Unverified50% confidenceBenchmarkExact time
Most frontier models exhibit extremely high variance when running the same task repeatedly, potentially producing drastically different results.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/21/2026
First Seen
Valid until: 12/20/2026
Sources
Related Entities
Related Claims
Unverified大模型在逆向拆解任务中存在幻觉风险，更容易在常见框架标准用法上准确，而在非标准实现或项目特有约定上产生偏差74% similarUnverifiedAt equivalent parameter scales, diffusion language models typically still lag behind mature autoregressive models in perplexity and downstream task performance.73% similarUnverifiedThe same model paired with different harnesses can exhibit vastly different success rates on real coding tasks.73% similarUnverified当模型同时优化任务准确性和置信度校准时，两个梯度方向可能相互干扰，涉及灾难性遗忘和目标冲突问题73% similarUnverified不同任务类型的涌现阈值存在巨大差异，某些推理能力在模型规模跨越临界点时突然涌现，而标准代码补全呈现平滑饱和曲线73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/939569API
curl https://kongchang.com/api/v1/knowledge/claims/939569MCP
get_claim(id=939569)