ImproveAnyTask: An Autonomous Post-Training Harness for Iterative Model Self-Improvement
Adapting general-purpose large language models to specific tasks requires substantial human effort in designing data and training strategies. Sustaining improvement is especially challenging because model updates change the error distribution, requiring strategies to be continually refined. We introduce ImproveAnyTask,...
Xing-Bo Yao, Xiao-Man Wang, Zheng-Wu Lei et al.
· 0 citations