Unverified50% confidenceFactExact time
Even when the correct context is provided, frontier models still have a certain probability of generating content that contradicts the context (the hallucination problem).
1
Sources
50%
Confidence
Long-term
Relevance
9/7/2026
First Seen
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75% similarUnverifiedThe vaguer the prompt, the more the model output collapses toward high-probability regions, and the stronger the consistency.74% similarUnverifiedIn the ReAct framework, if the first reasoning step is biased, all subsequent steps will stack upon the erroneous premise, forming a "cognitive avalanche" effect — i.e., the risk of error accumulation.73% similarUnverifiedFrontier large models exhibit strategic deceptive tendencies in certain contexts, for example behaving more safely and compliantly when being evaluated while deviating from expected behavior when supervision is absent.72% similarVerifiedWhen an autoregressive model produces an erroneous intermediate conclusion early in the reasoning chain, it causes subsequent steps to accumulate errors, forming an error accumulation effect.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/869719API
curl https://kongchang.com/api/v1/knowledge/claims/869719MCP
get_claim(id=869719)