ckrapu

11 ms·

121.
by ckrapu 3y ago
Cool work and I hope to be able to use it some day. Most importantly though, does anyone know how they made the animation with the data sources? I feel like that came from something lightweight and convenient and I'd like to know what
122.
by ckrapu 3y ago
Yes, and my only commentary here is that the threshold for academic decision science is a lot higher than the one I have for managing my day-to-day life.
123.
by ckrapu 3y ago
These are excellent but I think #8 (statistics of small numbers) is one that, with some nuance, you can get more personal mileage out of if you're willing to tolerate being wrong some of the time. For example, I've noticed that of
124.
by ckrapu 3y ago
Potentially, yes! The scenario I am imagining is that context A as context for a query yields results 1, 2, and 3. Sometimes, finding neighbors of 1 (ie not necessarily in the top K w.r.t A) instead of going to results 4, 5, 6 might be bett
125.
by ckrapu 3y ago
I’m really interested in content explaining how to navigate graphs of embedded items for Q/A. Any resources on how to do this or arguments for why it’s a bad approach? For example, if my top K docs aren’t answering the question but eac
126.
by ckrapu 3y ago
An assumption here is that attractive men both pair with women at a higher rate and also have more children. I don't see any reason that's the case - the very largest families I am aware of in my extended social circle belong to m
127.
by ckrapu 3y ago
There’s a few companies making inroads here. Upcodes is one, IIRC. I’ve seen some work from Stantec that is perhaps more conventional ML and less LLMs. There’s also civilis.ai but I’m not sure how serious they are. I’m broadly interested in
128.
by ckrapu 3y ago
This is a bit of an embarrassment for that country in that they aren’t trusting their own people to actually build cars properly and have to build up from just reassembling cars. I do not expect them to hit their target of fully manufacturi
129.
by ckrapu 3y ago
I think I can see what's going on here - I used a shapefile with the boundaries of the world's countries (and their coastlines) which had some geometric simplification applied to it. This file was used to mask out any water (for w
130.
by ckrapu 3y ago
You pretty much hit the nail on the head. The application areas you mentioned are the same as the ones that I had in mind when developing this.
131.
by ckrapu 3y ago
Yup, and while the similarity search is perhaps the most visually appealing way to work with it, the real use (in my opinion) is in providing generic sets of geospatial features which are reusable across applications. I've built out ve
132.
by ckrapu 3y ago
Cool! Glad you got to see it working and that presentation was a nice reason to make sure everything was cleaned up.
133.
by ckrapu 3y ago
Thanks! And you are giving it too much credit here - it's just trained on one-hot encoded land cover (24 classes) from Copernicus. Using imagery directly would be # 2 on my list of to-dos after including elevation in the input data.
134.
by ckrapu 3y ago
I intentionally avoided using lots of ocean areas - this way I cut down the number of required sites for inference from ~100 million (at resolution 7 in the H3 system) to around 25 million.
135.
by ckrapu 3y ago
Sorry! The presentation could be better. I'll work on the FAQ.
136.
by ckrapu 3y ago
The embeddings were obtained using a CNN triplet loss model (~10M parameters) on the Copernicus land cover data. I haven't used DEM data yet but I have done generative modeling on DEMs in other work and would like to do that too: htt
137.
by ckrapu 3y ago
The original idea came from something I saw at work - we needed a way to build generic feature sets representing something about real estate, but beyond the data we had on prices, floors, and other house-specific details.
138.
by ckrapu 3y ago
Sure! The basic idea is that each hexagon is a discrete unit of space for which I obtain a vector embedding. This vector is supposed to represent a sort of data-based summary of that location, obtained in this case using deep learning. When
139.
Show HN: My demo for vector embeddings for the Earth's surface(louisquissetlabs.com)
106 points by ckrapu 3y ago|38 comments
140.
by ckrapu 3y ago
I've had to build out some version of a geospatial vector embedding / latent variable dataset for at least 4 separate projects now. Come see the viewer I've built on top of it! The embeddings come from globally available Cope
141.
by ckrapu 3y ago
I found a map of the area around my family’s farm in 1894 super quick. Awesome site!
142.
by ckrapu 3y ago
It would if you had some past history to work with. Let’s say that you measured the same object awhile ago and you want to incorporate that prior information in a principled way, while also letting the new measurements dominate if they give
143.
by ckrapu 3y ago
Any commentary on what this means for the wider world of MCMC?
144.
by ckrapu 3y ago
If the story of Earth were a novel and the current chapter introduced both global warming and the advent of carbon-based manufacturing, it would almost feel too contrived of a setup.
145.
by ckrapu 3y ago
When I think about carbon extraction and storage, I get the same feeling as when I read about the ancients discovering oil seeps and surface deposits of hydrocarbons. They knew that they could be harnessed for keeping fires lit and even som
146.
by ckrapu 3y ago
I think "childlike" comes close but misses the mark a bit. It's not that the LLMs are necessarily unintelligent or inexperienced - they're just too trusting, by design. Is there work on hardening LLMs against bad actors
147.
by ckrapu 3y ago
I love everything about how prompt manipulation is turning out to be a major weakness of exposing LLMs to users. It feels like this vulnerability reflects how LLMs are indeed a huge step not just towards machine intelligence but also towar
148.
by ckrapu 3y ago
“Partial support provided by Koch foundation” Color me a little unsurprised that an article downplaying the value of elite schools’ value-add lists such a funding source.
149.
by ckrapu 4y ago
Oh man, I’ve tried so hard to figure out how to prompt engineer the subtler things out. One of the hardest topics I came across was Mardi Gras. No matter what I did, GPT always wanted to (1) comment on how women like to flash while drunk in
150.
by ckrapu 4y ago
That’s not too much of a lift aside from working around the maximum tokens limit on the GPT API. Wikipedia articles are long enough to eat up the total token count just with the prompt.

More ›