Unverified50% confidenceOpinionExact time
Asking the demonstrator to adjust their posture may seem like a low-cost compromise, but it quietly changes the distribution of the training data.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedBehavioral cloning suffers from the distribution shift problem: during training the policy only encounters the state distribution covered by expert trajectories, and during deployment the accumulation of small deviations causes the actual state distribution visited to gradually diverge from the training distribution.76% similarUnverifiedIf the quality of the guidance source is poor (large δ), extended training will amplify the bias, in which case the guidance weight should be reduced or external guidance should be stopped altogether.75% similarUnverifiedThe cost of targeted optimization is far lower than that of large-scale training improvements, and it is very difficult for users to detect, especially when users evaluate models using only a small number of the most popular prompts.75% similarUnverifiedTraditional draft models are trained on narrow distributions targeting a single model, causing acceptance rates to drop sharply when workloads change, and requiring the entire pre-training process to be redone each time a target model is replaced.75% similarVerifiedFine-tuning typically requires only one percent or less of the original training cost to achieve domain performance close to that of a dedicated model.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1014612API
curl https://kongchang.com/api/v1/knowledge/claims/1014612MCP
get_claim(id=1014612)