Unverified50% confidenceFactExact time
In LLM prompt-based generation, mode collapse manifests as the model tending to repeatedly return to the 'safe' answer with the highest probability in the training distribution.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn LLM post-training, the model directly spews intermediate reasoning steps that should have been hidden to the user, or during RL reward computation, the reward model that should only evaluate the final answer sees the reasoning process, thereby distorting the optimization direction.78% similarUnverified提示注入利用的是模型无法可靠区分「数据」与「指令」的天然缺陷，这一缺陷根植于LLM的训练机制本身77% similarUnverifiedIf LLM judges that are blind to omissions are widely used for scoring, benchmarking, or guiding model training (e.g., RLHF, RLAIF), the entire optimization loop may be misled, causing models to learn to generate content that appears complete but is actually incomplete.76% similarUnverifiedWhen faced with questions outside the training distribution (out-of-distribution queries), models may degrade into guessing based on surface-level patterns.76% similarUnverifiedIf leading models are used to generate training data, they may systematically pass their own biases to the next generation, a risk known as 'model collapse'.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1129087API
curl https://kongchang.com/api/v1/knowledge/claims/1129087MCP
get_claim(id=1129087)