Thanks for the feedback! Will work on some updates for the hero backgrounds.
On your point about local code vs. API:
- We handle all the authentication, rate limiting, and API differences between providers, which becomes complex when working with multiple models. Our platform allows smart switching between different models depending on the message context.
- Our switching is optimized for performance. We provide detailed usage analytics and cost management across models so you can optimize which models to use where. This allows you to see what costs your users may be incurring / misbehaving.
We're building features that go beyond what's easily replicable locally, including some of the agent and evaluation capabilities you mentioned (rolling out over the next two weeks). OpenRouter is definitely doing great things in this space. We're taking a slightly different approach by focusing on bringing customization within the dashboard to allow for on-the-fly updates without pushing new code edits.
Appreciate the feedback and hopefully you get a chance to try out the tool!
This is correct. The brightness of the stars made them a bit difficult to identify by eye, I'm sure someone more familiar with sky charts could have done it. The constellations helped me align the sky on Stellarium (from their angles) and the north star helped me find the approximate latitude by using the angle from the horizon.
You don't actually need the focal length, it doesn't help accuracy that much but can help you line up the sky to the photo a bit quicker. Anyway, if we did, all that info was in the metadata anyway:
Camera: Nikon D5
F-stop: f/2.8
Exposure time: 5 sec.
Focal length: 28mm
Max aperture: 3
To be fair, a secretive aircraft would trigger thoughts about security in a hacker's mind. A camper van does not trigger any such thoughts. It may not be just clicks, with a camper van the whole thread might not even exist. I am not the original poster, so this is just my thought
So it's manipulative. Like a sensational headline that buries the mundanity at the end. Lead with "this is already publicly known" so people can make an informed choice about where to direct their attention.
If the point of the article/twitter thread was to reveal the location of the hanger, as if that was previously a secret, then yes perhaps the title is misleading.
But that's not the point of the article. The point of the article was to reveal the location of the hanger using the stars and other metadata, and to explain how that was done. The journey was the story, not the destination.
And that's where the disagreement is. It implied a secret by the nature of the object, enough so that the fact it wasn't needed to be clarified - but at the end. You only publicly reveal something once, and this already had been.
But then, those aren't his words and it's implicit so this is a grey area and not a huge deal. But I still think leaving the clarification to the end was deliberate.
The process can still be done using entirely the star pattern and no Google Maps. There wasn't any reason to exclude other information I could use, so I didn't.
You're right, it wasn't too difficult to guess the location without the stars. The base did happen to be right in-between the 34 and 35 degree lines (34.6 lat), and you can estimate pretty well with only one line
Little Rock AFB is a few tens of km north (34.9 deg lat) for what it's worth. There may be some others as well. The chances of getting a picture of the stars that clear in Arkansas is way lower than in the Mojave however...
Looking for higher resolution (3m), the only viable option really is Planet. Even at that, they're pretty iffy with their pricing models and distribution. I think they range around the $2 per square kilometre mark.
On your point about local code vs. API:
- We handle all the authentication, rate limiting, and API differences between providers, which becomes complex when working with multiple models. Our platform allows smart switching between different models depending on the message context.
- Our switching is optimized for performance. We provide detailed usage analytics and cost management across models so you can optimize which models to use where. This allows you to see what costs your users may be incurring / misbehaving.
We're building features that go beyond what's easily replicable locally, including some of the agent and evaluation capabilities you mentioned (rolling out over the next two weeks). OpenRouter is definitely doing great things in this space. We're taking a slightly different approach by focusing on bringing customization within the dashboard to allow for on-the-fly updates without pushing new code edits.
Appreciate the feedback and hopefully you get a chance to try out the tool!