John CarmackVerified account

@ID_AA_Carmack

Independent newbie AI researcher, Consulting CTO Oculus VR, Founder Id Software and Armadillo Aerospace

Dallas, TX
Joined August 2010

Tweets

You blocked @ID_AA_Carmack

Are you sure you want to view these Tweets? Viewing Tweets won't unblock @ID_AA_Carmack

  1. I’ll be doing a talk Saturday (2/1) at the Frontiers of Flight Museum in Dallas from 5-7 pm on STEM fields. Stop by if you are in the area!

    Undo
  2. Changing this now would probably be too traumatic, but I tend to think that VR controllers would be better with a crisp, binary trigger instead of a squishy analog one. Games like Audica and Pistol Whip would be objectively improved. Sound off if you would hate that.

    Undo
  3. I was puzzled by this (reading about SPSA): "uniform and normal distributions do not satisfy the critical finite inverse moment condition" Later, I figured out that it was just because they would cause divide-by-zero problems, unlike the strictly -1/1 Bernoulli distribution.

    Undo
  4. Nvidia should sort out the firmware and drivers to allow cards to be chained by NVLink alone, without PCIe, bypassing all the bios, chipset, and OS issues. A ring of hundreds of GPUs would serve many applications that don't need a lot of host bandwidth.

    Undo
  5. \ monopropellant. You would know after a few million $ of engine development (and many, many explosions) if it was worth going full scale. Chasing a conventional path is unlikely to provide meaningful competitive advantage over SpaceX.

    Show this thread
    Undo
  6. \ perfectly and no mistakes were made along the way, a small orbital launch might have been made for $20M, but that isn't at all likely. If I took another stab at it today, I would forsake reusability and try to build an expendable SSTO using crazy-dangerous mixed cryogenic \

    Show this thread
    Undo
  7. Not much has changed that would make it materially easier today than it was for SpaceX. There is more chance involved than most would like to admit -- their first launch could have made it, and the project cost would have been much lower. If everything had gone \

    Show this thread
    Undo
  8. Given the ever-widening compute / bandwidth gap, maybe we should throw a little bit of hardware at high performance "random" sequence generation. You can do a lot with seeded randoms if you are up against the memory wall.

    Undo
  9. Considered horizontally flipping my video data set for more variety, but realized that would be a good way to train a dyslexic AI.

    Undo
  10. I had forgotten the name of this algorithm, and had to search around a bit: Useful for sampling from distributions of initially unknown size or unknown weighted sums.

    Undo
  11. Retweeted

    My 5yo asked, “How big is a wolf?” So I Google it... Google’s first result is the option (in browser!) to place a realistic wolf in the room with us so we can walk around it and see for ourselves. Magical. The closest I have felt to a Young Lady’s Illustrated Primer.

    Undo
  12. Retweeted
    Jan 21
    Replying to

    Thank you for bringing up as an example. Those interested in the project, currently beta testing on the Ethereum mainnet, you can check the recently published user manual: The RNDR network is functional as of now:

    Undo
  13. If someone wants to put some rigor in my “order of magnitude” WAG, calculate the cost of mining Ethereum on AWS GPU instances.

    Show this thread
    Undo
  14. Crypto mining is a nice, public benchmark of what cost-optimized computation can do. For GPU workloads that could run on those systems,, the “crypto GPU cloud” is an order of magnitude cheaper than conventional cloud. Less obviously, it is also two orders of magnitude bigger.

    Show this thread
    Undo
  15. Start a VR game studio with . That sounds like a high outcome variance prospect.

    Undo
  16. ... comes when strictly using the web for quality reference material, but it would be a discipline challenge to avoid all the people and fluff vying for attention. It is easier to just say “none of that at all” than make constant judgements.

    Show this thread
    Undo
  17. ... more focused, but not magically so. The next day, when I resumed checking twitter / email / workplace / messenger, it was clear that it consumed a good chunk of the day, but I had my knowledge-of-the-world reference library back. Peak research productivity for me probably ...

    Show this thread
    Undo
  18. I was doing a remote study/think time this week, and I tried a “no screen time day” as an experiment, restricting myself to printed books and papers. It was inconclusive. I missed finding instant answers and chasing references at least a dozen times during the day. I was ...

    Show this thread
    Undo
  19. A SciFI book about an inadvertently sentient computer program from 1977. Similar vein to Hartling’s Avogadro books 35 years later. Not bad for the time — owning all the IBM 360s.

    Undo
  20. Interpolated 60 FPS is a nice feature, but someone should break down and actually change the code to let it run at unlocked frame rates and G-sync off the mouse update at 250 FPS on PC.

    Undo

Loading seems to be taking a while.

Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.

    You may also like

    ·