Bigger Isn’t Better: The Ethical and Scientific Vices of Extra-Large Datasets in Language Models
Bibliographic record
Abstract
The use of language models in Web applications and other areas of computing and business have grown significantly over the last five years. One reason for this growth is the improvement in performance of language models on a number of benchmarks — but a side effect of these advances has been the adoption of a “bigger is always better” paradigm when it comes to the size of training, testing, and challenge datasets. Drawing on previous criticisms of this paradigm as applied to large training datasets crawled from pre-existing text on the Web, we extend the critique to challenge datasets custom-created by crowdworkers. We present several sets of criticisms, where ethical and scientific issues in language model research reinforce each other: labour injustices in crowdwork, dataset quality and inscrutability, inequities in the research community, and centralized corporate control of the technology. We also present a new type of tool for researchers to use in examining large datasets when evaluating them for quality.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.365 | 0.648 |
| Meta-epidemiology (narrow) | 0.001 | 0.002 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.004 | 0.006 |
| Science and technology studies | 0.008 | 0.059 |
| Scholarly communication | 0.021 | 0.042 |
| Open science | 0.008 | 0.018 |
| Research integrity | 0.014 | 0.029 |
| Insufficient payload (model declined to judge) | 0.004 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".