Their last success was acquiring Instagram in 2012. Every new effort since then has been hemorrhaging money. They get away with it because they have two limitless money faucets in Facebook and Instagram, but their product strategy as a whole has been a disaster.
It would be incredible to think that Mark Zuckerberg genuinely thought their Metaverse/VR investment was going to be akin to Xerox's bayarea PARC campus (developer of modern networking / GUI &c). I guess both were ultimately profit-negative financial disasters.
Watching their demo video was the perfect encapsulation of "this was not made for users" I have ever seen. First of all the idea of hanging out in a digital world with Mark Zuckerberg is so bleak. I can't imagine a worse hang.
But other than that, it was all about working in a digital office, being advertised to, etc. They had this scene where one of Zuck's definitely-real friends is excited about "this new street art" on the digital wall that jumps off the wall and they interact with it. Imagine having popup ads that jump up at you when you're walking (gliding?) down the street!
I wish creators of local model inference tools (LM Studio, Ollama etc.) would release these numbers publicly, because you can be sure they are sitting on a large dataset of real-world performance.
Desktop computers being as open as they are is an anomaly. It only came to be because the systems originated from research labs and hacker cultures rather than rent-seeking corporations. And even though corporations (like IBM and Microsoft) did push them, there was a lot more emphasis on business rather than consumer use at the time.
Vendors keep them open today only because there is a historical exception, but make no mistake if the laptop computer was first introduced to the masses in 2008 you would be downloading apps through official stores and paying a 30% fee on all transactions and would only be able to do a tiny fraction of what is possible on them today.
To me the surprise isn't that the phone is locked down, but that Apple allows MacBook Neo to do so much. Just look at its iPad counterpart.
Everyone is hypothesizing government backdoors and whatever else but to me there's a simpler and more obvious reason - AI.
Companies started pushing E2EE a few years ago because users' private messaging data used to be a liability. Now that the data can be fed into LLMs for training and inference its value has gone up significantly, and the privacy and security tradeoffs are suddenly worthwhile.
PMs across the industry are pushing product decks with "conversational AI assistants" to get their next promotion. I've been in more than one of these meetings myself. If the data is encrypted then there's no way to build this kind of stuff.
It's around the same time they announced their Applied AI org under Boz, which is responsible for data for Avocado/Mango/Watermelon training now. The timing certainly doesn't help.
There is a product reason - AI features are fundamentally incompatible with E2EE. If they want to bring more AI generated experiences and content into Instagram then the data needs to be accessible by them.
Is that really a problem? All the examples on the repo page themselves show LLMs running unintended operations on the "correct" service and messing up your data. And that is very much still going to happen with this wrapper. If anything it is going to provide a false sense of security.
reply