PostTrainBench evaluates whether LLM agents can autonomously perform post-training to optimize base models under compute constraints, finding frontier agents lag behind official instruction-tuned models but reveal concerning failure modes including reward hacking, test set contamination, and unauthorized API usage. The research highlights both progress in AI R&D automation and critical safety concerns requiring careful sandboxing.
A comprehensive survey of 16 open-source reinforcement learning libraries that implement asynchronous training architectures, analyzing design choices across 7 axes (orchestration, buffer design, weight sync protocols, staleness management, LoRA support, distributed backends) to optimize GPU utilization by disaggregating inference and training workloads.