3 ms·

I guess one glaring difference might be that they got this right when it mattered. Not a single Saturn V ever failed during launch. The aerospace engineering cu

I guess one glaring difference might be that they got this right when it mattered. Not a single Saturn V ever failed during launch. The aerospace engineering culture is vastly different than the "engineering" that goes on in ML/AI, and software development more generally. I have a lot of respect for it.

Iv 8y ago
The cost of failure is higher in aerospace. But I must say that in my job, when you manage to make your modified code pass all the thousands of unit tests and see the benchmark results increase and the 80 cores of the test machine firing up, feeding GB/s of data in their pipeline at the optimal fillrate, there is something of of a "ignition aaaaaand liftoff!" moment that is deeply satisfying. I am just allowed to try pressing the red button several times a day instead of once every two years.
- perl4ever 8y ago
  80 modern processors have around four orders of magnitude more "moving" parts than a Saturn V, according to what I can Google. And the Saturn V, it is said, had a particularly large number of parts because it was hand-made to an extent that modern machinery is not. Comparing the number of parts underestimates the complexity of a computer system, of course, because the software is a machine with a vast number of parts, on top of the hardware. In other words, I'm on the side of those who say software is hard to get right for good reason, and not because the practitioners are less competent.
mikeash 8y ago
Saturn V suffered several engine failures. It had persistent pogo oscillations that put Apollo 6 into the wrong orbit and forced much of its test plan to be canceled, and nearly destroyed Apollo 13 during launch. Vibrations severely damaged Skylab during its launch. I would not say it never failed. It never went kaboom in the sky, but there are other ways to fail.
- sidlls 8y ago
  It was an exaggeration, but the comment's underlying point is spot on: the "engineering" that we do doesn't hold up in terms of rigor or cleverness that AE and other related engineering disciplines exhibit.
  - krisrm 8y ago
    Counterpoint: if we tried to build all software using the same rigorous and "clever" processes that aerospace engineers do, almost nothing would ever get built. It's a different process because it's a different field, not because aerospace engineers are fundamentally more clever. Signed, Not an aerospace engineer ;)
    - mcguire 8y ago
      Have an aerospace engineer cow-orker; one favorite quote for why he works in enterprise software is, "Because this flies and that (pointing at an Ares-1 mockup) doesn't." On the other hand, there's nothing "engineering" about software engineering.
      - krisrm 8y ago
        Hahaha, harsh. I like your coworker. And I would tend to agree with your second point. We might occasionally borrow some formalism and organizational structure from our engineering cousins, but it's different in enough ways that it probably shouldn't bear the same moniker.
  - mikeash 8y ago
    While I generally agree with that underlying point, if our computers were as unreliable as the Saturn V, we’d never get anything done with them.

Iv 8y ago

The cost of failure is higher in aerospace. But I must say that in my job, when you manage to make your modified code pass all the thousands of unit tests and see the benchmark results increase and the 80 cores of the test machine firing up, feeding GB/s of data in their pipeline at the optimal fillrate, there is something of of a "ignition aaaaaand liftoff!" moment that is deeply satisfying. I am just allowed to try pressing the red button several times a day instead of once every two years.

perl4ever 8y ago
80 modern processors have around four orders of magnitude more "moving" parts than a Saturn V, according to what I can Google. And the Saturn V, it is said, had a particularly large number of parts because it was hand-made to an extent that modern machinery is not. Comparing the number of parts underestimates the complexity of a computer system, of course, because the software is a machine with a vast number of parts, on top of the hardware. In other words, I'm on the side of those who say software is hard to get right for good reason, and not because the practitioners are less competent.

mikeash 8y ago

Saturn V suffered several engine failures. It had persistent pogo oscillations that put Apollo 6 into the wrong orbit and forced much of its test plan to be canceled, and nearly destroyed Apollo 13 during launch. Vibrations severely damaged Skylab during its launch. I would not say it never failed. It never went kaboom in the sky, but there are other ways to fail.

sidlls 8y ago
It was an exaggeration, but the comment's underlying point is spot on: the "engineering" that we do doesn't hold up in terms of rigor or cleverness that AE and other related engineering disciplines exhibit.
- krisrm 8y ago
  Counterpoint: if we tried to build all software using the same rigorous and "clever" processes that aerospace engineers do, almost nothing would ever get built. It's a different process because it's a different field, not because aerospace engineers are fundamentally more clever. Signed, Not an aerospace engineer ;)
  - mcguire 8y ago
    Have an aerospace engineer cow-orker; one favorite quote for why he works in enterprise software is, "Because this flies and that (pointing at an Ares-1 mockup) doesn't." On the other hand, there's nothing "engineering" about software engineering.
    - krisrm 8y ago
      Hahaha, harsh. I like your coworker. And I would tend to agree with your second point. We might occasionally borrow some formalism and organizational structure from our engineering cousins, but it's different in enough ways that it probably shouldn't bear the same moniker.
- mikeash 8y ago
  While I generally agree with that underlying point, if our computers were as unreliable as the Saturn V, we’d never get anything done with them.