Word embedding demo pages are nice but they are very limited. Instead students should be programming their own word analogy programs. Here's an example in Snap! that supports 20,000 words in 15 languages:

https://ecraft2learn.github.io/ai/snap/snap-no-logging.html?project=word%20analogy&noRun&editMode 

It is part of a whole guide around working with word embeddings: https://ecraft2learn.github.io/ai/AI-Teacher-Guide/chapter-5.html

I have some forthcoming publications on programming word embeddings in Snap! including this four page paper from the Snap! conference..

 

On Wed, 8 Jan 2020 at 13:53, Dave Touretzky <dst@cs.cmu.edu> wrote:
1. Here's an online demo based on word2vec that lets you do the
"king + (woman - man) = queen" experiment  yourself:

  http://turbomaze.github.io/word2vecjson/

While that example works, it doesn't always do so well on examples you
might expect it to get right.  But remember, it wasn't designed for
analogies.  Because it runs in the browser, this version uses the
smallest (1000 word) version of word2vec; it doesn't know some common
words such as "cow" or "toaster".  A larger version of the model might
do better.

2. Here's a delightfully disturbing fictional account of how deep fakes
upended someone's life:

  https://www.technologyreview.com/s/614942/deepfake-girlfriend-fiction-story/

AI isn't advanced enough yet to hold the kinds of conversations
described in this story, but given the remarkable tolerance people
show for chatbots (which are really stupid), that might not matter.

-- Dave
To unsubscribe from this list please go to http://www.simplelists.com/confirm.php?u=HgcrDKbrGvVfL7TMw7K0LcD05I6H3rlD