2,240 karma · joined August 2, 2013
How do these AI systems account for causation?
It's the fundamental issue I have with the penetration of these sorta "AI" systems in daily life: from what I understand about the basic maths, they're ultimately all seeking correlation and real-world data used to improve models seems to achieve verification, not validation.
I'm guessing I'm not the only person who reads this article and experiences a visceral reaction.
This is what I would call "ugly" tech.
There are quite a few constellations that you can benchmark by: - Astrocast (IoT) [1] - Fleet (IoT) [2] - Iceye (Earth Obs) [3] - Planet (Earth Obs) [4] - Spire (Earth Obs, Navigation) [5] - Capella Space (Earth Obs) [6]
And many more ...
I think the whole Mars One story will make it into a Hollywood script, because there's been more drama than most good movies I've watched in recent times.
There are thoughts about the project that I'd rather not share publicly, especially since I also know one of the main authors of the MIT study that became the focus of a lot of the critique about the feasibility/viability of the entire project.
Safe to say, Mars exploration is hyper-complex because of complexity in a whole host of dimensions, not just technical/scientific. Mars One exemplifies that.
I'm involved in Mars analog simulations missions through the Austrian Space Forum [1], so I've got a pretty good reading of the complexity that goes into planning & operational excellence.
If I had to sum up the whole Mars One journey in one word, it would be: underestimation.
[1] https://oewf.org
Within my domain of expertise (space engineering), the problem is even worse. Not only do suppliers provide haphazard information about parts through PDF documents, there's very little, if any, real convergence on what to actually provide to the engineer.
I'm confident that these highly unstructured PDF datasheets are going to go the way of the dinosaurs, but it's quite challenging to integrate any smarter solutions into the business chain.
My domain of expertise is the space industry, and it's stunning how much of a mess datasheets are. As an engineer on different missions, I couldn't understand why I was spending so much time trying to "undo" all of the work the suppliers had done to create these documents, to be able to get the data into into the tools that I needed to use and actually start my real work.
In the space industry at least, there's a concerted effort going on to try to replace these PDF documents with electronic, machine-readable equivalents. Through the European Cooperation for Space Standardization (ECSS) [1] and work being done by agencies like ESA [2], and private industry [3], but really it's not yet taking hold for a lot of operational reasons.
We're actually working on trying to digitalize the entire global supply chain by converting PDF datasheets into something that allows more intelligent engineering work. We're currently serving "level-1" through our website: https://satsearch.co.
We've also prototyped a way to cast the datasheets into something more usable with ESA and a number of other partners. We've currently serving this digitalized data through our API: https://api.satsearch.co, which is being integrated into industry tools.
The real dream for me is to get to the point where the engineer can work with this kinda data natively within their tools and skip the entire PDF datasheet fiasco. Ideally, the data can then even be operated on my multi-objective optimization algos to speed up the entire design process.
Disclaimer: I'm one of the co-founders at satsearch and we started our startup precisely because I was just sick of dealing with PDF datasheets.
[1] http://ecss.nl/glossary/eds/
[2] https://indico.esa.int/event/252/contributions/3844/attachme...
[3] https://indico.esa.int/event/22/contributions/1997/attachmen...
Pretty remarkable that a company like Amazon allows for experiments like Ground Station. They have a pretty clear business model and I think that it will help to open up the sector even further, in the sense that now you can just grab your satellite data in pretty much the same way you operate on AWS otherwise.
It's also interesting, because Amazon came somewhat late to the game and sorta blew open the market. Friends of mine founded Leaf Space [1] and they've sorta built a similar concept from the ground up, deploying antenna around the world.
Then you've got efforts like KSAT Lite [2], InfoStellar [3], ATLAS [4], and RBC Signals [5], so it's a pretty hot market at the moment.
I'm excited about AWS entering into the market because I believe it'll accelerate development of some pretty cutting-edge technologies like laser satcom, that will change the way we think of data from space.
[2] https://www.ksat.no/en/news/2016/january/ksat%20lite-network
[3] https://www.infostellar.net
I kinda draw the analogy of how the Raspberry Pi went from being a learning tool, to something that hackers/makers saw as an opportunity to innovate on. CubeSats went through a similar transition.
The work that then led to commercial activities came from a lot of different people have similar ideas about how commoditization can help to reduce costs and allow for riskier projects.
There's actually a strong bias in the media that doesn't reflect the diversity of projects, missions, and technologies on the market at the moment. The idea of using a platform like CubeSats to open up the space sector is a lot more widespread than I think more people are aware of.
There's definitely a changing attitude towards what can be launched into space. The "cheap" satellites are not going to necessarily achieve the extreme performance needed for some missions, but by being more commercial, there's a new wave of missions that target specific, market-driven use-cases, and they don't all need the latest/greatest/fanciest tech.
Additionally, the orbit you fly in affects this conversation a lot. For Low Earth Orbit <1000 km, I'm confident that the market is going to be dominated by commodity hardware, except for flagship missions perhaps.
Higher altitudes require greater reliability due to radiation and are just more expensive orbits to get to. So I think it's going to take a little longer. Having said that, there are a bunch of companies tackling the higher orbits, like SpaceX with Starlink [3], and even YC alum Astranis [4] in GEO.
The way I like to think of it is that the space sector is maturing from a craft to a profession, and with that, commoditization offers a clear path towards scalability.
[2] https://www.planet.com/company/approach
[3] https://en.wikipedia.org/wiki/Starlink_(satellite_constellat...
Groups like the Libre Space Foundation [1] and LibreCube [2] are doing a great job of opening up the whole stack (software+hardware for space, launch, and ground systems). Projects like Ardusat [3] are also helping to "democratize" space, to use a cliche. Ardusat actually led to the founding of Spire [4], one of many commercial EO constellation companies that have benefited from the process of commoditization.
The "giant leap" really is closing a lot faster than people realize.
If you're interested in exploring the kind of hardware on the space market, check out https://satsearch.co. We've currently got 5,000+ products & services for the space market listed, with the goal being to use data to crack open the supply chain, enabling spin-in, spin-out, tech transfer, etc.
Disclaimer: I'm one of the co-founders at satsearch.
[3] https://www.kickstarter.com/projects/575960623/ardusat-your-...
Firstly, digital filing is the norm and for "simple" returns, just works.
Secondly, if you ever have a question, you can just call the tax authority on their free phone number, and pretty much ask them anything related to individual or business tax returns.
I've been messing for a while with my own boilerplate C++ project [0], since I got sick of having to recreate the CMake set up for every app I created. It's taken a lot of CMake magic to get it working and I'm still not totally happy with the config using ExternalProject_add to automagically deal with dependencies that aren't present.
I've been evaluating Hunter [1] and wondering whether to step over.
If I understand correctly, the big advantage of Bake is that it completely replaces CMake for dependency management (at least that's what I take away from the readme)?
Exciting!
I'm much more used to clicking on links that are organically ranked high in DDG or Google, when I am specifically looking to buy something. I have to admit, I barely make a handful of purchases per year, so perhaps I'm just totally disconnected from this whole topic.
On the one side, you can't get through most days without an article about SpaceX, Blue Origin, Planet, Spire, etc. popping up, but on the other hand, the market is still pretty much dominated by the traditional large players.
We're already "weird" by the fact that we're ONLY building software, and no hardware. Further more, our software is not in itself a product, since we're building a marketplace of sorts. So what we're really trying to attack are the business and engineering processes.
Although we haven't been actively fundraising, I've talked to quite a few investors, and nobody seems to care, because if you're not building a satellite, a rocket, a space station, or something similar, you're not really in the game.
We're banking on the fact that commercialization of the space industry is coming, and so business and engineering processes will have to adapt.
Convincing people of this to the point where they're willing to put money on the table is proving super difficult.
The solution for us has been to not fundraise and try to minimize burn to the point where we can bootstrap for a very long time.
From this article, it seems we're in the "behavior change" category. The article only briefly touches on this, and I'd be interested to understand how to be effective in communicating that behavior change to investors, especially in cases where things like TAM are largely "? x ?".
Anyone with additional tips/resources?
I cannot think of many things more horrific. It's not AI that scares me, it's people. The nonchalance with which this sentence is woven into a buzzword-dominated article shocks me to the core.
People are truly horrific.
This has persisted for more than a day.
No response to my report submitted through the help center yesterday, except the standard auto-reply.
Looks like there are quite a few people who have disappearing data and there seems to be no acknowledgement of the issue on the FB side.
[1] https://techcrunch.com/2018/03/01/astranis-emerges-from-stea...
I thought I'd kick off the new year by submitting an open-source project of mine that I haven't been tracking on Github at all until recently. I happened to check my Github account a few weeks ago and saw that it had quite a few stars (at least for my standards). I figured maybe it's useful to more people.
Moreover, I'd be happy for any and all feedback. This project is the culmination of fighting with CMake and C++ for a few years and I use it as the basis for most new codebases that I build.
I have a few ideas on how to expand it further and would be grateful for input, even if it means ripping the project to shreds.
Thanks!
> longer-term it is about the data and using it for other technologies.
I'm not sure what this means. Better adtech? Or is there something else unique about this data that I'm not seeing?
The other thing that I really appreciated was the fact that everyone in the batch that participated in the calls was treated individually by our mentor, Gleb. Not entirely sure how he achieved this but I can definitely say that if you put the hours in, the program helped organized thoughts and streamline energy.
To close out, my mandatory plug :)
We're building the first global marketplace for the space industry: "Digi-Key for space". Our presentation page: https://www.startupschool.org/presentations/465.
If anyone is interested in NewSpace, feel free to reach out!