YouTube’s AI-led Video Summary

Deep Fakes be damned! Artificial intelligence (AI) is now busy revolutionizing video editing by slicing-and-dicing long videos and repurposing them as shorter ones like Reels. At a time when the story of Opus Clip, a generative AI video editing tool is already doing the rounds, we now have YouTube kicking off its tryst with the technology by generating summaries. 

A notice on Google’s Support Page dated July 31, suggests that auto-generated summaries for YouTube videos will only appear in a chosen few English-language videos to start with. And they will only be visible to a small subset of users to give a brief overview of the contents in the video without actually replacing its existing description written by inferior mortals like us. 

A quick look at what Google is offering

Here’s what the page says, “​​We’re starting to test AI auto-generated summaries on YouTube, so that it’s easier for you to read a quick summary about a video and decide whether it’s the right fit for you.” Just in case, anyone wants to participate in the experiment, head over to YouTube.com/new and pass some tests. Of course, a paid subscription could be a must. 

A report published by The Verge quoted Google spokesperson Jen Jamie to suggest that the new feature was part of similar feature tests undertaken by the company to build or refine the best experience for viewers and creators. Viewers checking out English Vlogs, shopping and how-to videos on smartphones are the ones likely to see the auto-generated summaries. 

Opus Clip is fun, but are the results fun too?

One of the first responses we came across had a ring of truth. If a creator is writing a summary only to get the video indexed by Google, then AI can easily replicate it. However, what if the video creator wants to tug at our emotional strings or give information that’s beyond what’s on the video? Say the video is about a specific performance in a game of soccer, can the AI summary actually capture that? Most likely not! 

Which is also the question that many are asking about the video repurposing tool Opus Clip. Founded by Young Zhao and Grace Wang, the company uses a curation model powered by ChatGPT4 to analyze long-form footage and pick out nuggets from it and arrange them into a shareable viral short clip format. 

The company claims to be targeting content creators seeking to convert podcasts and webinars into clips that are easily shareable on YouTube, Instagram and WhatsApp. The service itself isn’t dissimilar to Adobe’s effort at scene detection via the AI-powered Sensei tool. However, Opus Clip seems to be targeting non-editors through its marketing.

Now comes the challenge. Anyone who sought to create videos would tell you that editing isn’t actually fun (unless you love it for that it is) as one needs to review, transcribe and then make the cuts for bite-sized versions. The challenge though is how accurate would these videos be? Well, we suggest you test it out yourself as we are doing. It costs $13 a month!! 

Looks like Google is turning to AI with a vengeance

Of course, Google’s penchant for giving out information on a need-to-know basis means we aren’t sure whether AI-led summaries would replace video descriptions as they are now and how a user will be able to differentiate it from one written by the creator. Of course, AI-led summarizer tools aren’t new – check out Clipnote.ai and Skipit.ai for more. 

For the moment, opinion is divided about YouTube’s summaries as well as others that claim to be finding a nuanced summary of what a video is about. A raging debate on Reddit presented two sides of the challenge. One claimed that for longer videos, the summaries were far from accurate while another suggested that for videos under a minute, they were good enough. 

A few Reddit users claimed that the summaries were as accurate as any other AI-led work that includes writing content or creating codes. “At best, these are about 30 to 40% on track and the rest requires human intervention,” says another user. As for YouTube, the latest test just adds to a bag of AI-led tricks such as quizzes for educational videos and a dubbing tool. 

This is one space we’d like to wait and watch. For, Google seems to be going overboard with its AI testing ever since Microsoft bet them to it with their OpenAI investments and incorporation of generative AI into their Office Suite.