In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond.
BIG-bench currently consists of 204 tasks, contributed by 442 authors across 132 institutions. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale.
0 kommentar(er)
