OpenAI says it paused reinforcement learning on its latest models "intended for deployment" for the last two weeks. This excludes internal models not intended for deployment. The phrasing also implies that pause is over and training has now resumed, but they never say that directly. They do clearly say their largest planned RL training run remains on hold, but this language also excludes pre-training. This post is annoyingly slippery throughout, possibly for legal reasons, or with congressional…