This is a fantastic website. Currently, it primarily displays satellite positions, but I think it would be even better if it could be displayed or filtered by orbit type. The current results clearly show the differences between different orbit types—the Central Park observations you provided already clearly demonstrate these differences—the Sentinel-2 satellite passes by around 10:50 local time (sun-synchronous orbit), the GOES-19 satellite continuously observes from geostationary orbit, and the International Space Station/Tiangong satellite passes by at different times each day (inclined low Earth orbit). This website would be even more educational if it could color-code or label the satellites according to orbit type (sun-synchronous orbit/geostationary orbit/inclined low Earth orbit/medium Earth orbit), or even label the local time for nodes for sun-synchronous orbit satellites.
Beautifully done, at so many levels: the pincushion view of just how much stuff is up there; the rendering of the orbit of each satellite; the ground track of its aperture; the time series of its passes. So many data sources seamlessly integrated. Even if Claude wrote every line of code, knowing what to ask for is quite a feat.
I think this is a great counterexample to AI sites that suck. It's not the Claudisms and the borders-and-pills that make them suck, because this has all that but is fantastic.
Nice site. Likely very few people will care this much, but with a couple of very easy changes you would make this feel a LOT less vibe coded.
The giveaways:
* Instrument Sans (your logo)
* Highlighted period in your logo
* Highlighted italicized text in your header text (also in Instrument Sans)
* Short punchy AI-written sentences. Just merge that into one sentence! ("Enter any place on Earth to see every satellite that passed over it in the last week and the pictures they took there.")
* The flashing orange square in the upper right (every single vibe coded site has a circular or square version of the LIVE icon)
And then, just one more reason to switch from Instrument Sans: the massive kerning gap between "W" and "atched" in the logotype.
When the page is first loading and the logo is shown in the middle of the screen, the period is pulsating. I think the intention is that the colored period is to evoke the look of a glowing red LED light from a camera that is recording. It fits with the name “watched”.
Now whether it looks better putting it as a period at the end of the word in the logo like that, instead of as a solid circle ahead of the word, is perhaps debatable.
My eyes rolled, at first. Brilliant website, love it. But it is the visual equivalent of Claude-isms in text. I have seen hundreds of pages in this style.
Is it good enough: absolutely.
AI looks, feeling and tone get repetitive and distracting quick. Fixing that isn’t deceptive. It adds - even more - taste and steering away it away generic, and there are some great suggestions above to avoid that generic AI trap. That said… does look great, nice job.
Kinda strange take to me. Is it deceptive to edit something to improve it? Is there some self-deprecating element to this, "I used AI so I'll wear crappy copy as a badge of shame"?
No, it's not deceptive to edit something to improve it. If you edit something specifically to get rid of the "giveaways" for vibe coding, as mentioned above, that's more along the lines of deceptive.
> "I used AI so I'll wear crappy copy as a badge of shame"
If that's your view on stuff made with AI, maybe don't use AI to make stuff. Generating it and pretending it isn't generated doesn't mean it's not generated. That's more of a question of how you yourself value something made by AI, not necessarily how others perceive it. I did in no way imply that using AI for something is shameful, I merely said that specifically taking measures to cover it up is deceptive. I don't see how that's a strange take.
For dev projects I see AI just as a tool in the process of making something and was coming from the perspective that a lot of tools leave rough edges in the material that need sanding out to get the product perfect.
Sanding jaggies which came from any tool in the process is fair game.
I do agree it's at least distasteful to one-shot something and try and remove the telltale signs; but for me this particular project is not doing that at all. Look at the repo history, the author has no problem showing Claude writing commits.
For data sources, it would be cool to add Sentinel-1 since SAR sees through clouds, plus the Umbra and Capella open data archives for some high-res passes.
The first half of my second career (third maybe) in software was spent in satellite ground processing. My first true culminating project, so to speak, was when the portion of the US Maritime Domain Awareness meant for use by the Navy went to initial operational capability.
I checked out your site briefly. It's mildly amusing to note differences between that and a real system in use by military units. It'd be nice to be able to use a selector of some sort rather than typing in a location. The globe itself is most obvious, but a flat map works. If you're going to suggest arbitrary locations underneath the text entry box, you should probably suggest places you actually have imagery for. I don't care what looked at the hotel I'm in this morning and don't know coordinates for places off the top of my head, but I can find them on a map.
The real version of a system like this doesn't try to draw orbits and estimate what was overhead when. The imagery archives already have geolocation information you should be able to query directly instead of trying to project backwards like that. There are at least three downsides to the way you've done it. One, users typically don't care whether an image was collected from orbit. They just want to know what happened in location X at time Y. If you have aerial imagery or even ground imagery you can correlate with, augment, or entirely replace orbital imagery if it doesn't exist, that's just as useful. Two, satellites don't need to be overhead to collect on a location. The exact angles they can collect from are classified (and of course depend on atmospheric conditions and potentially other obstructions), but they don't have to be overhead. Three, not all satellites have fixed orbits. You can't always solve Kepler's equations to figure out where they've been from their current velocity. Some make changes to their own orbits.
Seeing things like this puts me in weird headspace. I think a lot of you get dazzled by how easy it's become to put together software. I want to reiterate the "not the bottleneck" phrase you probably keep hearing. Being able to do something like this relies on decades of past effort to put the sensor platforms in orbit in the place, create imagery from the data they collect, and archiving that imagery in a way that users can query. MDA itself was first announced by George Bush in his 2004 budget request. We went live with our first users in 2016. The effort involved was massive, but we had to invent new sensors, put them in orbit, test them in controlled, uncontested waters, figure out how to simulate data feeds that didn't yet exist in order to create software processing capabilities without being to run against real sensor feeds during the development, figure out how to chip, compress, and redact the massive files we produce so they can be disseminated over radio to tablets on ships in the middle of nowhere to field commanders that don't have top secret clearances with compartments for the collection method, or to partner nations that don't have US clearances at all.
The web frontend was such a small part of this that we never even retrained or hired JavaScript developers. The original non-browser frontend was a Java Swing desktop UI, so they just ported it to use the Google Web Toolkit, which transpiles Java to JavaScript, and that code was touched at a rate of maybe a few hundred lines a year, in contrast to every other part of the system involved in orchestrating the batch workload scheduling, order queueing, dissemination logic, everything that made it possible to process more data per hour than any other computing system on the planet, or at least they told us that in 2009. I'm sure that is no longer true thanks to LLM training. But man, we basically wrote our own Borg before Kubernetes existed (they of course migrated to Kubernetes eventually, which is why I'm in a hotel at KubeCon NA). We wrote our own filesystem because we couldn't get the IO throughput we needed from what the proprietary Unixes of the 90s offered. We wrote our own network protocols. We scrounged for replacement parts on eBay to service end of life SGI Altix boxes while the government squabbled over how to fund and who to award a contract to build a classified cloud, meanwhile canceling hardware refresh funding to already running operational systems. There are so many battle stories I really hope I can tell someday before I die when they're declassified. I'm sure the commercial providers you're querying have similar stories.
This is a nice effort, by the way. I don't want to come across like I'm downplaying it. I just want the next generation as they get awed at what they can do in days or even hours from a browser to understand you can only do that because of the supporting infrastructure and data that is out there. Bringing that data into existence took a lot of work. It was done by people. Without it, you're drawing ellipses on a screen. We had pocket computers that could do that with a few keystrokes 30 years ago.
This is cool! Is there an API to get the latest picture for a location? I'd like to set that as my browser new page background or integrate it into Home Assistant.
This is what I need: I like to see the latest developments in my area, but other web interfaces are hard to use. The only thing I wish for is the higher resolutions which I’d like to pay for with reasonable price
Relevant to your use case: Lucky Satellite Pic [1]
Super minimalist and Sentinel-2 imagery only, but simple to use and allows for easy download of full resolution satellite images without requiring any account creation.
This is very cool, and a bit creepy too, but mostly cool
Really liked the UX, even on a small phone screen it was easy to use. I enjoyed the big typing cursor, it was a unique surprise (almost 100% of text form inputs have either a very small vertical line cursor, or nothing at all - the big cursor feels refreshing)
From an user standpoint it's nice and works well. Nevertheless, using a boredom is not an aliby to use vibe coding only. Still, I like it. What times are we living in?
I love the idea. However the UI is a lot of bad choices made by LLMs. One thing I particularly dislike also when I work with LLMs is that they tend to make extremely redundant ad sometimes downright lazy UI choices:
- use centerdot to divide information where other means would be more readable
- many different fonts and shades for each bit of information, that *looks* ok but does not really relate well to that information
- redundancy (did I tell you it was clouded in 3 different places on the page?)
Essentially it feels like you asked a graphic designer to make a cool looking interface for a movie, that doesn't need to be too functional.
Trying to type in the location field is almost impossible. There is something way off in the way the block cursor reacts to character spacing compared to the text in that field.
Beautiful site! The comments section on a lot of HN posts is starting to remind me of posts for cars for sale on Bring a Trailer. On HN commenters seem to always need to qualify with "Even though it's AI...." and on a BaT post for a Porsche for sale commenters always need to qualify "When I got MY 911 GT3..."
for most places in the world ~10m/px is the best public image res, some lower (ex. rural areas) and some higher, am working on sourcing higher res for the areas where it's available tho
executes perfectly on my hincty android phone running a stripped down browser, everything is fast, search is educational, as in who knew there are so many closely named places, and quickly confirmed that a number of my spots of interest are ,yes, very often clouded over.
the contrast between this and the adver-servailence web is stark, stark to the point of making me wonder if the only counter to the deluge of slop is to make an app that filters and finds basic info and displays it like 2012, everything is still there, but the insane efforts to convert every single web event into a sale makes doing any online research or planning more trouble than it's worth
Can you please make your substantive points without swipes and fulmination, as the guidelines ask (https://news.ycombinator.com/newsguidelines.html)? It's quite a bummer to check in on a thread and see an aggressive attack sitting at the top, sucking all the oxygen out of what was otherwise a curious and fun discussion.
People are upvoting this submission because it's obviously a cool project. Yes, we ask commenters and submitters to hand-write the text they post to HN, and yes, Show HNs with genai writeups mostly aren't that interesting. Occasionally, however, one is cool, fun, and creative. HN is better off with these, even though few of us want to read generated text.
I actually emailed the submitter this morning, asking them to replace that text with something handwritten. We do that all the time, trying to salvage cool projects from getting flagged (either by the software or by users). In this case I never heard back—presumably the email didn't get through.
I decided that the text wasn't so bad and moved it to the top and re-upped the post anyway. Perhaps that was a mistake—but if I hadn't, this post would probably never have gotten attention. Given how enthusiastic the community response has been (and no, it is not a voting ring), that would have been a loss for the site.
Society is in a complicated process of figuring out the best ways to use AI. It's not clear what the conventions will end up being. Many people have many takes. Some of them break HN's rules—but often they haven't the foggiest idea there even is such a rule.
How about we don't assume deceptive intentions or seek to send others to hell during what is obviously an entirely-in-flux situation which will probably be going on for years?
Please delete my account. I want no association with this site if you're choosing to ignore your own rules to actively promote bot-generated spam over human content.
The nonsensical "I brute force checked it" paragraph is 100% Claude. Nobody writes like that, and no human ever includes this level of irrelevant reassuring detail.
To "deceive more people"? Take off the rage goggles and consider other alternatives like "they normally type like this. so when they had the LLM edit it that's how it remained", "because they normally type like this, its in the memory that's how 1st person text should be generated", or even just "because that's how they type, that's why they prompted it that way" before accusing blatant 1 mm deep attempts at deception as the driver.
But yes, the rule about LLM generated text should at least be abided by and preferably enforced regardless if the intent was malicious.
Amazing.
If someone puts a second-take, or a twentieth, it doesn’t matter what tools are used…
Human verification of human resonance.
That’s always the defining feature.
The giveaways:
* Instrument Sans (your logo)
* Highlighted period in your logo
* Highlighted italicized text in your header text (also in Instrument Sans)
* Short punchy AI-written sentences. Just merge that into one sentence! ("Enter any place on Earth to see every satellite that passed over it in the last week and the pictures they took there.")
* The flashing orange square in the upper right (every single vibe coded site has a circular or square version of the LIVE icon)
And then, just one more reason to switch from Instrument Sans: the massive kerning gap between "W" and "atched" in the logotype.
I'm being picky, of course. Nice job!
When the page is first loading and the logo is shown in the middle of the screen, the period is pulsating. I think the intention is that the colored period is to evoke the look of a glowing red LED light from a camera that is recording. It fits with the name “watched”.
Now whether it looks better putting it as a period at the end of the word in the logo like that, instead of as a solid circle ahead of the word, is perhaps debatable.
Is it good enough: absolutely.
AI looks, feeling and tone get repetitive and distracting quick. Fixing that isn’t deceptive. It adds - even more - taste and steering away it away generic, and there are some great suggestions above to avoid that generic AI trap. That said… does look great, nice job.
> "I used AI so I'll wear crappy copy as a badge of shame"
If that's your view on stuff made with AI, maybe don't use AI to make stuff. Generating it and pretending it isn't generated doesn't mean it's not generated. That's more of a question of how you yourself value something made by AI, not necessarily how others perceive it. I did in no way imply that using AI for something is shameful, I merely said that specifically taking measures to cover it up is deceptive. I don't see how that's a strange take.
For dev projects I see AI just as a tool in the process of making something and was coming from the perspective that a lot of tools leave rough edges in the material that need sanding out to get the product perfect.
Sanding jaggies which came from any tool in the process is fair game.
I do agree it's at least distasteful to one-shot something and try and remove the telltale signs; but for me this particular project is not doing that at all. Look at the repo history, the author has no problem showing Claude writing commits.
This site is really good one as you can see exact image taken by a satellite.
Well done.
For data sources, it would be cool to add Sentinel-1 since SAR sees through clouds, plus the Umbra and Capella open data archives for some high-res passes.
I checked out your site briefly. It's mildly amusing to note differences between that and a real system in use by military units. It'd be nice to be able to use a selector of some sort rather than typing in a location. The globe itself is most obvious, but a flat map works. If you're going to suggest arbitrary locations underneath the text entry box, you should probably suggest places you actually have imagery for. I don't care what looked at the hotel I'm in this morning and don't know coordinates for places off the top of my head, but I can find them on a map.
The real version of a system like this doesn't try to draw orbits and estimate what was overhead when. The imagery archives already have geolocation information you should be able to query directly instead of trying to project backwards like that. There are at least three downsides to the way you've done it. One, users typically don't care whether an image was collected from orbit. They just want to know what happened in location X at time Y. If you have aerial imagery or even ground imagery you can correlate with, augment, or entirely replace orbital imagery if it doesn't exist, that's just as useful. Two, satellites don't need to be overhead to collect on a location. The exact angles they can collect from are classified (and of course depend on atmospheric conditions and potentially other obstructions), but they don't have to be overhead. Three, not all satellites have fixed orbits. You can't always solve Kepler's equations to figure out where they've been from their current velocity. Some make changes to their own orbits.
Seeing things like this puts me in weird headspace. I think a lot of you get dazzled by how easy it's become to put together software. I want to reiterate the "not the bottleneck" phrase you probably keep hearing. Being able to do something like this relies on decades of past effort to put the sensor platforms in orbit in the place, create imagery from the data they collect, and archiving that imagery in a way that users can query. MDA itself was first announced by George Bush in his 2004 budget request. We went live with our first users in 2016. The effort involved was massive, but we had to invent new sensors, put them in orbit, test them in controlled, uncontested waters, figure out how to simulate data feeds that didn't yet exist in order to create software processing capabilities without being to run against real sensor feeds during the development, figure out how to chip, compress, and redact the massive files we produce so they can be disseminated over radio to tablets on ships in the middle of nowhere to field commanders that don't have top secret clearances with compartments for the collection method, or to partner nations that don't have US clearances at all.
The web frontend was such a small part of this that we never even retrained or hired JavaScript developers. The original non-browser frontend was a Java Swing desktop UI, so they just ported it to use the Google Web Toolkit, which transpiles Java to JavaScript, and that code was touched at a rate of maybe a few hundred lines a year, in contrast to every other part of the system involved in orchestrating the batch workload scheduling, order queueing, dissemination logic, everything that made it possible to process more data per hour than any other computing system on the planet, or at least they told us that in 2009. I'm sure that is no longer true thanks to LLM training. But man, we basically wrote our own Borg before Kubernetes existed (they of course migrated to Kubernetes eventually, which is why I'm in a hotel at KubeCon NA). We wrote our own filesystem because we couldn't get the IO throughput we needed from what the proprietary Unixes of the 90s offered. We wrote our own network protocols. We scrounged for replacement parts on eBay to service end of life SGI Altix boxes while the government squabbled over how to fund and who to award a contract to build a classified cloud, meanwhile canceling hardware refresh funding to already running operational systems. There are so many battle stories I really hope I can tell someday before I die when they're declassified. I'm sure the commercial providers you're querying have similar stories.
This is a nice effort, by the way. I don't want to come across like I'm downplaying it. I just want the next generation as they get awed at what they can do in days or even hours from a browser to understand you can only do that because of the supporting infrastructure and data that is out there. Bringing that data into existence took a lot of work. It was done by people. Without it, you're drawing ellipses on a screen. We had pocket computers that could do that with a few keystrokes 30 years ago.
Super minimalist and Sentinel-2 imagery only, but simple to use and allows for easy download of full resolution satellite images without requiring any account creation.
[1] https://luckysatellitepic.com/
Really liked the UX, even on a small phone screen it was easy to use. I enjoyed the big typing cursor, it was a unique surprise (almost 100% of text form inputs have either a very small vertical line cursor, or nothing at all - the big cursor feels refreshing)
Thank you
Your headline "Enter any place on Earth. See every satellite that passed over it in the last 7 days, and the actual pictures they took.".
Ok great ! But wait, what's this small-print ....
"pulls the real Sentinel-2 / Landsat / NASA images from those passes."
Oh. Yeah, about those "actual pictures they took" ...
Cue sad-face. :(
Bait-and-switch to low-res alternatives is probably not what most people want or are expecting from your headline boast.
This photo of the SF Bay this morning is a great representation of the fog that settles in over the area:
https://www.watched.earth/s/3zg2u78u
Very cool tool
I imagine, that there is someone with access to this sort of thing, with live imagery from any satellite they click on, in 8k.
edit: thanks for the work
_Most_ satellites. However, nothing associated with the NRO is in the public TLE catalog.
People are upvoting this submission because it's obviously a cool project. Yes, we ask commenters and submitters to hand-write the text they post to HN, and yes, Show HNs with genai writeups mostly aren't that interesting. Occasionally, however, one is cool, fun, and creative. HN is better off with these, even though few of us want to read generated text.
I actually emailed the submitter this morning, asking them to replace that text with something handwritten. We do that all the time, trying to salvage cool projects from getting flagged (either by the software or by users). In this case I never heard back—presumably the email didn't get through.
I decided that the text wasn't so bad and moved it to the top and re-upped the post anyway. Perhaps that was a mistake—but if I hadn't, this post would probably never have gotten attention. Given how enthusiastic the community response has been (and no, it is not a voting ring), that would have been a loss for the site.
Society is in a complicated process of figuring out the best ways to use AI. It's not clear what the conventions will end up being. Many people have many takes. Some of them break HN's rules—but often they haven't the foggiest idea there even is such a rule.
How about we don't assume deceptive intentions or seek to send others to hell during what is obviously an entirely-in-flux situation which will probably be going on for years?
Edit: please see https://news.ycombinator.com/item?id=50434775 as well because you're breaking the site guidelines in many places at the moment, and that is not cool.
But yes, the rule about LLM generated text should at least be abided by and preferably enforced regardless if the intent was malicious.