John Carmack Stepping Down As CTO of Oculus To Work On AI (theverge.com) 41
Oculus CTO John Carmack announced Wednesday that he is stepping down from the augmented-reality company to focus his time on artificial general intelligence. The Verge reports: Carmack will remain in a "consulting CTO" position at Oculus, where he will "still have a voice" in the development work at the company, he wrote. Recent comments from Carmack suggest he may have soured on VR. Carmack was a champion of phone-based VR for years at Oculus, but in October, he delivered a "eulogy" for Oculus' phone-based Gear VR. And in a video for receiving a lifetime achievement award this week at the VR Awards, he said that "I really haven't been satisfied with the pace of progress that we've been making" in VR.
Skynet now has it gun shooting code! (Score:4, Funny)
Skynet now has it gun shooting code!
id software's Skynet (Score:2)
Skynet: Prepare to meet your doom ! <evil laught> ... fuck.
Human: - iddqd. Sorry, what where you saying again, Skynet ?
Skynet:
finally, better Q3A bots (Score:3)
they were tough, but not that tough ..
All the fun code in VR is probably over... (Score:2)
Time to move to the place where there's fun code to be written.
Re: (Score:2)
Abrash worked on Quake, and he was even picked up for the Oculus.
Doom (Score:2)
Really enjoyed reading about Carmack in 'Masters of Doom'. AI could use some fresh insights and I'm sure his nose to the grindstone persona will benefit the field.
Re: (Score:3)
idk man. the tricks carmack pioneered were elegant af and absolutely necessary at the time when you had to wring everything out of every clock cycle, but in absolute terms they only increased performance by a few factors. what's the point in AI, where you can wait a year or two and get a 2x improvement from hardware alone.
and beyond that, he wants to do general AI. good fucking luck; all the noses to all the grindstones in the world won't bring that. it needs a fundamental breakthrough or a hundred.
i wish h
Re: (Score:2)
that "improvement from hardware alone" point was a bit weak; after all, performance back then increased 2x every three years or so. not much worse.
let me bolster it. to be clear, state-of-the-art ML algorithms (particularly stochastic gradient descent for neural nets) have already been parallelized not perfectly, but at least to the point where, if the bottleneck were compute power, DeepMind, Google, or even a few of the larger universities could easily just brute-force it (as they are already doing, and ha
Re: (Score:2)
Carmack's greatest skill was his ability to distill useful information that wasn't particularly obvious from the literal wording of the relevant datasheets & scraps of documentation. Most of what he (and Michael Abrash) did seems obvious NOW, in RETROSPECT, but at the time (especially if you relied only upon IBM's official public documentation), it really wasn't.
Carmack was also willing to take risks and be a rebel, and the fact that he released his first games as shareware gave him freedom that he woul
Re: (Score:2)
It was different times. I think you are absolutely correct about the shareware angle. All the big titles of the era sorta implemented try before you buy that way. In those days of 80386s and 80486 systems clocked at 16-66MHz with who knows what array of 30pin SIMM modules installed what their performance was like and any number of SVGA cards with varying degrees of VESA compliance you weren't surprised if a given title didn't run or did not run well. You did not really blame the publisher either you just
Re: (Score:1)
Have you played many "AAA" games lately? The Battlefield, Fallout, and Elder Scrolls series are notoriously unstable and barely playable for a week or two after launch-- unless you're on AMD hardware, in which case it might be a month before the game runs proper
Re:Doom (Score:4, Insightful)
Carmack is notable because he was first. He is an implementation guru, and was able to achieve things on hardware that were *just* barely possible with that generation of processing speed / memory / storage, etc. I've spent hundreds of hours in the Quake (and Quake 2) sources, porting it back in the early 2000s to Pocket PC [slashdot.org] which was a feat of its own, again, to get the game to run on hardware on which it was *just* possible at that point in time (to be very specific, the hurdle for PDAs was that they did not have a math coprocessor, and could not do floating point math at the hardware level, so I spent many, many hours converting floating point data and operations to fixed point math).
Carmack took constructs that had already been "invented", sometimes decades earlier, like with the BSP tree, and figured out the magical blend of real-time processing coupled with pre-processed data, to get a desired feature set running on hardware in which it was just possible. He would experiment often and would totally discard and re-implement code that may have taken dozens of hours to write with an entirely new methodology when he encountered some performance bottleneck or feature limitation that was deemed unacceptable. Where most developers have sacred cows in their code, or code that is known to work and be bug-free, that often prevents them from making drastic changes, Carmack would literally just toss entire subsystems worth of code and start over, even as deadlines neared.
One of the greatest things he achieved with Quake in particular was the preprocessing of data by their custom toolchains, to produce data that could be consumed and rendered real-time on contemporary devices. Examples are the algorithms used to generate balanced BSP trees, precalculating which nodes of the map were visible from which other nodes in the bsp (the "vis" tool), and pregenerating lighting shadows.
In some ways, like with Bill Gates, he was at the right place at the right time, to capitalize on his skills to get a step ahead of everyone else and do something groundbreaking. How well his strengths and the kinds of processes he embraces will result in success in other endeavors isn't clear. I'm not exactly keeping score, but he hasn't exactly turned into an Elon Musk, for example, because Carmack is an implementer and not necessarily a leader.
Re: (Score:2)
Re: (Score:2)
He ain't hebbie.
Re: (Score:2)
I wonder if, with Carmack leaving, Abrash will also leave. Not that he will necessarily follow, but Abrash doesnt stick around once the interesting stuff is done.
Re: (Score:2)
Re: (Score:2)
That's with CPUs. AI mostly uses GPUs and those are still improving pretty fast.
Short bursts of VR (Score:2)
Looks like this VR interest bubble burst in record time.
I suppose the next one will happen when Apple poots out it's AR glasses
and everyone and their sister copies it cheaper and with extra cheese.
I give that bubble 18 months at best.
Re: (Score:2)
Looks like this VR interest bubble burst in record time.
It'll keep bubbling as long as VR is closer to V than R.
Re: (Score:2)
VR getting gamers to play computer games standing up suffering nausea and vertigo instead of sitting down in comfortable chairs. They never did the great playoff, who would win in an eight hour FPS match, people using VR versus people using PCs a keyboard and a mouse (they probably did behind the scenes and it was probably a slaughter, those using VR getting wiped out and the longer it went the worse they got).
Augmented reality is just smart glasses hooked into smart phone, with sensors fitted to your finge
Re: Short bursts of VR (Score:2)
Even though it's a bit elaborate, swivel chairs and slipring/wireless should have been the standard solution. Shipping wired shit which couldn't handle free rotation poisoned the market.
Re: (Score:2)
Shipping wired shit which couldn't handle free rotation poisoned the market.
Which is why I was expecting the death of VR to happen yet again, until the Oculus Quest arrived. Wired VR requiring a powerful PC (and a closed operating system from a felonious organization) was an absurdity that did absolutely nothing to promote the genre. Wireless, completely self-contained VR (which the Quest is) was the next necessary revolution in VR technology that we had all been waiting for, and I am not disappointed.
I went from, "VR sucks!" to, "VR is awesome!" solely because of the Quest. If J
JC should stop... (Score:1)
... jumping around. He is good at making games, he really should be back in the gaming industry. I think he's had an itch as a nerd to scratch and is jumping around because he's lost. He seems to have lost touch with reality a bit. I knew back when he joined occulus VR tech was not going to be anywhere near ready in the timeframe Carmack wanted it to be. When most games are released unfinished, buggy and with bad camera's in the AAA space... what exactly is there to look forward to when your gamer cu
Re: (Score:2)
No, he was good at making game engines. iD were successful because of the team. DOOM had an engine build by John Carmack, character design and aesthetics by Adrian Carmack, level design by John Romero, and music by Bobby Prince. John Carmack wasn't great at game design as a whole. iD were also known for not making the most of John Carmack's engines. There were always games that looked more impressive built with licensed (and often modified) iD engines. He'd need to join up with another dream team to r
Re: (Score:1)
No, he was good at making game engines.
Good engine's are what make good games, it's not everything but competence with tools and engines matter. We've seen the decline of level editors and open sourcing games since JC left the industry and iD was sold. Things I'd like to see return.
Pace of AI (Score:4, Interesting)
"I really haven't been satisfied with the pace of progress that we've been making" in VR.
Wait until he finds out how slow the pace of AI research has been.
Winter is coming.
Re: (Score:2)
The REAL tragedy is that Carmack is undoubtedly tied up by noncompete agreements for the next few years, so he can't go out & use his fame to drum up enough venture capital to start a NEW VR company willing to spend whatever it takes to significantly raise the hardware bar and make everything that presently exists today obsolete overnight (the way the original Nexus One blew away every single phone manufacturer's roadmaps for the next five years... pre-N1, we'd get a few more kilobytes of ram, another h
Re: (Score:2)
Re: (Score:2)
You'd absolutely need something on par with the best mobile workstation available today from someone like Lenovo, HP, or Dell. It wouldn't be cheap, and would be outrageously expensive if you allocated 100% of the cost to "VR"... but in reality, VR would be just one of many things you'd use that computer for, even if VR were the primary driver for buying one THAT expensive.
That's the biggest problem with the "Quest" approach. If you're dedicating hardware ENTIRELY and EXCLUSIVELY to VR, you'll be limited to
Re: (Score:2)
High refresh rates alone doesn't solve the problem with people getting sick. Artificial locomotion is much more likely to get you sick than a "low" 90hz refresh rate. The amount of folks getting sick at 90hz (while avoiding artificial locomotion) is quite small.
There are ultra high resolution HMDs [varjo.com] already (Valve Index and Pimax also support higher than 90hz) but it's far from the missing piece of the puzzle.
What VR really needs is better controllers, haptics, and interaction systems. A compelling VR experie
Re: (Score:2)
Carmack doesn't seem to be interested in PCVR though. Whenever he gives his yearly talks at Oculus Connect events he almost exclusively talks about mobile VR. First it was GearVR, then Go and now Quest. I think he was having more fun playing with lower end hardware and making it do more than what anybody thought possible than worrying about the bleeding edge.
Sure, he probably wants all those things you mentioned but I don't think he left because they weren't launching a product with high end specs.
Re: (Score:1)
Its all fun until the funding stops again...
the usual VR (Score:2)
VR has been going nowhere for... decades. Every time the hype starts up, there's new products, new applications, new venture capital - and then it dies down again when people start to realize that even if humans are visual creates, graphics isn't everything. The few half-hearted attempts to add haptics usually come too little, too late, and the next VR hype is over.
If I were an investor, I'd calmly put money into a company that tries to solve the non-visual aspects of VR, let them tinker on it for the 5-10
Re: (Score:1)
VR will never go anywhere until they solve the inner ear problem.
Re: (Score:2)
Wonder what happened to this [uploadvr.com]... That being said. I used to get sick in VR when using artificial locomotion very (under a minute) quickly. Now it's pretty rare I get sick at all unless it's an unusually poorly implemented solution.
Finding your VR legs, higher refresh rates, and more intelligent movement algorithms can do quite a bit.
I blame Facebook. (Score:1)
And I knew Carmack made a mistake going to work for them. I guess now he knows it too.
Magic Number (Score:1)
Title WRONG, he will be working on AGI. (Score:2)
Facebook bought Oculus (Score:1)
So who cares if Carmack is leaving, wouldn't you, to me as a Rift owner it's over anyway.
kiss of death (Score:2)