Three CDS Researchers Featured in Major Story on LLMs’ Impact on NLP
Large language models have fundamentally transformed how computers understand human language, making years of specialized research obsolete almost overnight. Three current and former CDS researchers played key roles in documenting this seismic shift, earning prominent spots in a major new oral history about the dramatic evolution of the field of natural language processing (NLP).
Quanta Magazine’s “When ChatGPT Broke an Entire Field: An Oral History” featured CDS Associate Professor Tal Linzen, CDS Associate Professor Sam Bowman, and former CDS Research Scientist Julian Michael among 19 researchers interviewed about how large language models reshaped natural language processing.
“It used to be that you had to design a totally different system for each of the tasks that you worked on,” Linzen explained in an interview with CDS. The emergence of foundation models changed everything. Researchers no longer needed to engineer specific features for each language processing task; instead, they could train one large model and adapt it with minimal additional effort.
The transformation enabled entirely new research directions. Linzen now works on projects like training language models to adapt to individual users over multiple interactions, researching questions that he “would never have imagined it would even make sense to ask, because the models were so bad a few years ago.”
CDS was particularly well-positioned for these changes, partly due to the timing of its founding in 2013. “When the Center was started, we had so many junior people that for a time we had trouble manning the tenure committees,” Linzen explained. “But in retrospect, this was a strength. Everyone was at the stage of their career that they were most forward-looking, and were really paying attention to the zeitgeist.”
That period, said Linzen, from 2015 to 2020, was right when deep learning was starting to make an impact. So while many academic departments had to abandon established methods, CDS researchers were already aligned with where the field was heading.
In the Quanta article, Bowman, who has since joined Anthropic, described how the field “completely reoriented” after ChatGPT’s release. Michael, recently of Scale AI, recounted watching “entire types of research, not just specific projects” become obsolete.
The article positions these CDS researchers among the key voices documenting one of the most significant technological shifts in recent academic history — a transformation they not only witnessed but helped shape through their foundational work on modern language models.
By Stephen Thomas
Have feedback on our content? Help us improve our blog by completing our (super quick) survey.
