M.G. Siegler •

Meta's Sloppy AI Social Network

The new 'Meta AI' is exactly what you'd expect, in ways good and bad...
Meta's Sloppy AI Social Network

In the end, Meta launched a social AI product before ChatGPT could become one. And it's... pretty much exactly what you'd expect it to be. A service that's both aiming to catch up with the competition while also leveraging Meta's unique strength in social media. It's both pretty well made but also a bit boring.

Does it have any sort of shot in displacing ChatGPT? I mean, as a pure-play AI tool, no. But it could blunt the potential social growth vector that OpenAI was apparently exploring – though it's also not clear if that would have been a meaningful lever for them. But if we learned one thing from everyone Studio Ghibling their families – it's maybe! At least for a time, thanks to viral moments. Then again, leveraging other, existing networks for such growth is seemingly a better path.

Regardless, I suspect Meta's maneuver here may actually be useful to OpenAI in one way: showing them what not to do...

Here's Alex Heath's overview of the new Meta AI for The Verge:

Meta’s standalone ChatGPT competitor is mostly what you’d expect from an AI assistant. You can type or talk with it, generate images, and get real-time web results.

The biggest new idea in the Meta AI app is its Discover feed, which adds an AI twist to social media. Here, you’ll see a feed of interactions with Meta AI that other people, including your friends on Instagram and Facebook, have opted to share on a prompt-by-prompt basis.

This "Discovery feed" is more or less exactly what I suspected OpenAI was going to do when the reports first surfaced that they were working on a social product. I guessed that it was less a stand-alone product and something more akin to the public feeds we've long seen in other AI products, primarily those built around image creation. As I wrote a couple weeks back:

To be honest, this sounds less like a Xitter competitor, and more like the public feeds of Midjourney and other AI image creation tools. These feeds are interesting and smart because they provide both inspiration for users of the services and also overcome the "blank prompt" problem – i.e. if you can literally create anything in the world, what should you enter in the prompt box?

While this is useful for images, it's arguably just as important for any AI query as these tools go more mainstream. Anecdotally, a lot of people I know – non-tech people – who have tried ChatGPT, aren't even sure what to ask at first. So a feed showcasing such things would potentially go a long way in user on-boarding, if nothing else.

So yeah, that's what I suspect OpenAI will launch, perhaps now imminently, given that this is what Meta just launched. But actually, in using it, and Meta being Meta, their version is more social than I suspect ChatGPT's would (and should) be. With Meta AI, you sign-in with one of your Meta accounts and this creates a public profile. With it, you can not only share your AI prompts and outputs, but also "like" those from other people. And comment. And even remix those prompts from others – this is seemingly the most important social signal in this early version of the product as your profile displays how many remixes you've spurred.

Beyond my vision for pure utility – again, overcoming the blank prompt problem – I'm not sure how interesting an actual social network for AI outputs is. My early instinct is that while it might be sort of fun for image remixing, regular prompts will get completely lost here – thus eroding the very utility I'm thinking about above. And even then, once the novelty wears off, will people keep sharing their images for others to see?

What if Meta brings over the Facebook and/or Instagram graphs? Again, maybe that's fun for a time, but I'm not sure how sticky that product is. And there are perhaps some very real downsides if people don't fully understand what they're sharing – as is clearly already happening! It's not just AI slop, it's AI sloppy.

So instead, maybe Meta intends for such content to be shared outwards, back to their more robust social networks? But there, beyond perhaps a few fun viral moments – those Studio Ghibli-style images – one suspects those networks will quickly start to complain about being inundated with such content. It's one thing to have a network built for that content – as this new Meta AI service clearly is – it's another to shove this content into an established network.

Then again, Meta loves doing that shit. No shame. Just game. They don't care if the existing user base throws up from this forced feed, if there's a chance to snuff out a rival, we will be guinea pigs getting A/B tested until the cows come home. And Meta has clearly long been laying the groundwork for such tests...

And if and when it turns out that you don't want your Instagram feed shoved chock full of AI-generated images from your friends, they'll try, for about the 5th time, to see if you might like your friends themselves to be AI-generated. This hasn't worked with celebrities. And it continues not to work with celebritieseven if they're sexting you, it seems. But maybe, just maybe, it's all a matter of timing and framing, the thinking goes. Meta has a lot of surfaces on which to attack their three-billion-plus users. Something might stick... Probably not. But maybe!

Anyway, this new Meta AI version is just step one – well, again, a 5th attempt at that same first step – it's a new approach from where to search for their AI path. Clearly, they needed a stand-alone answer as simply shoving their AI into the existing multi-billion-user surfaces may have provided them with some numbers to tout, but it wasn't giving them numbers they could actually be proud of – or ones that would be sustainable, long-term. See also: Google+ back in the day.

As I wrote a couple months back on the rumors of a Meta AI app:

For the record, I doubt they'll use the Llama branding, which is of course their "open-source" (read: open-weight) model name. I think it would be fun, with great app icon potential, but I suspect we'll get a bland old "Meta AI" app.

Still, branding aside, this makes sense to me. While Zuckerberg keeps touting the amazing usage Meta AI is seeing within their products, I'm not really buying it and I know I'm not alone there. Per above, it really feels like the way Google used to tout the massive Google+ usage – after they shoved it into every single product via notifications and other annoying surfaces. Meta replaced their own search functionality in apps for billions of users with Meta AI. Of course people are using it! The question is if they're getting any real utility out of it (or even know they're using it).

Further, the arms race is seemingly shifting from purely being about LLM pre-training to winning the hearts and minds of consumers. And, of course, their wallets. ChatGPT is far ahead in all of those areas right now. Meta is middling at best – despite having what many consider to be the best "open-source" (read: open-weight) models. Well, at least until DeepSeek entered the fray...

And:

Eventually, perhaps you bundle the Meta AI app back into the core products. And this is Meta, you can be sure they're going to unbundle and bundle again until the cash cows come home. But really, you probably just do both here. This is the same playbook that Google now seems to be running with Gemini. It's baked into their products – including, notably, Search – and exists as a stand-alone app.

This is the wild west right now, you need all the guns you can get your hands on. Having a shotgun is great, and may save you in a shootout, but it's overkill in the initial duel. You need a trusty, nimble pistol too. ChatGPT is Wild Bill Hickok in this scenario. And we're on about pace number six of the ten before we turn...

Meta, like Google and even Microsoft, clearly has ChatGPT envy. It's understandable! And so this new approach is also understandable. But that doesn't mean it will work. Fittingly, it may serve as a testbed for OpenAI to know what they should and should not do when it comes to "social" and ChatGPT.

As for Meta and AI, I think they probably need to do something far more interesting and inspiring. Something beyond incorporating user data from Facebook and Instagram. Sure, AI ad optimization may pay off, literally (as it needs to), but they're clearly aiming for something larger. Something outside of the prompt box.

Perhaps something literally outside...1

Meta Is Unbundling... Again
The strategy behind the stand-alone apps for Reels and Meta AI…
RIP Meta’s Celebrity AI Chatbots (2024 - 2024)
BRU, we hardly knew you…
An AI in Every Pot
Meta aims to scale AI to their masses…
Meta’s Next Big Blunder
Social AI “characters” are so obviously and hilariously going to fail
A ChatGPT Social Feed Makes Fun and Practical Sense
Nothing too crazy beyond a finger in the eye of both Meta and Xitter…

1 With that in mind, it's interesting that this new Meta AI app was born out of the companion app for their Meta Ray-Ban smart glasses...