Practice online against Fritz, starting from the Foisor - Krush position in the article. Click the Move Now button to switch colors.
https://fritz.chessbase.com/?fen=1Q6/8/8/q7/8/4k3/1KP5/8%20w%20%E2%80%93%20%E2%8...

But how valuable is it to play out a drawn position against an engine? In a winning tablebase position, I believe most engines will select the candidate with the smallest Depth to Conversion (DTC), conversely when losing it chooses the largest DTC. But in a drawn tablebase position, how should/does an engine select between drawing candidates? Interesting to me would be a learning AI (sic) for human vs computer training, which learns the candidates against which humans fare worst. Second generation AI would learn recurring patterns behind those candidates, for predicting the results when it hasn't been able to gather statistics yet. Current tablebases are optimized for disk space. My idea would require a modified database to capture the statistics, and it would be online in order to maximize the learning opportunities.