Unverified50% confidenceSolutionExact time
LongZero adopts a two-stage "compress first, then expand" design: first the model synthesizes questions, answers, and supporting evidence in a compact structured form, then the evidence is expanded into multi-document long contexts.
1
Sources
50%
Confidence
Long-term
Relevance
10/22/2026
First Seen
Sources
Related Entities
Related Claims
Unverified结构化文档生成是一个多约束优化问题，需要模型具备long-horizon reasoning（长期推理）能力72% similarUnverifiedCommon experimental embellishment techniques in papers include: tuning hyperparameters for one's own method while keeping outdated parameters for baselines, writing custom efficient Triton kernels for one's own method while using vanilla implementations for baselines, and placing the question before the context to artificially improve compression results.68% similarUnverifiedThe MAX level essentially extends the model's chain of thought within a single inference, allowing the model to perform more internal iterations before output, rather than invoking more parallel resources.68% similarUnverifiedLongMemEval primarily tests a model's ability to locate and correctly use specific information in very long conversation histories, covering both single-hop and multi-hop question-answering scenarios66% similarUnverifiedThe model uses stacked multi-kernel convolutions to combine sentence-level embeddings into inter-sentence representations, and employs bidirectional LSTM to model long-range dependencies.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1084156API
curl https://kongchang.com/api/v1/knowledge/claims/1084156MCP
get_claim(id=1084156)