This story was originally published on HackerNoon at: https://hackernoon.com/understanding-scaling-law-through-data-science-lenses. Despite the immense promise of LMs, initial endeavors to apply pre-trained LMs to downstream tasks have encountered significant challenges. Check more stories related to data-science at: https://hackernoon.com/c/data-science. You can also check exclusive content about #data-science, #scaling-law, #artificial-intelligence, #tokenizing-raw-text, #discrete-tokens, #embedding-vector, #token-embeddings, #power-laws, and more. This story was written by: @tianchengxu. Learn more about this writer by checking @tianchengxu's about page, and for more stories, please visit hackernoon.com. Despite the immense promise of LMs as task-neutral foundation models, initial endeavors to apply pre-trained LMs to downstream tasks encountered significant challenges.