Allen Institute for Artificial Intelligence
145 episodes listed below
PRESS PLAY
**The podcast is currently on hiatus. For more active NLP content, check out the Holistic Intelligence Podcast linked below.** Welcome to the NLP highlights podcast, where we invite researchers to talk about their work in various areas in natural language processing. All views expressed belong to the hosts/guests, and do not represent their employers.
Based on 4 episodes we listened to · September 2026
NLP Highlights makes AI research useful by tracing model results back to choices about data, training, access, and evaluation.
NLP Highlights treats research papers as working material, asking researchers to explain the choices behind language and multimodal systems. Bridger and Pradeep Dazeghi favor direct technical questions: how training data should be organized, what lower-resource teams can use,…
Language-model recipes get their knobs questioned before you have to swallow the finished product.
Press play if you want
Short reviews from the PlayNext desk, based on the episodes we processed.
A focused argument for treating data design as the real lever in LLM training.
The discussion makes a steady case that model progress depends on treating data as a design problem, not just a compute problem. It follows Sachin Gururangan from UW NLP to his next role at LLaMA, then into adaptive pre-training, where time, language family, language, and domain…
A clear, concept-led look at multimodal AI, organized around a four-part framework for imagining.
The conversation frames computer vision as more than recognition, pushing toward AI that can imagine as well as interpret. It also follows Dr.
An attentive look at using NLP to make science easier to study.
The episode treats science of science as a practical problem: how computational methods can help study the process of science, and where language tools can make that work easier to handle.
A clear look at why open-sourcing language models is treated as a research choice.
Open-sourcing language models is treated here as a research question, not a finished product. The conversation with AI2's ALMA leads stays on the choices that shape a model and its release: how to structure the training recipe, where instruction data belongs, whether human…
Reality, examined past the shortcut.
Global news, sharpened into consequence.
Tech power, counted in costs and constraints.
TED's stage, one speaker at a time.
Philosophy with technical consequences.
Bridger and Pradeep Dazeghi talk with Kyle Lowe from the Allen Institute for Artificial Intelligence about science of science, a field Kyle describes as computational sociology for studying how science progresses.
NLP Highlights interviews Sachin Gururangan, who is finishing a University of Washington PhD and joining Meta's LAMA team next week.
Dr.