cosbynator

defined https://www.thisworddoesnotexist.com/w/Elonzi%20Scheme/eyJ3I...

Thanks! I just fixed it

Just pushed a change to update the blacklist (collapsing hyphens / spaces to single words); it'll never be perfect but hopefully slightly better!

What about REFACTOROLOGY https://www.thisworddoesnotexist.com/w/refactorology/eyJ3Ijo...

Haha, no but that's neat. The generation process is a sampling procedure so "anything goes" and it can stumble on unlikely samples from time to time. I should probably filter out recursive ones!

Good call, when you press "generate your own" it tries to detect these and displays a little warning but my blacklist is not complete. My model is bolted on top of GPT-2 which used the web as a training set which is probably why some of these are popping up.

Amazing! I found https://www.thiswaifudoesnotexist.net/ there a while back

Add to the list of reasons I wish I could change my HN name

My model is bolted onto GPT-2 which is trained on internet text; I think some of these examples are coming from there.

I'll give it a go! Seems pretty easy to replace the blacklist test with a WHOIS lookup.

Yup, press the "Generate your own button" on the site. If you want to go the other direction (definition to made-up word) you can hit up my twitter bot @robo_define: https://twitter.com/robo_define

Your noble spirit embiggens the smallest man

For slang, I'm training another modelusing urban dictionary as the datasource. I hope to release it someday but there is a lot of work to be done to clean the data. The articles are huge, user-generated and full of racism.

For example if it consumes lots of data that uses the pronoun "he" for doctor, it's much more likely to spit out "he" in a medical context.

Since the model lacks world knowledge for new words, sometimes it also ascribes words to a specific cultural origin / group that is completely wrong.

Both Alex and I are from Toronto (we are the engineers on Hyperlapse). I shot it on a little visit at the start of the summer and it became our go-to demo. That water taxi is particularly windy, bumpy and beautiful.

My post wasn't really aimed at a technical audience. It was more of "look what cool stuff we can do without looping in a human". I've given a slightly more technical introduction in my post on ranking universities: http://blog.argteam.com/coding/university-ranking-wikipedia/

It really is just PageRank. The source code is available on github https://github.com/cosbynator/WikiRank

Good point - this is a view of the English speaking world as an aggregate. I'm hoping to re-run this experiment on non-US Wikipedia sites. If there are any interesting differences I'll try to put together a new visualization :)

Cool! I didn't know about this. The score I compute is actually a PageRank, which takes into account the quality of inbound links rather than just the raw number. I'd be interested to try to see how well this correlates to page visits!

That's certainly a downside. I've been meaning to run this on non-english Wikipedias to see the similarities and differences. A subject for a future blog post :)

You definitely aren't wrong, but it is worth noting that the PageRank is calculated based on all wikipedia articles, and not just those of movies (so influence into other areas will hopefully bubble up). "Birth of a Nation" actually does show up quite prominently for the pre 1930s era.

I've actually computed a PageRank vector for all articles in Wikipedia. If you are interested, Coursera is about 0.0013% of Harvard which would put it very close to the bottom of the ranking.

I'm completely biased in this (I know the founders), but this is the real deal. I've recorded shaky, all over the place video handholding my iPhone on a bike and it came out like this: http://luma.io/v/B2-

It is pretty neat stuff.

There was a 5 page limit :) Also, there was nothing to prevent anyone from releasing source code, I'd imagine that a lot of projects are out there if you look carefully on GitHub