‘It’s a whole new kind of blerp’: YouTube’s AI-enhanced reply suggestions seem to be working as well as you might expect

Have you heard of the dead internet theory? It’s essentially the idea that the internet is primarily populated with bots and AI-generated responses, intentionally designed with the overall goal of minimising human interaction and promoting products. The theory posits that humans on the internet are relatively rare, but bots responding in a manner just like them are everywhere. Including YouTube.

An interesting concept, no doubt. However, there’s one fly in the ointment: Even huge titans like Google struggle to produce AI-generated responses that make any sense. 404 media has been speaking to Clint Basinger, a YouTuber who’s been testing out “editable AI-enhanced reply suggestions” on the platform—and according to Basinger, the results so far have been variable at best.

The suggestions work in roughly the same way Gmail creates optional “smart reply” suggestions to your emails, except more in-depth. The AI responses appear to be based on actual comments from the creator in an attempt to mimic their tone, which on paper sounds like a perfectly reasonable idea. Context is everything, however, and without sufficient contextual data, the AI seems prone to making things up.

In a video demonstrating a Duke Nukem branded energy drink powder (god help us all), Basinger struggled to find a scoop inside the packaging. A commenter suggested that the scoop might be buried in the powder, to which the AI suggested the response:

“It’s not lost, they just haven’t released the scoop yet. It’s coming soon.”

Later on in the video, Basinger shakes the container, and a commenter suggests that he should have had a tighter grip on the lid. According to the YouTuber, the AI suggested they respond with: “I’ve got a whole video on lid safety coming soon, so you don’t have to worry!”

Mind you, I suppose you could argue it’s providing ideas for new content, at the very least. Lid safety videos are the hot new thing, after all, and… I kid, I kid.

It’s not just context the AI seems to struggle with. Occasionally, it’s repeating words. Basinger’s other channel is called LGR Blerbs, and upon viewing a new video, another commenter said “Nice. Back to the blerbs.”

“It’s a whole new kind of blerp,” suggested the AI.

Yep, a completely made-up word there. Anyway, Basinger seems to take all this in good humour, although they report one incident where things take a slightly darker turn. When another commenter expressed delight that Basinger was posting to their second channel, the AI suggested they respond:

“Yeah, I’m a little burnt out on the super-high-tech stuff so it was refreshing to work on something a little simpler.”

While YouTube does provide advice for its creators in regards to burnout, I doubt it’d want its AI to reference the fact that creators working on the platform can experience high degrees of stress and strain keeping up with the churn.

So, as for that dead internet theory? It’d be particularly obvious with this bot I think, although that’s not to say more advanced ones (or perhaps even worse ones, given some of the horrors seen in the YouTube comment section) aren’t out there, auto-generating away.

And as a useful tool for creators? I can’t quite see the idea of AI-generated responses catching on. After all, aren’t you, the audience, there to listen to what your creator of choice has to say, not what the AI thinks they might say?

And goodness knows what the auto-comments would look like with a more controversial training subject. I shudder at the thought.

Still, I digress. Time to cool my circuits, and… oh no, dear reader. The game is up. You’re really here though, aren’t you? Right? Please tell me I’ve been writing this article for someone, at the very least.