Meta’s Favorite Product Isn’t AI. It’s the Copy Button.

An hour after I posted my piece about Meta’s Super Intelligence memo, a friend pinged me and pointed out that once again, Mark Zuckerberg has subsumed someone’s idea and phrase and presented it as his own: personal super intelligence. Not surprising, given that “copy” and “subsume” are on-brand for Zuck, who historically has not been shy about taking ideas and presenting them as his own.

The phrase “personal super intelligence” was coined by Character.AI CEO and co-founder Noam Shazeer, who was also one of the authors of the seminal paper, “Attention Is All You Need.” This paper is widely regarded as the work that inspired today’s AI chatbots. When the company announced its Series A funding, in their press release, they noted:

We understand the importance of providing an AI that truly feels like your own, and that’s why our AI is customizable. From personality to values, you can customize your AI to suit your needs and preferences. Whether you want a sympathetic ear or an analytical problem-solver, there’s an AI at Character.AI that’s right for you.

There is an eerie similarity in how Zuck presented his version of “personal super intelligence.”

As profound as the abundance produced by AI may one day be, an even more meaningful impact on our lives will likely come from everyone having a personal superintelligence that helps you achieve your goals, create what you want to see in the world, experience any adventure, be a better friend to those you care about, and grow to become the person you aspire to be.

Meta’s vision is to bring personal superintelligence to everyone. We believe in putting this power in people’s hands to direct it towards what they value in their own lives. Personal superintelligence that knows us deeply, understands our goals, and can help us achieve them will be by far the most useful.

As I was saying, Zuck knows how to copy and subsume well. He has been doing it since the beginning of Facebook. And that is why I think it is in the company’s DNA.

To understand this pattern, look no further than Meta’s secret weapon: a small startup it acquired called Onavo. This app essentially informed Facebook about which apps consumers had installed, how frequently they opened those apps, and how long they stayed inside them — functioning as “an early-detection system on whether an app was gaining popularity.”

Now, to replicate the Onavo strategy for the AI era, the company spent billions to invest in data-labeling startup Scale.AI and acquire a 49 percent stake in the company. Scale was labeling data for pretty much all large AI platforms. This is a small price to pay. Why?

• Onavo: Showed which apps were popular and how much they were used.
• Scale AI: Shows what capabilities competitors are building, how well their models perform, and what data they need.

Since Scale AI was helping the likes of OpenAI, Google, and Microsoft prepare data they use to train cutting-edge AI models, Meta can now understand what models are struggling with what tasks, or what new capabilities they are developing. In general, it can understand where the industry is heading. No surprise, the likes of xAI, OpenAI, and Google have said sayonara to the data labeler. Of course, the upshot of this is that now its rivals have to scramble to find a new source of “labeling,” giving the company more time.

Let’s start with the disputed origins of Facebook. Cameron and Tyler Winklevoss, along with Divya Narendra, claimed that Zuckerberg stole their idea for a social networking site called ConnectU (originally HarvardConnection) while he was hired to work as a programmer for their project in 2003. There is even a movie about this — The Social Network. The Winklevoss brothers, not themselves paragons of virtue, got paid and eventually became billionaires.

Over its history, the company has done this repeatedly. Even the two things we most associate with Facebook, the like button and the newsfeed, were not created by the company. The newsfeed was a consequence of the acquisition of Bret Taylor’s FriendFeed. The like button can trace its roots to Digg and Vimeo.

The company that has suffered the most as a result is Snap. In 2012, Facebook launched Poke, a Snapchat copycat. It allowed users to send photos, messages, and videos that disappeared in ten seconds or less. It failed. Again forgotten in time is the fact that Facebook launched “On This Day” in March 2015 after seeing TimeHop hit 6 million daily users and having a moment of panic. 

Fast forward to 2016, and the company launched Instagram Stories, nearly identical to Snapchat Stories. The rip-off was complete feature-by-feature. Facebook did a good job of cloning Snap’s stories, and rolled it out across all its products. They essentially copied Snap, feature-by-feature — from the disappearing Stories format to camera interfaces to face filters and lenses. The most important thing it copied — the advertising formats within Stories. These days everyone is talking about Meta’s AR glasses. Again, an idea blatantly ripped off from Snap, which was the first company to really experiment and launch AR Glasses.

We have seen it try to copy TikTok with Instagram Reels, launched back in 2020. It did a good job of copying the interface. However, it has failed to even come anywhere close to TikTok’s secret sauce — its scary algorithm. Reverse engineering an algorithm is not that easy.

I bring up this recent history because it shows a pattern. Whenever there is a new competitor with a new online and social behavior, Meta management views it as a systemic risk to its attention-based economy. Instead of trying to reinvent, it simply copies and rolls out those features to its massive audience. Thanks to its large audience scale and massive infrastructure, it leaves the smaller rivals and original innovators as nothing more than mere footnotes.

Facebook’s real innovation has been in perfecting and scaling existing concepts rather than creating entirely new behaviors. Their strategy appears to be: watch others innovate, then copy and execute better with their massive user base and resources. Out in Silicon Valley, we call this “Zucking.”

As I pointed out yesterday, the biggest challenge for Meta is not that it doesn’t have the resources or the mettle. It is just that it still doesn’t know what to copy. Scale.AI might help, but the reality is that no one, including OpenAI and Anthropic, knows the how and why of AI and human behaviors around it. Chatbots are a start.

Taking the idea of “personal super intelligence” from Shazeer might have been the easy part. Actually “zucking” it would be much harder.

Popcorn, anyone?

July 31. 2025. San Francisco

Decoding Zuck’s Superintelligence Memo