"Meta is reportedly investing nearly $15 billion in the data-labeling firm Scale AI and taking a 49% stake in the startup, while also bringing on CEO Alexandr Wang to help lead a new “superintelligence” lab within the company."
More eyebrow raising
"Zuckerberg's hands-on involvement is unprecedented. He's reorganized Meta's Menlo Park office so the new team sits near him and is offering compensation packages between seven and nine figures (whoa) to poach talent from OpenAI and Google."
(!)
And people wonder why I am depressed! I should have done that MSc in AI in 2004-5 rather than the MBA. The problem is that I couldn't have got a loan to do an AI MSc whereas I could an MBA. Plus , in 2005, there weren't many actual AI jobs around and the chances of getting one even with a MSc in AI would have been pretty slim. So, having an MSc in AI would benefit me (a bit), but not really then. Although it would have been fun to do. I have done lots of Coursera courses and some Udacity stuff, but I don't know far it gets you. I have the NVIDIA Jetson Orin Nano now, so I can do some real stuff that I post to GitHub. I should have done that months ago. The DGX Spark will have 15x the grunt.
But who's getting $100M/year? These aren't tech billionaires as such. Get a first in CompSci at Cambridge (2012), do a PhD at the UCL Gatsby Unit, join DeepMind (2016), transfer to Google Brain on an L-1B (2018), get an O-1 visa, join OpenAI (2020), spent the last 5 years doing cutting edge research. That would make you about 36 now (assuming you had a year off to intern at Microsoft Research in Cambridge). I'm 57 and I was never that good at maths.
So, you've probably been leading a team at OpenAI (or equivalent company) for half a decade or so. You are probably as much a manager as a researcher. Your presence presumably helps to recruit and retain others, most of whom are probably going to be 10 years younger and on seven figure salaries. Still, this is utterly remarkable in its way. The economic justification is that you are only a handful of $100M salaries to the team leaders, most people will be on closer to $1M with a tranche at $10M or so. But that's not chump change. Even if you are paying the thick end of a billion dollars a year in salaries (and x3 that for the compute), it's still worth it you can get to ASI in 2 or 3 years. In terms of proto-AGI, there's clearly a lot of scope for different models to catch up with what other models already do better. ASI, in principle, could eat everyone's lunch, the old AGI/SL4 list idea of a singleton or monolith, that you can only have ultraintelligence because ve will simply out-compete any rivals. I don't know about that, but then I guess you drone strike the opposition's air-gapped data centre or penetrate using more traditional humint methods so that Meta pwns OpenAI and the US pwns China (or vice versa) and Zuckerburg ends up caesar at least until he pwned by his own ASI.
But is a $100M enough? You can't take it with you. In the long run we're all dead, but that long run might be quite a short one right now. The default assumption is that 2035 will look like 2025 only more so. But 2025 looks nothing like 2015 in a lot of ways (Brexit, Tr*mp I, electric cars, COVID, mRNA, Ukraine, proto-A(G)I, Gaza, Tr*mp II, the end of the post-WWII/post-Cold War dispensation, the Return of History), so I am not sure at all that that works anymore. If you really wanted to survive until 2035, you would probably want to make some pretty extreme precautions and how much are they going to cost and how much are they going to have to pay you to live in a nuclear submarine for the next few years?
But, of course, this was what I wanted. We are getting it, it seems. And there is nothing I can do, but be a bystander as we hurtle into history. And there's probably nothing I could ever have done. Even if I had done a PhD in AI in the late 1980s/early 1990s, I just wasn't good enough at maths and research to make it as an academic researcher. I wasn't going to be Murray Shanahan even if I could have written his 2015 book. There's the whole futurologist angle, which is what I said I wanted for so many years and this blog is, I suppose, an actual suppose, an actual instantiation, over 21 years, of that ambition. It's not nothing, but I wonder if there ever were a time that I could have leveraged it into something more. It seems hard now to imagine that it's not later than we think, but perhaps 90 will be the New 50.
Add comment
Comments