Unverified50% confidenceFactExact time
The architecture of current MLLMs typically trains language encoders and visual encoders separately before fusing them, causing models to struggle with deep cross-modal reasoning.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTemporal multimodal large language models (TS-MLLMs) leverage the reasoning capabilities of large language models to handle question-answering tasks, but struggle to genuinely capture dynamic temporal patterns, typically providing implicit reasoning that lacks explicit explanation of the logic behind conclusions.73% similarUnverifiedIn the instruction fine-tuning phase of many MLLMs, task descriptions often immediately follow the evidence, which may cause the model to learn an inductive bias of prioritizing the most recently read evidence before responding.71% similarUnverified在代码数据上训练的LLM不仅提升了编程能力，还显著增强了模型的逻辑推理能力，代码训练和逻辑推理能力之间存在正向迁移效应71% similarUnverifiedThe division-of-labor architecture of a planning model plus a coding model separates the two subtasks of intent understanding and symbolic precision, which can mitigate attention drift in long code generation by a single model69% similarUnverified提示词与模型版本存在深度绑定关系，不同版本模型在RLHF训练数据、指令微调策略上的差异会导致相同提示词产生不同输出，这种现象被称为「提示词漂移」69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/961214API
curl https://kongchang.com/api/v1/knowledge/claims/961214MCP
get_claim(id=961214)