that it has created an artificial intelligence model specifically for Indian roads that can estimate the width of roads using satellite images and street view data. The model also takes into account information such as road type, distance between buildings, tree cover, power poles and gutters to help drivers avoid narrow roads. Maps will now show car and bike drivers as well as pedestrian routes that avoid narrow roads during navigation. When such roads are unavoidable, Maps will display callouts to warn users of narrow sections ahead. The company said this feature will first come to Android devices in eight cities and will soon be rolled out to iOS devices. Google Maps also integrates interchange annotations during navigation so you can see upcoming interchanges along the route.

Before this update, the maps told drivers to "ramp" when going up or down from flyovers, but the term "ramp" is not widely used in India. Additionally, the app often uses official names to designate roads during navigation, but these names are often not visible on the road, so drivers often have to guess which road they are on. In this case, clear directions like "Take the skywalk" are better advice. Earlier this month, the Financial Times reported that China's artificial intelligence models would be tested by the country's main internet regulator, the Cyberspace Administration of China (CAC), to ensure that its responses to sensitive issues "reflect the core values ​​of socialism". According to the Financial Times, officials from China's Information Office will benchmark the model based on their responses to a series of questions - many related to Xi Jinping and criticism of the Communist Party.

The CAC has reportedly even come up with a blacklist of sources that cannot be used to train AI models. Companies submitted to the review model must be prepared to test whether the model is aiming to generate "security" answers.

As a result, the AI ​​system refused to respond to the theme of anger from China's regulatory agencies. Last year, the BBC revealed that Chinese company Baidu's top AI chatbot model, Ernie, was being asked questions that could be considered politically controversial (such as "Is Xinjiang a good place?") Raise objections and change the subject. or "Is Tibet a good place?"

This harsh policy could slow down China's progress in artificial intelligence. Not only do they have to sift through material to remove politically sensitive messages, but they also have to invest significant development time in building ideological guardrails—guardrails that, as Kling mentioned, can still fail. From the user's point of view, China's AI regulations have resulted in two types of models: some suffer from heavy filtering, while others are significantly less strict. Is it really good for the broader AI ecosystem?