311–320 of 930 posts
Do you all realize you're arguing about nothing?
Good for him for doing something he seems excited about. Maybe we should all stop gossiping and go do something we're excited about too.
Who will fund the necessary computing resources? If not FB, then he will surely be joining or starting a different org
Robotics has the same issues, but you spend all your time fussing with the mechanical machinery. Carmack is a game developer; he can easily connect whatever he's doing to some kind of game engine.
(Back in the 1990s, I was headed in that direction, got stuck because physics engines were no good, made some progress on physics engines, and sold off that technology. Never got back to the AI part. I'd been headed in a direction we now think is a dead end, anyway. I was trying to use adaptive model-based control as a form of machine learning. You observe a black box's inputs and outputs and try to predict the black box. The internal model has delays, multipliers, integrators, and such. All of these have tuning parameters. You try to guess at the internal model, tune it, see what it gets wrong, try some permutations of the model, keep the winners, dump the losers, repeat. It turns out that the road to machine learning is a huge number of dumb nodes, not a small number of complicated ones. Oh well.)
What a joke. Carmack is going to sit at home and solve what teams of scientists can't do in decades. I'm complaining less about Carmack wanting to spend his time doing this and more about the comments here acting like he is some 10000x research scientist.
He invented modern graphics as a practical problem by himself as the sole researcher. Given the tools at the time that may have been a harder problem.
That is a ridiculous exaggeration. Carmack was clever enough to gain ~1 year advantage in performance over his competitors for the Doom engine, using Binary Space Partitioning, which was first applied to 3D graphics in 1969, before he was born. The Quake engine got a significant performance boost from Michael Abrash, who is a specialist in code optimization.
AGI is very much a research problem. It's not going to be solved with a clever hack.
Earlier quoted context omitted.
He had a lot of help behind the scenes and has been credited with things that aren't his. I respect his achievements more as a regular smart guy than a bonafide genius. He described the math in rocketry as being basically solved in the 60s and video games being far more complex as a project, so that was really a step down in difficulty. His VR role is the same field as his primary skills, impressive work but not an e…
Millions*- A cursory Google search suggests that he has a net worth of 50MM.
This doesn't surprise me at all. He went on a week long cabin-in-the-middle-of-nowhere trip about a year ago to dive in to AI (that's all this guy needs to become pretty damn proficient). (edit: I'm not claiming he's a field expert in a week guys, just that he can probably learn the basics pretty fast, especially given ML tech shares many base maths with graphics) As recent as his last Oculus Connect keynote, he exto…
This may be his biggest impediment. ML has gotten very far with looking at problems as linear algebraic systems, where optimizing a loss function mathematically yields a good solution to a precisely defined (and well circumscribed) classification or regression problem. These techniques are very seductive and very powerful, but the problems they solve have almost nothing in common with AGI.
Put another way, Machine Learning as a field diverged from human learning (and cognitive science) decades ago, and the two are virtually unrecognizable to each other now. Human learning is the best example of AGI we have, and using ML tech as a way to get there may be a seductive dead end.
I'll take an opportunity to plug a paper I recently published on comparing relative intelligence. The punchline will illuminate the low-hangingness of the fruit in this field.
Suppose X and Y are AGIs and you want to know which is more intelligent. For any interactive reward-giving environment E, you could place X into E and see how much reward X gets; likewise for Y. If X gets more reward, you can consider that as evidence of X being more intelligent. But there are many environments, and X might do better in some, Y in others. How can you combine those pieces of evidence into a final judgment?
The epiphany I had (obvious in hindsight) is that the above situation is actually an election in disguise. The voters are interactive reward-giving environments, voting (via their rewards) in an intelligence contest between different AGIs. This allows us to import centuries of research on voting and elections! In particular, by using theorems about elections published in the 1970s, I was able to provide an elegant notion of relative intelligence.
The notion I provided is elegant enough that some theorems can even be proved with it, for example, formalizations of the idea that "higher-intelligence team-members make higher-intelligence teams". Which emphasizes the low-fruit-hanginess of the field: as obvious as that idea seems, apparently no-one was able to prove it with previous formal intelligence measures, probably because those previous intelligence measures were too complicated to reason about!
Here's the paper: https://philpapers.org/archive/ALEIVU.pdf
Earlier quoted context omitted.
A general AI would give you all 3
That argument is like saying if we had better theorists you wouldn't need to build particle accelerators. Nature has brute facts that can only be discovered through observation. I would be surprised to see evidence that an intelligence, no matter how smart, could reason to the fundamental properties of neutrinos without massive physical real world experiments and piles of observational data.