Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models

  • 类型:arxiv
  • 标识:2206.04615
  • 链接:https://arxiv.org/abs/2206.04615
  • 主题:multimodal
  • 主分类:evaluation
  • 形态:benchmark
  • 被引:2608
  • 被引来源:Semantic Scholar
  • S2被引:2608
  • OpenAlex被引:551
  • 影响力被引:189
  • TLDR:Evaluation 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 finds that model performance and calibration both improve with scale, but are poor in absolute terms.
  • OpenAlex ID:W4281690148
  • OpenAlex DOI:10.48550/arxiv.2206.04615
  • DOI:10.48550/arxiv.2206.04615
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://arxiv.org/pdf/2206.04615
  • OpenAlex更新:2026-08-23
  • 待LLM分类:否
  • 成熟度:research
  • 场景:scaling、capability-evaluation、big-bench
  • 标题中文:Beyond the Imitation Game:语言模型能力的量化与外推
  • TLDR中文:在 BIG-bench 上对 OpenAI 的 GPT 模型、Google 内部稠密 Transformer 架构及 Switch 风格稀疏 Transformer 进行评估,模型规模跨越百万至千亿参数,结果显示性能与校准均随规模提升而改善,但绝对水平仍然欠佳。
  • 来源文件
  • [OpenAlex discover]
  • [S2 enrich]