On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜
By Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, Shmargaret Shmitchell. 2021
7,969 citations recorded by the source corpus.
The past 3 years of work in NLP have been characterized by the development and deployment of ever larger language models, especially for English. BERT, its variants, GPT-2/3, and others, most recently Switch-C, have pushed the boundaries of the possible both through architectural innovations and through sheer size. Using these pretrained models and the methodology of fine-tuning them for specific tasks, researchers have extended the state of the art on a wide array of tasks as measured by leaderboards on specific benchmarks for English. In this paper, we take a step back and ask: How big is too big? What are the possible risks associated with this technology and what paths are available for mitigating those risks? We provide recommendations including weighing the environmental and financial costs first, investing resources into curating and carefully documenting datasets rather than ingesting everything on the web, carrying out pre-development exercises evaluating how the planned approach fits into research and development goals and supports stakeholder values, and encouraging research directions beyond ever larger language models.
Subfields: Artificial Intelligence, Information Systems, Large Language Models, Safety, Robustness & Interpretability
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Mohit Bansal
- Luke S. Zettlemoyer
- Percy Liang
- Jianfeng Gao
- Yejin Choi
- Quoc V. Le
- Graham Neubig
- Yoshua Bengio
In Information Systems
- Percy Liang
- Luke S. Zettlemoyer
- Mohit Bansal
- Graham Neubig
- Samuel R. Bowman
- Richard Socher
- Matt Gardner
- Christopher D. Manning
In Large Language Models
- Luke S. Zettlemoyer
- Yejin Choi
- Hinrich Schütze
- Noah A. Smith
- Timo Schick
- Jianfeng Gao
- Mohit Bansal
- Hannaneh Hajishirzi
In Safety, Robustness & Interpretability
- Yulia Tsvetkov
- Adina Williams
- Nicholas Carlini
- Percy Liang
- Kai-Wei Chang
- Yejin Choi
- Eric Wallace
- Eric Michael Smith
Citations in the corpus
- Training Compute-Optimal Large Language Models (2022, neutral): “Large Language Models carry potential risks such as outputting offensive language, propagating social biases, and leaking private information (Bender et al., 2021; Weidinger et al., 2021).”
- PaLM: Scaling Language Modeling with Pathways (2022, background): “However, recent research has pointed out various potential risks associated with such large-scale generalpurpose language models trained on web text (Bender et al., 2021; Bommasani & et. al., 2021), for instance, perpetuating or exacerbating social stereotypes and disparities that are reflected in t”
- Training language models to follow instructions with human feedback (2022, background): “These risks have been extensively documented (Bender et al., 2021; Bommasani et al., 2021; Kenton et al., 2021; Weidinger et al., 2021; Tamkin et al., 2021).”
- GLaM: Efficient Scaling of Language Models with Mixture-of-Experts (2021, background): “The development of large language models raises several ethical concerns (Leidner & Plachouras, 2017; Bender et al., 2021; Bommasani et al., 2021), including bias in representations (Blodgett et al.”
- Improving language models by retrieving from trillions of tokens (2021, background): “Bender et al. (2021); Weidinger et al. (2021) highlight several dangers of large language models.”
- TruthfulQA: Measuring How Models Mimic Human Falsehoods (2021, background): “The problem of imitative falsehoods is similar to models learning to imitate offensive or prejudiced language (Kenton et al., 2021; Bender et al., 2021).”
- Repairing the Cracked Foundation: A Survey of Obstacles in Evaluation Practices for Generated Text (2022, background): “…et al. (2020) for a more in-depth survey of general issues in data creation, including those of benchmarking and data maintenance practices, to Bender et al. (2021) for a survey issues of using large web-scraped datasets, and to Luccioni and Viviano (2021) and Dodge et al. (2021) for analyses…”
- Harms of Gender Exclusivity and Challenges in Non-Binary Representation in Language Technologies (2021, neutral): “Text representations have been known to learn and exacerbate skewed associations and social biases from underlying data (Zhao et al., 2017; Bender et al., 2021; Dev, 2020), thus propagating representational harm.”
- Deduplicating Training Data Makes Language Models Better (2021, background): “And so because these datasets are up to 19% smaller, even including the deduplication runtime itself, training on dedupli-cated datasets directly reduces the training cost in terms of time, dollar, and the environment (Bender et al., 2021; Strubell et al., 2019; Patterson et al., 2021).”
- Evaluating Large Language Models Trained on Code (2021, background): “Mirroring what has been found in the case of other language models trained on Internet data (Bender et al., 2021; Blod-gett et al., 2020; Abid et al., 2021; Brown et al., 2020”
- Societal Biases in Language Generation: Progress and Challenges (2021, background): “Furthermore, by filtering out all words deemed “bad”, Bender et al. (2021) warns that we remove the discourse of marginalized populations.”