 |
ML Newsnew | past | best | rss | submit | about |
| Implicit Bias in Large Language Models with Concept Learning Dataset (arxiv.org) | |
| 1 point by PaulHoule 346 days ago | hide | past | pdf | discuss on HN | |
|
In plain words: A set of small puzzle tasks teaches a model a made-up rule from examples, then tests it to reveal hidden habits. This exposed a tendency to assume statements stay true when a word is swapped for a broader one, which plain questioning barely showed. Abstract · Uncovering Implicit Bias in Large Language Models with Concept Learning DatasetWe introduce a dataset of concept learning tasks that helps uncover implicit biases in large language models. Using in-context concept learning experiments, we found that language models may have a bias toward upward monotonicity in quantifiers; such bias is less apparent when the model is tested by direct prompting without concept learning components. This demonstrates that in-context concept learning can be an effective way to discover hidden biases in language models. | |
| add comment on HN |