An oral history of how ChatGPT disrupted the field of NLP
This oral history from Quanta Magazine documents the disruption in the Natural Language Processing (NLP) field. When I first entered data science, NLP was a distinct specialisation. The colleagues I knew in that area were highly intelligent and research-focused, so reading about their collective experience with this shift is a useful reference.
The article, When ChatGPT Broke an Entire Field: An Oral History, uses interviews with researchers to tell the story of the NLP community’s journey from the introduction of the Transformer architecture to the release of ChatGPT.
Key moments in the timeline
The article outlines a clear progression of events that reshaped the field:
- 2017: The “Attention Is All You Need” paper introduces the Transformer architecture, though its full impact was not immediately recognised.
- 2018: Google’s open-source model, BERT, begins to shatter performance records on NLP tasks, leading to a “benchmark boom”.
- 2020: OpenAI releases GPT-3. Its scale and capability to perform complex tasks with simple prompts was a shock to many researchers, leading to what one described as a “career-existential crisis”.
- 2020-2022: Debates intensify around model “understanding” and the ethics of scale, exemplified by the influential “Stochastic Parrots” paper.
- November 2022: The public launch of ChatGPT acts as a “Chicxulub” moment, making many existing NLP research problems obsolete overnight.
- 2023-Present: The field reorients completely. NLP is largely absorbed into the broader AI landscape, with a focus on prompting, evaluation, and building foundational models.
Summary of the disruption
The release of GPT-3 in 2020 was the first major shock. As one professor noted, tasks that once formed the basis of a five-year PhD could suddenly be replicated in a month.
I’m trying out all the things that my recent Ph.D. students had done as their dissertations, and just realizing — oh my God, the thing that had taken a student five years? Seems like I could reproduce that in a month.
— Christopher Callison-Burch, University of Pennsylvania
The public launch of ChatGPT was the final, decisive event. It effectively ended entire categories of academic research.
In a day, a lot of the problems that a large percentage of researchers were working on — they just disappeared.
— Iz Beltagy, Allen Institute for AI
The aftermath has been a complete reorientation of the field, with researchers grappling with obsolete work, intense media attention, and a new reality driven by corporate-funded, large-scale AI.
The article effectively documents the human and professional consequences of rapid technological change. It confirms my early impression of the field being full of deep thinkers, and it is notable to see them navigate a revolution they helped create.
If you achieve so much, you also have to accept that the debates are going to be heated. How else could it be?
— Christopher Potts, Stanford University
