News for hackers. Stuff that matters.

A comment on Reinforcement Pre-Training.

Re:Reinforcement Pre-Training by babelfish on Tuesday June 10, 2025 @12:21PM (#44238627)
So marginally better (and occasionally worse) performance for an order of magnitude larger training costs…?
[ Parent | on HN ]
Re:Reinforcement Pre-Training by watsonmusic on Tuesday June 10, 2025 @12:33PM (#44238758)
14b model performs comparably with 32b size. the improvement is huge
[ Parent | on HN ]
Re:Reinforcement Pre-Training by 85392_school on Tuesday June 10, 2025 @12:49PM (#44238901)
are we only comparing them in terms of text completion accuracy? does it also improve perf on benchmarks?
[ Parent | on HN ]