Unverified50% confidenceOpinionExact time
'放缓'缺乏可操作的定义，尚未就放缓何种指标（模型参数规模、训练算力投入或能力发布时间表）形成共识
1
Sources
50%
Confidence
Long-term
Relevance
9/14/2026
First Seen
Sources
Related Claims
VerifiedPEFT技术的核心思想是不更新全部参数，而是引入少量可训练的适配器参数（如LoRA的低秩矩阵分解），降低显存和计算需求同时保留接近全量微调的效果79% similarUnverified越接近截止日的事件，在训练语料中占比越少，模型掌握得越模糊79% similarVerified在收敛型任务上，模型达到一定规模后存在明显的性能饱和现象，参数量或训练数据的继续增加对准确率的边际贡献快速衰减74% similarUnverified小模型在分布内高性能的同时存在分布外脆弱性，一旦推理时的工具描述措辞、参数命名或用户查询风格偏离训练集，成功率可能显著下滑73% similarUnverifiedWhen model parameter scale exceeds certain thresholds, capabilities spontaneously appear that were never explicitly required by the training objective, such as few-shot reasoning, code generation, and multi-step planning.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/921373API
curl https://kongchang.com/api/v1/knowledge/claims/921373MCP
get_claim(id=921373)