ckrapu

Sorry! The presentation could be better. I'll work on the FAQ.

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:

https://www.linkedin.com/in/christopher-krapu/overlay/157690...

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.

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 you put the search on a hex, it looks up the vector for that hex and then performs a similarity search on all other vectors within the circle and shows the ones which are most similar in terms of land cover. The dependence on land cover / land use data is just because that was easy to get.

As other folks have pointed out here, raw satellite imagery is also a potential input source for this. I'm playing around with other sources and really want to integrate something like GeoVex (https://openreview.net/forum?id=7bvWopYY1H) into the embeddings as well.

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 Copernicus land cover data.

I found a map of the area around my family’s farm in 1894 super quick. Awesome site!

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 a very different result.

If you encode that language into a linear Gaussian model, you get KF.

Any commentary on what this means for the wider world of MCMC?

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.

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 sometimes associated them with the supernatural. They just lacked the tech and industrial background to use them.

We know that carbon is essential for life and if we ever engineer useful organisms to build structures or produce macroscale goods (as opposed to pharmaceuticals or food), then clearly carbon will be one of the main limiting factors, along with energy and water. It just feels like we will one day look back at this and think “why didn’t we just try to use it instead of burying it?”

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 during the training process?

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 towards AI which behaves similarly to people. After all, isn't prompt manipulation pretty similar to social engineering or a similar human-to-human exploit?

“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.

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 the streets and (2) how sternly it disapproved of this practice, violating instructions I had for sexual or vulgar content AND moral prescriptions.

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.

Yes! It is from DALLE-2.

I’m about to have my first child and I want to read to my baby. I only have so much patience for kids books, and I want to tell my kid about the real world. This site is a light UI over some prompt engineering to parse and rewrite Wikipedia articles. I’m open to any feedback! I’m not trying to market or monetize anything - I just want to get parents to read more to their kids.

A wise move for a people and nation well-familiar with the depredations of their neighbor.

https://en.m.wikipedia.org/wiki/Genocide_of_the_Ingrian_Finn...

Yeah, and the Finns hated it. My ancestors immigrated to the US to avoid being drafted from Finland into the imperial Russian army.

I disagree. It’s true that we are used to thinking of gravitation as a theory that was double checked with data, but it is also pretty clear that good physical arguments are strong priors. Then, with experimental data, we have a classic prior + data posterior update which does cast all of this as a statistical model.

I love reading about mcmc. Would you mind linking your thesis?

So how did all the fish and other animals die then?

I ride their cars every week in Chandler. For a joke, they sure have delivered a nice experience.

All snarkiness aside, I really enjoy riding in the cars because the driving strategy is interesting. Waymo routes appear to vastly prefer more right turns over fewer left ones (at a 3-1 ratio, by my estimates) and also is strategic about avoiding routes with large numbers of parked cars and pedestrians, even if the traffic flow is light.

This book shows up every year or so on HN and I will never stop upvoting it. My favorite textbook of all time, along with my favorite quote:

“Marginalization is a macho activity, appropriate only for those fluent in definite integration”

Nature hasn’t provided glyphosate, accessible nitrogen by the megaton, and millions of square miles of artificially drained soil. Vertical farming isn’t much more of an aberration than the average row crop operation in America.

The real downside: maybe I suck at writing and AI will forever describe my work as trite and insipid to all future beings.

Yes, Stokes flow all day! Also, most marshy places are flat, and most flat places are notoriously windy so the usefulness would be generally quite low.

However, my sense is that the real dealbreaker is that it’s impossible to find a suitably insulating material. A sphere with r=30m is already over a hectare of surface area.

The American interior generally does not benefit from the same moderating influence of major water bodies as Scandinavia does. The winters are quite different. That's why you can have frequent awfully cold conditions as far south as Oklahoma (and sometimes Texas) despite being as far south as Egypt or Jordan.

I sort of handwaved the drag away by claiming this really only makes sense at low velocities. That said, if there's a stiff headwind it might be very difficult to move forward.

One thing that occurred to me here is that perhaps you could shape the balloon like an airfoil to help increase the lift in the right kind of crosswind. Then you'd have to figure out if the increase in thermal losses were sufficiently small.

Yeah, the density number looks inappropriate. While looking more closely I found that my reference value of 1 kg lift per cubic meter was off; apparently Wikipedia says its closer to 4. That means the balloon size goes from radius of 20m. to 30m. and the fuel tank size goes from 100 kg to 400 kg.