Pages

DeepMind Collaborates With YouTube On AI Research Models To Produce Auto-Generated Chapters For Videos

The partnership between YouTube and DeepMind goes as far back as 2018. And the duo has been busy making some great progress. Now, we’re hearing about a new partnership that is related to AI research models.

Yes, both tech giants have united to produce auto-generated chapters that are being seen across countless YouTube videos. But they’re not stopping yet.

As of now, 8 million videos have auto chapters but they hope the figure will increase to nearly 80 million in the next year.

For the past 3 years, YouTube and DeepMind have been busy on label quality models. These are better seen as an amalgamation of what happens when a video mixes within advertising guidelines so that ads can be displayed.

The company says they’ve been responsible for launching a huge portion of the app’s life traffic. And during that time, it wouldn’t be wrong to say that they’re an integral partner that can use AI technology to compress videos using opening codecs.

DeepMind says it can carry out a range of functions without compromising on quality like bitrate reduction at the expense of absolutely nothing and that includes no quality reduction.

But the recent initiative of the two partners linking together for AutoChapters shows that they’re not just done yet and this partnership has miles to go before we can think of bidding farewell.

The company also outlined how they were able to go about their business and do just that. For starters, they made use of transformer models that produced huge chapter segments. Shortly after that, we saw the arrival of timestamps on the video through a typical two-step system.

After that, AI technology introduced a multi-model system that could carry out a variety of functions at once. This includes the processing of words, images, audio, and more, not to mention the production of different chapter titles.

We’ve seen Deepmind also hand a number of other big YouTube projects like recommendations on its app store, Google Maps’ predictions, and also the cooling system embedded in data centers.


Read next: Google Is Not Happy With The Practice Of Paying Publishers For Content Appearing In Search As It Undermines Trust

No comments: