Unverified50% confidenceBenchmarkExact time
Under the earlier idealized actuator model, the policy successfully recovered and maintained standing in approximately 97% of randomized evaluation episodes.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/17/2026
First Seen
Valid until: 10/31/2026
Sources
Training Whole-Body Recovery Policies for Humanoid Robots: From Reward Design to DC Motor Modeling
redditr/reinforcementlearning10/17/2026
Related Entities
Related Claims
UnverifiedAfter changing the actuator dynamics, the old 97% success rate data is no longer directly comparable, and the current policy is being retrained and evaluated under a more difficult model.69% similarUnverifiedRetry rate refers to the proportion of cases where the model's first output is not adopted and the developer needs to ask again or adjust the prompt and try again; it is ubiquitous in real workflows because the model's one-shot success rate is far below the ideal value assumed by benchmarks.67% similarUnverifiedDistilBERT retains approximately 97% of the original model's performance on the GLUE benchmark, with inference speed improved by approximately 60% and model size reduced by 40%.66% similarUnverifiedAn ultimate deception technique exploiting refusal reversal—constructing a narrative in which the model first refuses and then outputs the chain of thought—can raise the attack success rate to approximately 99%.66% similarUnverifiedRanked by question difficulty, persistence rate was 91.4% on the 60 questions that all models answered correctly, but only 73.5% on the 52 questions that at least one model got wrong, indicating that sycophancy tracks uncertainty.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1052201API
curl https://kongchang.com/api/v1/knowledge/claims/1052201MCP
get_claim(id=1052201)