Unverified50% confidenceOpinionExact time
In 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.
1
Sources
50%
Confidence
Long-term
Relevance
9/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe so-called modality preference observed in previous research may be partly or largely attributable to positional effects, and the conclusion that "MLLMs trust text more" may need to be re-examined after controlling for the order variable.75% similarUnverifiedThe core assumption of the research is that the residual stream representations inside the model have already implicitly encoded information about whether a task will succeed.74% similarUnverifiedMLP层充当模型的长期事实记忆库，其键值对是预训练阶段固化在权重中的静态参数73% similarUnverifiedMLflow的评估框架采用了LLM-as-Judge范式，用一个强模型来评判另一个模型的输出73% similarUnverifiedWhen faced with an ML system design question, one should first clarify requirements and frame the problem, rather than jumping directly to model selection.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/946100API
curl https://kongchang.com/api/v1/knowledge/claims/946100MCP
get_claim(id=946100)