21 datasets found

A large dataset for measuring bias in language models, including nearly 600 descriptor terms across 13 different demographic axes.

The dataset used in this paper consists of three tasks: Multi-Genre Natural Language Inference (MNLI), Quora Question Pairs (QQP), and Stanford Sentiment Treebank (SST-2).

Larger language models have higher accu- racy on average, but are they better on ev- ery single instance (datapoint)?

The authors used a variety of datasets for question answering, including TriviaQA, Natural Questions, CountryQA, and Jeopardy questions.

The BIG-Bench Hard dataset is derived from the original BIG-Bench evaluation suite, focusing on tasks that pose challenges to existing language models.

The dataset used in the paper is not explicitly described, but it is mentioned that it is a preference dataset for language models.

The dataset used in this paper is a collection of regular expressions and grammars for constraining language models.

The dataset consists of subtitles of television shows provided by the Flemish public-service broadcaster VRT. The dataset is used to train language models of spoken Dutch.

The dataset used in the paper is LLaMA, a large language model.

The dataset used in the paper is a grammaticality judgment task featuring four linguistic phenomena: anaphora, center embedding, comparatives, and negative polarity constructions.

Finetuned language models are zero-shot learners

A self-recognition test for language models using model-generated security questions.

The dataset used in this paper for multilingual language models

Corpus pairs dataset for LABDet, a robust and language-agnostic bias probing method to quantify intrinsic bias in monolingual PLMs.

Minimal pairs dataset for LABDet, a robust and language-agnostic bias probing method to quantify intrinsic bias in monolingual PLMs.

Sentiment training dataset for LABDet, a robust and language-agnostic bias probing method to quantify intrinsic bias in monolingual PLMs.

The dataset used in the paper to simulate human behavior in various experiments, including the Ultimatum Game, Garden Path Sentences, Milgram Shock Experiment, and Wisdom of...

A language model that demonstrates capabilities in processing and generating human-like text.

The dataset used in this paper is a large language model, GPT-4, and its residual stream activations.

The dataset used in the paper is a collection of email and blog posts from 20 distinct authors, with a focus on few-shot alignment of large language models.