Yuda Song

@yus167

On the academia faculty job market. PhD @mldcmu. Previously @ucsd_cse @UcsdMathDept

following
followers

Unified training proposes one algorithm for every foundation model training stage

In posts, @yus167 expects simpler infrastructure and says static data and reinforcement learning could mix at any point during training.

9 Sources, 2h ago, first seen 2h ago

Models trained with Harness Learning are claimed to nearly double performance on challenging unseen tasks

A post describes training a model to revise agent harnesses using execution feedback. It says the model can then improve harnesses for unseen tasks while its weights stay fixed at test time.

5 Sources, 29d ago, updated 29d ago

Featured in

Unified training proposes one algorithm for every foundation model training stage

Models trained with Harness Learning are claimed to nearly double performance on challenging unseen tasks

Featured in

Unified training proposes one algorithm for every foundation model training stage

Models trained with Harness Learning are claimed to nearly double performance on challenging unseen tasks