Unverified50% confidenceOpinionExact time
Knowledge benchmarks (such as MMLU, TriviaQA, and NaturalQuestions) test a model's ability to memorize and recall factual knowledge, and knowledge scores show a strong positive correlation with parameter count.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2022年Meng等人的「知识神经元」研究表明,事实性知识主要存储在MLP层的键值对中,通过修改特定MLP层的参数可以精确编辑模型的知识75% similarUnverifiedMMLU的生物医学部分主要来源于教材和标准化考试题库,主要考察概念性知识记忆而非数据分析推理能力74% similarUnverifiedSAGA performs knowledge distillation at the external memory level, trading precision for abstraction, at the cost of introducing a dependency on the LLM's inductive capabilities.73% similarUnverifiedThe knowledge of large language models is encoded in a distributed manner across the parameter weight matrices, and the larger the number of parameters, the more long-tail knowledge can be retained.71% similarUnverifiedThe evaluation suite covers multiple dimensions including factual knowledge recall, logical and mathematical reasoning, scientific and statistical reasoning, general knowledge transfer, and ethics and multimodal tasks.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1137471API
curl https://kongchang.com/api/v1/knowledge/claims/1137471MCP
get_claim(id=1137471)