14 ms·

In the online world, it seems like there are two big things happening right now: - Neural networks / ML (eg GPT-2) Definitely nowhere near its potential for be

In the online world, it seems like there are two big things happening right now: - Neural networks / ML (eg GPT-2) Definitely nowhere near its potential for being applied to a wide variety of areas. Find a niche you like and apply there. - Security / Privacy (eg Telegram) Rapidly growing demand pretty much everywhere. Bonus points if you can make your product great for standard users and at the same time hackable/customizable for people who want to do that. Capitalize on both legs of the pareto distribution. That all being said, if you are ambitious and talented without an all-consuming passion for software, I highly recommend you find something you can work in hardware. Since the '70s or so most industries have been basically frozen, besides computer hardware/software. Yet in the meantime materials science and engineering design has advanced considerably, both of which form the basis for innovation in new technologies. This is why SpaceX was able to build components for 10-100x cheaper than the leading suppliers in the early 2000s. I work at a startup in nuclear fission, particularly because this tech is at <1% of its potential right now. The same could be said for many other areas. Here's some ideas you might find interesting, that I think could work in the next decade or so: - Supersonic air travel - Electric air travel - Nano/micro-scale metallurgy and materials for industry - Biological materials - Gut/microbiome - Genetic engineering - Nuclear fission / fusion - Carbon capture - Cross-laminated timber (CLT) for construction - Indoor farming / optimizing farming in general - Synthetic meat / meat alternatives

el_dev_hell 7y ago
> That all being said, if you are ambitious and talented without an all-consuming passion for software, I highly recommend you find something you can work in hardware. Since the '70s or so most industries have been basically frozen, besides computer hardware/software. Yet in the meantime materials science and engineering design has advanced considerably, both of which form the basis for innovation in new technologies. This is why SpaceX was able to build components for 10-100x cheaper than the leading suppliers in the early 2000s. I honestly feel like reading this is a moment that I will l recall in about 8 years and think: "Damn, I wish I listend to that comment about hardware from johnmorrison".
- GhettoMaestro 7y ago
  How do we act on it, instead? (not being a dick)
  - johnmorrison 7y ago
    General answer: Become a specialist, or drive funding/regulatory pushes towards these new hardware technologies. Figure out what you see as the barriers to their commercialization, then figure out what component of that problem you'd love to work on, and become an expert in that. My biased answer: If you're in the position to spend a lot of time learning and experimenting with new stuff (e.g. a highschool or uni student), I'd recommend becoming an expert somewhere along the liquid fueled fission reactor stack. Chem eng, nuclear eng, rare earth extraction, uranium extraction, working with molten salts, working with liquid metals, anything like that. If all goes well, I'll be needing people in just about every specialization there.
    - acidburnNSA 7y ago
      I love how you say you're working on fission because it's at <1% of its potential. That's a really wise perspective. > If all goes well, I'll be needing people in just about every specialization there. How far along are you in the Oak Ridge guys' elaborately detailed Molten Salt Reactor development program plan from 1974 [1], as far as it's relevant? If parts of it are irrelevant, how significant are the things you have to do that they didn't? [1] https://www.osti.gov/biblio/4227904 https://www.osti.gov/biblio/4227904
      - johnmorrison 7y ago
        There are some things that have become irrelevant for now, particularly the need for a stellite alloy. Most of the things we are doing differently from ORNL in our division for thermal MSRs is based on the progress in materials and CAD since the winding down of the ORNL MSRs. There are a lot of promising mostly mix-able design choices in MSRs that have become apparent in the last few decades, so we are never fully committed to a particular set.
ncmncm 7y ago
The most pressing problem for humanity is global climate disruption. Arguably, working on anything that doesn't help there is not just wasted effort, but actively harmful. Wind, solar power, energy storage still have huge room for improvement. Direct solar or wind -> liquid fuel will be essential to adapting fast. Windmills can drive production of liquid ammonia on farms (needing only air and water inputs), for use directly as fertilizer and fuel, without need for a grid attachment and without blocking sunlight needed for the plants. Ammonia is not a very dense fuel (e.g. terrible for aircraft), but that doesn't matter for farm machinery. Direct solar -> hydrogen has been demonstrated, with bio-reaction for hydrogen + CO2 -> liquid fuel suitable for aircraft. Direct hydrogen-fueled aircraft are feasible, and more efficient than with kerosine, but the design cycle is too long. Wind turbines will be wearing out as the blades erode. No-moving-parts screens might be the next generation, extracting power by releasing ions to be carried away from an electric field. The old towers will still be useful, and the rare earths can be mined out for other uses, probably vehicle motors. Batteries are a very material-intensive storage medium. Underwater air storage does not need exotic materials or tech, and the pressure at depth makes the strength of materials needed minimal, other than piping. We need to replace huge amounts of refrigeration equipment with versions that don't rely on HFCs, and get the HFCs incinerated. One gram of HFC traps as much heat as 2500 g of CO2, and lasts centuries in the atmosphere. Once vented, it cannot be recaptured. Ammonia-cycle systems need to be made safe enough for general use, and HFC versions outlawed. Without massive progress in the next decade, civilization will probably collapse by 2035.
- opencl 7y ago
  HFCs are already being phased out in favor of HFOs (though you could certainly argue that this should be happening more quickly), is there a strong reason to go for ammonia based systems over HFO?
  - ncmncm 7y ago
    Thank you, the atmospheric half-life of HFOs is short enough to make it a viable alternative to ammonia as a refrigerant. However, the existing stocks of HFCs, in installed systems, is enough to match, if vented, the heating from all the CO2 currently in the atmosphere.
ryanlol 7y ago
>- Security / Privacy (eg Telegram) But Telegram offers just as much security and privacy as Slack.
- eitland 7y ago
  1. They are vey different. Those who arent aware of the large differences shouldn't comment as if they are they are subject experts. Telegram might be full of holes but it is a very different thing to Slack. (I use both for different things.) 2. I have my own issues with Telegram (the refusal to introduce sustainable funding, pushing of crypto currencies, possibly also marketing it as more than it is) but except that there's a lot to like about it and I think it is a decent starting point if someone with strong cryptographic skills wants to start with a decent client and go ahead to provide something better (either introduce e2e-encryption or improve, document and provide some way to verify the current solution.)
  - johnmorrison 7y ago
    E2E encryption is kind of a fucky thing, some suggest it is fundamentally impossible for central web services and even mobile / desktop apps. Basically, if there is a third party involved in the code besides Alice and Bob, the two can never guarantee E2E encryption. Web part is simple, as there are countless ways you can get malicious code delivered from what you think is the correct, safe, web server. Same concept applies to app store / desktop program updates, but with a slightly slower and more difficult attack prospect. Anyways, Telegram does seem to have the best encryption of any major service, and it's what we use at my company for almost all internal communications.
    - ryanlol 7y ago
      >E2E encryption is kind of a fucky thing, some suggest it is fundamentally impossible for central web services and even mobile / desktop apps "some suggest" How about you actually name some competent people who suggest this? >Basically, if there is a third party involved in the code besides Alice and Bob, the two can never guarantee E2E encryption. Oh, this is a downright insane, dishonest argument. Perfection is impossible, so we shouldn't even try! >Anyways, Telegram does seem to have the best encryption of any major service, and it's what we use at my company for almost all internal communications. Why do you think this? This is such a fundamentally ridiculous claim, I find it absolutely fascinating that someone might arrive at this conclusion.
luc4sdreyer 7y ago
>Here's some ideas you might find interesting, To expand on your key ideas: it looks pretty comprehensive but if you're interested in making the world a better place as well, have a look at 80,000 hours [1]. They've been thinking about this question for at least 8 years and it's quite extensive. They have a simple quiz that might also be an interesting starting point [2]. [1] https://80000hours.org/key-ideas/ https://80000hours.org/key-ideas/ [2] https://80000hours.org/career-quiz/#/ https://80000hours.org/career-quiz/#/
- vinliao 7y ago
  Hey luc4sdreyer, thanks for the link! I'm about to graduate college soon and I find the 80000hours website extremely helpful. Randomly encountering this site is one of the best thing that happened this week, thank you very much!
closeparen 7y ago
As someone using ML in production at scale, it's not the sophistication of the models themselves, but all the "boring" operational stuff around them. - Keeping them fed with clean, high quality, low latency feature data. - Understanding the impact of an ML intervention on the overall system or business process, in aggregate and for all the relevant subpopulations. - Understanding why scores may be drifting. These are very challenging alerts to investigate. - Resource efficiency at very high QPS or in mobile contexts.
- alexhutcheson 7y ago
  + Running constant live experiments on actual traffic to make sure it continues to do what you think it does.

el_dev_hell 7y ago

> That all being said, if you are ambitious and talented without an all-consuming passion for software, I highly recommend you find something you can work in hardware. Since the '70s or so most industries have been basically frozen, besides computer hardware/software. Yet in the meantime materials science and engineering design has advanced considerably, both of which form the basis for innovation in new technologies. This is why SpaceX was able to build components for 10-100x cheaper than the leading suppliers in the early 2000s. I honestly feel like reading this is a moment that I will l recall in about 8 years and think: "Damn, I wish I listend to that comment about hardware from johnmorrison".

GhettoMaestro 7y ago
How do we act on it, instead? (not being a dick)
- johnmorrison 7y ago
  General answer: Become a specialist, or drive funding/regulatory pushes towards these new hardware technologies. Figure out what you see as the barriers to their commercialization, then figure out what component of that problem you'd love to work on, and become an expert in that. My biased answer: If you're in the position to spend a lot of time learning and experimenting with new stuff (e.g. a highschool or uni student), I'd recommend becoming an expert somewhere along the liquid fueled fission reactor stack. Chem eng, nuclear eng, rare earth extraction, uranium extraction, working with molten salts, working with liquid metals, anything like that. If all goes well, I'll be needing people in just about every specialization there.
  - acidburnNSA 7y ago
    I love how you say you're working on fission because it's at <1% of its potential. That's a really wise perspective. > If all goes well, I'll be needing people in just about every specialization there. How far along are you in the Oak Ridge guys' elaborately detailed Molten Salt Reactor development program plan from 1974 [1], as far as it's relevant? If parts of it are irrelevant, how significant are the things you have to do that they didn't? [1] https://www.osti.gov/biblio/4227904 https://www.osti.gov/biblio/4227904
    - johnmorrison 7y ago
      There are some things that have become irrelevant for now, particularly the need for a stellite alloy. Most of the things we are doing differently from ORNL in our division for thermal MSRs is based on the progress in materials and CAD since the winding down of the ORNL MSRs. There are a lot of promising mostly mix-able design choices in MSRs that have become apparent in the last few decades, so we are never fully committed to a particular set.

ncmncm 7y ago

The most pressing problem for humanity is global climate disruption. Arguably, working on anything that doesn't help there is not just wasted effort, but actively harmful. Wind, solar power, energy storage still have huge room for improvement. Direct solar or wind -> liquid fuel will be essential to adapting fast. Windmills can drive production of liquid ammonia on farms (needing only air and water inputs), for use directly as fertilizer and fuel, without need for a grid attachment and without blocking sunlight needed for the plants. Ammonia is not a very dense fuel (e.g. terrible for aircraft), but that doesn't matter for farm machinery. Direct solar -> hydrogen has been demonstrated, with bio-reaction for hydrogen + CO2 -> liquid fuel suitable for aircraft. Direct hydrogen-fueled aircraft are feasible, and more efficient than with kerosine, but the design cycle is too long. Wind turbines will be wearing out as the blades erode. No-moving-parts screens might be the next generation, extracting power by releasing ions to be carried away from an electric field. The old towers will still be useful, and the rare earths can be mined out for other uses, probably vehicle motors. Batteries are a very material-intensive storage medium. Underwater air storage does not need exotic materials or tech, and the pressure at depth makes the strength of materials needed minimal, other than piping. We need to replace huge amounts of refrigeration equipment with versions that don't rely on HFCs, and get the HFCs incinerated. One gram of HFC traps as much heat as 2500 g of CO2, and lasts centuries in the atmosphere. Once vented, it cannot be recaptured. Ammonia-cycle systems need to be made safe enough for general use, and HFC versions outlawed. Without massive progress in the next decade, civilization will probably collapse by 2035.

opencl 7y ago
HFCs are already being phased out in favor of HFOs (though you could certainly argue that this should be happening more quickly), is there a strong reason to go for ammonia based systems over HFO?
- ncmncm 7y ago
  Thank you, the atmospheric half-life of HFOs is short enough to make it a viable alternative to ammonia as a refrigerant. However, the existing stocks of HFCs, in installed systems, is enough to match, if vented, the heating from all the CO2 currently in the atmosphere.

ryanlol 7y ago

>- Security / Privacy (eg Telegram) But Telegram offers just as much security and privacy as Slack.

eitland 7y ago
1. They are vey different. Those who arent aware of the large differences shouldn't comment as if they are they are subject experts. Telegram might be full of holes but it is a very different thing to Slack. (I use both for different things.) 2. I have my own issues with Telegram (the refusal to introduce sustainable funding, pushing of crypto currencies, possibly also marketing it as more than it is) but except that there's a lot to like about it and I think it is a decent starting point if someone with strong cryptographic skills wants to start with a decent client and go ahead to provide something better (either introduce e2e-encryption or improve, document and provide some way to verify the current solution.)
- johnmorrison 7y ago
  E2E encryption is kind of a fucky thing, some suggest it is fundamentally impossible for central web services and even mobile / desktop apps. Basically, if there is a third party involved in the code besides Alice and Bob, the two can never guarantee E2E encryption. Web part is simple, as there are countless ways you can get malicious code delivered from what you think is the correct, safe, web server. Same concept applies to app store / desktop program updates, but with a slightly slower and more difficult attack prospect. Anyways, Telegram does seem to have the best encryption of any major service, and it's what we use at my company for almost all internal communications.
  - ryanlol 7y ago
    >E2E encryption is kind of a fucky thing, some suggest it is fundamentally impossible for central web services and even mobile / desktop apps "some suggest" How about you actually name some competent people who suggest this? >Basically, if there is a third party involved in the code besides Alice and Bob, the two can never guarantee E2E encryption. Oh, this is a downright insane, dishonest argument. Perfection is impossible, so we shouldn't even try! >Anyways, Telegram does seem to have the best encryption of any major service, and it's what we use at my company for almost all internal communications. Why do you think this? This is such a fundamentally ridiculous claim, I find it absolutely fascinating that someone might arrive at this conclusion.

luc4sdreyer 7y ago

>Here's some ideas you might find interesting, To expand on your key ideas: it looks pretty comprehensive but if you're interested in making the world a better place as well, have a look at 80,000 hours [1]. They've been thinking about this question for at least 8 years and it's quite extensive. They have a simple quiz that might also be an interesting starting point [2]. [1] https://80000hours.org/key-ideas/ https://80000hours.org/key-ideas/ [2] https://80000hours.org/career-quiz/#/ https://80000hours.org/career-quiz/#/

vinliao 7y ago
Hey luc4sdreyer, thanks for the link! I'm about to graduate college soon and I find the 80000hours website extremely helpful. Randomly encountering this site is one of the best thing that happened this week, thank you very much!

closeparen 7y ago

As someone using ML in production at scale, it's not the sophistication of the models themselves, but all the "boring" operational stuff around them. - Keeping them fed with clean, high quality, low latency feature data. - Understanding the impact of an ML intervention on the overall system or business process, in aggregate and for all the relevant subpopulations. - Understanding why scores may be drifting. These are very challenging alerts to investigate. - Resource efficiency at very high QPS or in mobile contexts.

alexhutcheson 7y ago
+ Running constant live experiments on actual traffic to make sure it continues to do what you think it does.