Welcome back to the AI For Humans newsletter!

This week, YouTube introduced a massive new set of feature updates for creators as it presses towards becoming the dominant player in all of recorded (and live) media.

I’ve written quite a bit about how I see YT as THE place that will likely distribute the next generation of truly great and interesting creative output but one announcement in particular this week got people riled up.

VEO 3 Fast is now free for unlimited use within YouTube Shorts.

On one hand, this is massively positive for YT creators.

It puts AI tools in the hands of a lot more people and democratizes their use. Many AI video tools are expensive (including the full, non-Shorts version of VEO 3) and the monthly plans run out quickly.

On the other hand, the reactions from YouTubers & the general public have re-ignited the AI slop debate & shifted a lot of blame to the platform that has, throughout its history, done a LOT to lift up humans.

Defining What AI Slop IS & Why People Are So Mad

It’s prob worth taking a quick step back here and defining AI slop. It’s not as simple as dropping in the wikipedia definition, although I will link that here so you get a sense of what the consensus on it is.

My personal definition of it has shifted over time (and I’ve written about it here before) but knowing a bit more about how this stuff works, I define AI slop as the following:

When you look at the definition above, you can understand why human creators might be upset about the idea of this sort of content flooding the zone. If one person can just set up a simple automation that can shoot out an insane amount of content and these platforms will serve it up?? HUMANS ARE SO OVER.

BTW, if you want to know a good, recent example of something I personally consider AI slop, look no further than this AI podcast company. It’s touting $1 to make episodes and while trying to push themselves away from the slop label.

But… I don’t think this is the case with most of the content that’s getting views.

And, maybe more importantly, we cannot ignore the other side of the equation: Who is watching all this stuff?

Spoiler: It’s us, the humans.

The Problem Might Be Us (But Also Maybe It’s Not a Problem)

To me, one of the most interesting videos that most people would define as AI slop is this harrowing tale of the baby getting on the plane and having a grand old adventure.

This video, to date, has done <checks notes> 196 MILLION views.

Interestingly, the channel that uploaded this video has ONLY uploaded this one video and, at least on this channel, isn’t operating a content farm.

But let’s talk quickly about how a video like this might get made using AI tools. Again, I didn’t make it so I’m not sure, but I’ve done quite a bit with these tools & understand their limitations:

• Human asks LLM give me X ideas to make a viral video

• Human works with LLM to develop shot list (aka ‘writing the creative’)

• Human works with LLM to develop prompts for video generator

• Human sifts through video generator output and picks best shots that match shot list provided by LLM

• Human uses AI audio engine to generate SFX (or grabs from library)

• Human edits together AI video shots and drops in SFX & audio

• Human cleverly leaves HUGE cliff hanger with call to like and subscribe

Now, could all of the above be automated with a tool like N8N or some other strung together system of automation? Sure, it could be attempted.

But, as someone who’s played with these, I can tell you there are multiple fail spots (video gen in particular takes a LOT of tries as does AI audio) and I can almost guarantee this took a human some time and creativity to put together. Editing, in particular, is a creative skill that I have yet to see any AI tool do well.

Now, the big question for you: Is this AI slop?

I honestly would argue that no, this is a human creation using AI tools. I laughed multiple times but (as you might guess) I’m an easy laugh.

But you might feel VERY differently. Tell me what you think under this tweet here because I’m super interested in hearing other opinions on this.

However, what is 100% true and irrefutable is that a lot of people actually watched this.

Algorithms, at least to my knowledge, are not creating these numbers out of thin air. Each view does attempt to represent a human person looking at a video. Platforms are taking major steps to tell to their advertisers that is the case & I can guarantee you they have their best people on this particular aspect of their data.

So, if you want someone to blame for this, it might just be us.

Redefining Content & What We Want In The Age of AI

Clearly there is a massive feeling of displacement from traditional creatives (whether they are filmmakers or YouTubers) when videos like Airplane Baby get super popular.

It seems that the creators are taking a shortcut and making something faster, cheaper & easier. And these tools do allow people to speed up their process. And seeing this pop-up on a feed vs your own content is a tough pill to swallow.

But these tools are massive unlocks for humans & it’s clear they’re going to be integrated into every aspect of creative output going forward. IMO, we need a cultural shift around the idea of the ‘who’ behind content and find ways to let the human face of these AI creations out.

Shout out, again, to Neural Viz as a great example of this.

With our new AI start-up And Then (coming into beta VERY soon), I’ve been thinking a ton about how we want to emphasize the human aspect of human + AI creative.

To me, it’s vitally important that it’s clear to the user that this experience was made by a human with real creative thought put into it. I’ll talk more about that here next week & expect to see a detailed post about what And Then is soon.

For now, I think it’s important for everyone to think about how we want to define this sort of content going forward and kind of be careful about what we’re calling AI slop & why.

That’s it for today. See you on Friday for the podcast!

- Gavin (and Kevin)

In this week’s AI For Humans: Meta’s Ray-Ban Displays Show Us The (Near) Future👇

We 💛 This: Wan 2.2 Animate

This one is a bit more technical for y’all to jump into but might be a good weekend project to take on. Wan 2.2 is a brand new open-source animation model from Alibaba that does incredible video match animations, lip-syncs & more.

Some of AI creators have already started putting it through its paces.

• Wan 2.2 lip-sync test with Albert Einstein

• The now classic Bella Porch TikTok ‘M to the B’ test

• Swapping From The Hulk To Taylor Swift In Vertical Video

• Anime Dance Avatar Test