Totally right to be wary as it’s a “generate plausible text” tool and not a “go find the right answer” tool. That said, it’s always fun to play and see where things are and see where the limits of things are!

Have you considered asking it to give cuisine=* suggestions based off the website for the restaurant? Having the language model do parsing of that form could be super helpful.

I am reminded of some work someone posted in the OSM Discord some time back about using AI text tools to crawl website= tags to scrape and generate opening hours, phone number, and social media into data tags.

As we work through this, I’m getting more and more interested in a general deduplication/simplification/deletion of common US import tags. Many objects have more than one synonym (or roughly synonym with the same value) which leads to maintainance issues (ex: merged nodes with different GNIS IDs, but they are in different synonymous keys.)

Some cleanup that would immediately help reduce confusion/headaches: The 6 ways to tag GNIS ID. The many ways to encode name data NHD:GNIS_Name, gnis:name, the GNIS tags that are just is_in:* tags of a different format, etc

I wonder if there’s precedent for merging out synonyms etc. Something to ask around about I suppose.

When I did the Indianapolis import I definitely hit issues where they disagreed. In those cases I had to find ANOTHER source of data to tie break, typically street side data. You can also see notes I left around Indianapolis like “what is the name of this street ‘foo’ or ‘bar’?”. Because I figured it would eventually get seen by a local and they can help be my arbiter.

UPDATE! I have completed the address conflation of NAD data into Indianapolis. I’ll need to post an update to the dataviz for it to be reflected but… one city down, many to go.

This is extremely helpful and always nice to have written down in an accessible location. Thanks for writing it up!

Thank you for sharing! Seems all too easy to miss the correct accent marks. Reminds me of this excellent diary about finding and fixing spelling in the “cuisine” tag which you may find interesting.

Oh absolutely. I was curious about “genus” and it’s only in the list because Latin was one of the “languages” that happened to survive my sorting. I highly doubt anyone is actually accidentally submitting Latin into the database. Casting a wide net means you’ll very often find false positives!

I’m super new to this kind of editing work. I’ll definitely review my edits. Keeping on task (correcting typos etc) and resisting the urge to fiddle (in many cases somewhat haphazardly) is a real issue. This is all extremely well thought out and I appreciate the review and feedback from y’all.

Just finished running this workflow for my local area. What a great idea. Some cuisine tags that aren’t in the list but made sense to me were: “gelato”, “pho”,”terikayi”. There’s hits for all those in tag info but would love to know if folks think there’s ways to improve those entries.