Safe Superintelligence Inc. Vouches for Alignment
VC’s are still betting on AI, not shying away from pouring cash like it’s a poker game in Vegas. No doubt, with the recent tech layoffs (and some uncertainty in the consumer market) who else will foot the massive compute bills AI researchers are generating (unless there’s an absurd record digital currency mining operation laying low somewhere)?
Lucky for them, these VC’s have Ilya Sutskever (and a few other AI mavericks like Daniel Gross & Daniel Levy). Sutskever is a former OpenAI executive who is pretty much writing his own Wonka tickets (more like…billion dollar tickets). Safe Superintelligence (SSI) is the new maverick AI startup that pushes the limits of the status quo. At the time of writing this article (September 2024), SSI’s URL (https://ssi.inc) points to a landing page with no product, just basically a brief explanation of an ambition to pave the way for the next big AI step (with an air of JFK’s famous Rice University speech).
With less than 300 words written on their static page, the second biggest paragraph outlines a focus on: “…no distraction by management overhead or product cycles…”, basically flashing a pride of no production management pressure. Sounds like the perfect no lab coat wearing (nonconformist) Silicon Valley company to me!
Now here’s the trick question; how in the world do you explain a "no product deployment pressure" vibe to a VC? Unless Andreessen Horowitz & the rest of these "wizard sponsors" bought in the "non-profit" in disguise element when it comes to writing PR’s (remember how OpenAI started…and where they are now.)
Let’s not forget who Ilya is though - the guy studied at the University of Toronto & was mentored by Geoffrey Hinton (arguably the Yoda of neural networks himself).
After co-authoring the game-changing Imagenet classification paper, Ilya got hands-on experience at the top AI lab in the world at the time - Google Brain, before Elon Musk convinced him to switch teams. So he’s basically on the same AI level as a Samurai from the Shogun, but instead of a sword he’s armed with PyTorch (and a Zelda attitude). Jokes aside, all AI labs deserve their own recognition, whether it’s Meta, Google, Anthropic etc., but here’s the catch-22; none of them seem to have agreed on the true definition of AGI, as if they’re at a party playing the piñata together (each wearing VR headsets running on different OS’).
A few billion dollars for a startup is a lot of cash to imagine, but with how fast LLM’s are evolving - that barely gets you in the high stakes room. So SSI’s laidback, flip-flop wearing hippie “no-distractions” atmosphere description might look like a real chill place to work for the most recent grad (however I think it won’t take long before we start seeing some really confused poker faces in the media).
Their page does mention that they are looking for talent, but these days, the next breakthroughs almost always happen in someone’s garage, or some gloomy college lab where a kid is crunching numbers while eating instant noodles at two in the morning. At least that’s the impression of how the 2017 paper “Attention is All You Need” on Transformers (which pretty much is the magic wand of GPT’s) was written.
So at the end of the day - SSI may be a fearless move, and the founders are clearly aiming high. Forget AGI - let’s just take it above & beyond and call it...: Super. We’ll see if Nick Bostrom buys in.