Interesting you can crash the new preview models by asking them to reduce a very large array of words into common smaller set of topics and providing the output

Interesting you can crash the new preview models by asking them to reduce a very large array of words into common smaller set of topics and providing the output as JSON object with the parent topic and each of its sub topics in an array… gpt-4 preview will just start repeating one of the sub topics forever or timeout