In this video, Zach dives into the differences between UDFs and built-in SQL functions in Spark Streaming, using a benchmark that processes 5 million random numbers. He explores how UDFs can be slower, with results showing about a 10% speed-up when using built-in functions, but this can vary based on caching and the complexity of operations. Zach also experiments with increasing the row count to 100 million and even 1 billion to see how performance changes. He encourages everyone to run similar benchmarks and observe the results for themselves, as there are nuances that can affect performance.