ScaleAI quality concerns & Meta Superintelligence Labs (MSL) Falling Apart

Did Zuckerberg's dream team become a nightmare? Did he get scammed by Scala AI and dirty data? Have you seen this? Well, not really, but it sounds really exciting because it's Mark Zuckerberg getting scammed. So, it is exciting. So, you remember the dream team. How long's that been being put together? How long has Zuckerberg been dishing out hundreds of millions of dollars to pick up the likes of the key people from OpenAI, from Anthropic, from just the AI industry in general. How long? What? Two months. Yeah. And it's already falling apart. They uh purchased Scale AI or at least they purchased 49% stake in Scale AI, owning a minority stake in the whole entire company. But really, I think they were just after Andrew Wang to enroll him into Zuck's new efforts of the MSL, the meta super intelligence labs. What people are finding that are working with this in this MSL group is that the scale AI data is actually pretty terrible. They're actually turning to Scale AI's competitors to get better data using companies like Serge and Mer. I think you call that Merkore. So scale AI is is basically just I don't know maybe it's just not good enough and it's not necessarily the reason what they found was the data was produced by cheap labor and the the sort of quality they need at this point is from the likes of like doctors and lawy lawyers and scientists put together through this you know lowinccome um messy thing. So people are getting frustrated and people are actually leaving because of this. Now the senior well not nec again it's not because I think this is just piecing everything together here and you know one thing leads to another because we're only two months in and the senior vice president um has left because of it and other other members are are leaving. How does this work? To my like the way I understand there are these companies that are specialized in cultivating and collecting data actually they what what do they do? Do they go to like do they buy books, scan them and put them like in Dropbox and something like that? How does that do you know what I couldn't even tell you but we can we can hazard a guess and my guess would be yes. They're scanning books. They're doing research. They're collecting data from various resources. They probably ask them to gather information essentially, right? Categorize thing. I I believe there are there is also some tagging going on. So they tag uh information like images etc. It's not just like uploading things. They have to understand like they use the human to understand what the context like what is the context of this paper or whatever and add some notation that that's I believe what is happening using combination of automated tools and distributed network of human annotators in a process known as human in the loop data management. The company partners with clients to gather raw data in various types such as images, text, video, audio. Sounds like a lot of manual labor to do. So, this is just one of the pieces of this puzzle really when it comes to this this whole falling apart of the MSL group and Meta's efforts to to reach super intelligence. Is it to do with the the fact that they've lost faith in the the person who's leading their team? His own company is just a shell of a company. It's not very good. just not as good as they they maybe maybe Andrew Wang sold the dream and Zuckerberg took a sip of the scale AI Kool-Aid. I don't know. These are American idioms that I don't truly know. Comr is commenting saying that they use label data and then supervised learning uh which is used for training the model. They also use unsupervised learning technique and semi-supervised learning. Okay. So, there are like a bunch of different techniques. Nobody feels bad for Mark. So, yeah, he's tried to throw money at something and it's uh it's just not panning out for him. And to be honest, you know, we could probably go into the next article. Essentially, Meta are now looking to chat GPT, to open AAI, to Google to actually um partner with them in order to support their own meta super intelligence labs. It begs the question about OpenAI uh sorry, Llama being open source and if they're behind the scenes using Google and using, you know, um ChachiPT and stuff like that, is it truly open source? I think ultimately it's basically encouraging their teams to use third parties as much as possible to to work and um we haven't spoken about this but I'm close with some of the guys at Meta and I was chatting to him a couple of weeks ago and they full they're fully in on claw code they use clawed code on a daily basis to do their work at Meta now they're not part of MSL not that close and are they allowed totally allowed this is paid for this is this is their this is a subscription that's given to them. They're not they're not paying for them themselves. So, it just goes to show that Llama and Meta's own AI systems just aren't there yet. I know we kind of knew yet this, but it's just coming to light how yeah, how not up to scratch it is. So, um Zuckerberg's bet is is truly falling away. Meta also signed a six-year cloud computing deal with Google worth over 10 billion. and that's using their servers and stuff like that. And this is not, you know, I don't want to I don't want to try and um make something that's not already being done. iPhones use Samsung screens, you know, arguably better screens than Samsung songs only you their their phones use, you know, their own phones use. So, you know, this isn't crazy to know that you use other people. That's an interesting point and uh comparison that yeah like Samsung is making the the screens for iPhones, Sony is making the the camera sensors for iPhones. So based on this we so maybe the the companies who make the models might not be the winners but the companies that put together whatever model to to create a product they will be the winners. And we see this already. We see that these companies are you know starting to work with one but then switch to another company um based on like the quality and which changes like it feels like every week. Yeah, that's meta falling to absolute pieces in a nutshell. This is part of a larger conversation on my show Command AI which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.