Scale AI New Grad ML Research Probability Phone Screen

Their earlier project used a sequence-to-sequence model to expand mathematical expressions. They reported accuracy above 98%.

The interview instructions asked for NumPy, SciPy, and Matplotlib. During the screen, they worked on Adjacent Order-Statistic Gap Distributions. They generated ten random values, plotted the fifth and sixth ordered values, and studied the distribution of the difference between them. Browser research was allowed while sharing the screen.

This was the only probability and plotting task. There were no ML or deep-learning fundamentals questions. They were rejected after the round.