Unverified50% confidenceBenchmarkExact time
RLDS achieved an improvement of +11.5 points on ScienceWorld (95% confidence interval [+9.8, +13.3]) and +9.8 points on FrozenLake ([+7.0, +12.8]).
1
Sources
50%
Confidence
Long-term
Relevance
9/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRLDS reduces wall-clock time per step on ScienceWorld by 10.9% compared to scalar GRPO, because long trajectories amortize the fixed overhead of reflection and scoring.63% similarUnverifiedThe coefficient of determination R² for this power-law fit is 0.732, which improves to 0.821 after adding interaction terms.58% similarVerifiedZhipu released GLM-5.3, claiming its coding capability improved by 50% compared to GLM-5.2.58% similarUnverifiedWhen Gemini (Flash) was asked to evaluate the other models, GLM (Zhipu 5.3) scored the highest at 89 points, DeepSeek scored 87 points, and Qianwen was judged unsuitable.57% similarUnverifiedAccording to OpenAI's early 2024 technical report, by annotating sycophantic behavior in RLHF data and assigning negative rewards, the unconditional acquiescence rate of subsequent GPT-4 iterations declined.55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/946122API
curl https://kongchang.com/api/v1/knowledge/claims/946122MCP
get_claim(id=946122)