Pigeons are (deservedly) having a bit of a swing back from the “rats with wings” disparagement era, but I’m so, so psyched for the “bats are good, actually” renaissance 🦇
Posts by Josh Hadro
Also, there's probably an essay to be written (if it hasn't been done already) comparing the desolate urban depictions of Edward Hopper with the Roku City screensaver that is an obviously desirable and idealized "walkable city" environment that somehow also has no people in it.
An AI generated image of "Koru Suburb," a riff on the "Roku City" screensaver on roku devices. it depicts a bleak, desolate American suburban wasteland. Foreground: a shuttered strip mall with flickering half-dead neon signs, a vast empty parking lot with faded yellow lines and a lone shopping cart. Midground: a dark closed big-box store, a fast food pylon sign with several letters missing, utility poles with tangled wires. Background skyline: a water tower, a cell tower, and the underside of a highway overpass.
Everyone wants to live in Roku City, but I woke up last night and couldn't get "Roku Suburb" out of my brain so now you have to witness an approximation of it
This was stunning when I saw it 12.5 years ago, and remains a stunning and elegant use of geo and viz tools today!
“Neat things on the Web” IMO had its heyday from like 2005-2012ish, but there are still gems appearing like this
Try this one weird trick: elect a mayor who doesn’t despise the actual work of being mayor
“We have all these texts and manuscripts from the 18th and 19th centuries, what if we could engage with those authors through their words and thoughts in a modern interface?!”
[spends an afternoon working on it]
“Oh, I see why we don’t do this”
No shade to this Victorian chabot effort, but one of the main reasons libraries themselves aren’t doing this left and right is that the general public would be stunned at how casually racist/misogynist/etc every LLM along these lines would be without *extensive* post-training/alignment
Teach your kids about subway signaling and interlockings at home before they learn it from someone else on the playground
1896 volume brewery and malster directory entry for Lytle-Stoppenbach Co. (Maltster) in Jefferson Junction, WI
Ha, it is there, but it was called Lytle-Stoppenbach Company at the time!
Most volumes have an entry for it, e.g. hadro.github.io/brewery-guid...
and hadro.github.io/brewery-guid...
What a fun connection!
The Data Explorer is part of a larger pipeline tool I'm working on that takes any IIIF manifest or collection as input, and works through a series of steps to output structured data like this Data Explorer view, a map view, and more. Stay tuned!
And hopefully more intuitive to most people than browsing through photographs of pages of a book from the 19th century. If a tool like this prompts people to engage with the original digital collection item even just a little bit more than they otherwise would, that seems like a win.
It isn't perfect by any means, there are still some OCR or alignment errors. It's not meant to be a 100% surrogate for the content of the directories. But it seems like a valuable modern access point to the richness of this kind of resource!
Each row in the data explorer corresponds to one entry in the directory, shows you the exact entry spot in the digitized text, and if you click through a IIIF viewer will take you to the exact right spot on the primary source page.
Screenshot of the Data Explorer tool applied to 12 brewery directories, published from 1896-1918
I'm fascinated by digitized historical directories, and I'm a big #IIIF booster, so naturally I spent some recent evenings building the tool I wanted to see in the world: a "Data Explorer" view of 12 brewery directory guides from 1896 to 1918: hadro.github.io/brewery-guid...
100%, no reason to trust him, just thought it was funny
Not sure if you’re referencing this or no, but if not: one of the galaxy brain CEOs did it on purpose! futurism.com/artificial-i...
Whoa!
Screenshot of a brewery guide data explorer I put together recently
If nothing else, it makes it possible to explore the dozens of breweries that a century ago were all within a few miles of where I live!
I can code, but coding for personal projects lost its allure for a long time. It may lose its appeal again once I get through the things I have in mind, but for the time being I'm producing things I've always wanted to see in the world and didn't have the energy to code through.
On the way to building a proof-of-concept data explorer for those brewery guides, I've ended up building a somewhat generic tool that can take just about any directory-like digitized volume, send it through a couple of steps, and spit out a CSV that is more than decent enough to start exploring
I created a github repo in 2016 to store the data from these amazing directories: digitalcollections.nypl.org/collections/...
But once I got through the OCR and PDF creation process for all the volumes, I lost steam.
With a couple of new tools, I got to where I wanted to get in a matter of hours.
A lot of great posts recently on this topic of "vibe coding as enabler," including the two linked here.
For me it's manifested as a way to scratch a digital project itch I've had for a decade, a data explorer for a set of Brewery directories from 1899 - 1918:
hadro.github.io/brewery-guid...
That story is only partly about the swan never seen in NYC before, it’s also about the avid birders who noticed and dedicate time to paying attention to the birds that are here and the birds that come here and that’s beautiful!
One of the genuinely beautiful things about local journalism taken seriously is that it gives you daily views into things that people love about this city that are different from the things you already love about the city
a few years ago I went out for a walk, and ended up walking from Fort Greene to the Verrazano Bridge on 4th Ave. Then walked all the way home up 5th. About 13 miles all told. A++, would totally advocate dropping everything some random days and just going on Extra Long Walks
Congrats! Really grateful to have worked with you. This one of my favorite things I ever wrote, based on an invitation from you! www.nypl.org/blog/2017/11...
Another idea: one model coding, another set of models criticising/critique-ing with how they would have done it or how it can be improved and feeding back. Get them to document/log to a file all actions and comments. Interesting for model interaction.
Yes, I like that as an added evaluation layer, and would be interesting to compare what an LLM chooses to do from the outset versus what the same model claims it would do in the critique mode. You'd have to structure the problem domain carefully, but I think this would be fascinating!