Created a proofreading pipeline that masks manuscripts locally before sending to external LLMs
To have unpublished article manuscripts proofread by external AI, a mechanism was created to mask them locally before sending and restore them locally upon return. SB Intuitions' Sarashina3 was used, specifically the sarashina3-mini model.
A mechanism was created to mask unpublished article manuscripts locally before sending them to external AI for proofreading and to restore them locally once they are returned. SB Intuitions' Sarashina3 was used, specifically the sarashina3-mini model.
The reason for not sending the manuscripts as they were lies in the API terms of use. The terms stated that transmitted inputs and generated results could be used for service provision and improvement, which could be interpreted as including training. Because the unpublished manuscripts intended for proofreading contain reference URLs and strings such as identifiers, this mechanism was developed to avoid passing that information directly to an external source.
This item is a summary. The full, checkable version is at the original source:
Read the original at Zenn (LLM) ↗