GPT-3: A Breakthrough, but not Coming for Your Job

Summary

  • Recently, OpenAI released the largest language model to date, GPT-3, which is functionally similar to many of its predecessors and differentiated primarily by having 10x more parameters than the previous largest language model and being trained on a much bigger dataset.

  • This large quantitative difference allows a trained GPT-3 to achieve qualitative improvements over its predecessors; unlike the latter, the trained GPT-3 model can perform many different tasks without training specifically for those tasks.

  • It has met significant acclaim from both the press and technologists, who describe its myriad applications and also several key limitations.

  • Although GPT-3 represents remarkable progress, its limitations have gotten less attention, and are summarized here.

Continue Reading ->

Recent Articles: