Unverified50% confidenceBenchmarkExact time
The AXIS benchmark paper found that method-aware episode conditioning on average outperforms ordinary behavior cloning, while long-horizon label fine-tuning can further improve the success rate; in real-world experiments, simulation pretraining raised the success rate from 9.6% to above 31.7%.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAfter conditioned pretraining, using constant labels during fine-tuning improves average success rate, and this holds whether labels are used or not at inference time.75% similarUnverifiedThe number of training epochs directly affects recognition performance; the more thorough the training, the higher the detection accuracy, and it is recommended to increase the number of training epochs to 20, 50, or even 10074% similarUnverifiedThe practical recommendation from quality-aware pretraining is: conditioning is the default first choice, especially when episode-level quality or trajectory annotations can be provided.73% similarUnverifiedExamining the model's action sequences during training one by one can determine whether high scores result from genuine capability improvement or from exploiting loopholes in the reward mechanism.73% similarUnverifiedIn controlled single-run pre-training comparison experiments, the model trained on the Edu-QuRating-curated mixture achieved higher overall observed accuracy across nine benchmarks than the FineWeb-Edu baseline model, with performance gains concentrated on specific tasks.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1029791API
curl https://kongchang.com/api/v1/knowledge/claims/1029791MCP
get_claim(id=1029791)