Sample efficiency of training AI models has been increasing over the years
— And since “Finetuned Language Models Are Zero-Shot Learners”, technically that sample efficiency number crossed zero a while back, necessitating a refinement in how we calculate smaple efficiency, perhaps measuring the length of the zero-shot prompt instructions themselves, i.e. measuring sample efficiency via the “a word to the wise is sufficient” principle.
— And since “Finetuned Language Models Are Zero-Shot Learners”, technically that sample efficiency number crossed zero a while back, necessitating a refinement in how we calculate smaple efficiency, perhaps measuring the length of the zero-shot prompt instructions themselves, i.e. measuring sample efficiency via the “a word to the wise is sufficient” principle.
🔥2❤1👏1
Tool Documentation Enables Zero-Shot Tool-Usage with Large Language Models
“Our work provides an alternative to demonstrations: tool documentation. We advocate the use of tool documentation—descriptions for the individual tool usage—over demonstrations. We substantiate our claim through three main empirical findings on 6 tasks across both vision and language modalities.”
“we show that tool documentation is significantly more valuable than demonstrations, with zero-shot documentation significantly outperforming few-shot without documentation.”
From the author:
“Our new paper finds something quite neat: We easily scale up how many tools LLMs can use to over 200 tools (APIs, models, python functions, etc.) ...without any training, without a single tool-use demonstration!!”
Arxiv Link
“Our work provides an alternative to demonstrations: tool documentation. We advocate the use of tool documentation—descriptions for the individual tool usage—over demonstrations. We substantiate our claim through three main empirical findings on 6 tasks across both vision and language modalities.”
“we show that tool documentation is significantly more valuable than demonstrations, with zero-shot documentation significantly outperforming few-shot without documentation.”
From the author:
“Our new paper finds something quite neat: We easily scale up how many tools LLMs can use to over 200 tools (APIs, models, python functions, etc.) ...without any training, without a single tool-use demonstration!!”
Arxiv Link
❤3👍1🔥1👏1