Roger Craig used Polar to Win Jeopardy.

Polar has been rapidly adopted by alpha-geeks to combat information overload and looks like we’ve found another convert - this time Roger Craig who used it to win $77k on Jeopardy.

He’s a machine learning, data science, and AI practitioner who combined his computer skills and an interest in quiz games into something extraordinarily unique:

He won more than half a million dollars on Jeopardy!, including a whopping $77,000 in a single day.

He did this not just with raw intelligence, but by creating a system to help him fashion himself into a game-show superstar.

Roger and I have similar backgrounds - both machine learning and big data and it looks like we arrived at the same general strategy for managing knowledge. Specifically, use tools like natural language parsing (NLP), clustering, and spaced repetition to force material to be stored in your brain long term with high retention.

How Roger uses Polar and Anki

The next tool I use is called Polarized. A lot of the information I consume is in technical published papers, which are in PDFs, and Polarized helps with that.

It takes in PDFs and web pages, and allows you to highlight and annotate them. You can also export them, so you can turn the annotations into Anki cards.

So basically every paper that is of interest to me, I save it into one big directory, and then Polarized goes in and ingests all of the papers and makes them available for me to read and annotate.

What Roger did here was basically take his background in ML and built a set of tools to manage his learning.

I basically did the same thing when I first started working on Polar but I realized that this tool needs to be democratized.

We’re going to be building in more cloud learning and big data capabilities in Polar in the next few months including support for full-text search, clustering, auto-tagging, etc.

The entire point of Polar is that we’re trying to democratize this technology and make it usable to the average person.

As an aside, it’s been really interesting to see who’s been using Polar. Lots of PhD students, VCs, researchers, academics, scientists, etc. Very exciting to see such amazingly smart people using the product.

Flashcards and Spaced Repetition

The key benefit Polar provides is a fully integrated environment for reading and managing flashcards. You can work directly with PDF and web content, annotate it, create flashcards from your annotations, and then sync the flashcards directly to Anki.

What’s nice about this is that not only do you have a high quality index of all your notes and highlights but this data is sync’d with Anki.

Machine Learning and Spaced Repetition

UPDATED I managed to schedule some time to talk to Roger about his use of Polar and Anki on Jeopardy.

We both come from similar backgrounds so it was nice to compare notes about NLP and machine learning and how it can apply to optimized learning.

He first competed in Jeopardy in 2010 and NLP (and machine learning) has come a long way since them.

Back then, NLP technology was available but just VERY expensive. Now it’s essentially commodity and dirt cheap.

The NLP frameworks in Python can be used for free meaning if you’re an alpha-geek you can build something pretty straight froward by putting together a few scripts.

What we’re trying to do in Polar is democratize a lot of this technology and make it “just work” right out of the box.

I talked to Roger about using NLP and machine to automatically build flashcards from textbooks. There’s still a lot of active research in this area with question and answer generation from text corpora.

There might be some low hanging fruit by parsing text the user is reading and finding the first named entities mentioned and building flashcards directly from those first sentences.

Sentence boundary detection and NLP can be used to take a textbook and allow the user to easily generate flashcards with one click.

This is something I’m working on improving in future versions of Polar but first we have to improve usability.

We’re going to keep comparing notes and try to sync up again in another few weeks. There’s definitely a lot of potential for using machine learning to improve the efficiency of education.