We previously released an interactive Markov Chain Demo that allows
people to experiment with simple statistical models of language known as
Markov chains.  This is a good way to give K-12 students an inkling of
what large language models like ChatGPT are doing when they generate
text.

Today we're announcing an Activity Resource Guide for the demo that will
help teachers understand how to use it in the classroom.  It includes a
series of experiments you can perform to understand how statistical
language models are trained on text and are then able to generate new
text.  Because the demo supports bigram, trigram, and tetragram models,
it allows students to examine the tradeoff between model complexity,
text quality, and richness of variation in the generated text.

You can find the new guide on the AI4K12.org Activities page: https://ai4k12.org/activities/

Here's a direct link to the guide: https://ai4k12.org/wp-content/uploads/markov_chain-activity_guide.pdf

And here's a link to the demo itself: https://www.cs.cmu.edu/~dst/MarkovChainDemo/

Development of this guide was funded by a grant from NEOM Company.  Development of the
Markov Chain Demo was funded by National Science Foundation award IIS-2112633.

-- Dave Touretzky