Unverified50% confidenceBenchmarkExact time

For adapters trained only on behavior (without storing document content), the trained writer recovers only 1-2% of instructions.

1

Sources

50%

Confidence

Medium-term (~90 days)

Relevance

11/9/2026

First Seen

Valid until: 2/7/2027

Sources

Can the Skill Documents Hidden in LoRA Weights Be Read Back Out?

Related Entities

Executing Writermodel

Related Claims

Unverified指令微调不需要从头训练，通常仅需基础训练计算量的1%以下即可显著改变模型行为模式74% similar UnverifiedIn actual products, user input is not directly used as training data for instruction fine-tuning; behavior calibration is mainly accomplished through reinforcement learning.70% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/1149722

API

curl https://kongchang.com/api/v1/knowledge/claims/1149722

MCP

get_claim(id=1149722)