A new arXiv paper studies why aligned large language models tend to generate less varied text, linking the effect to probability concentration in their output distributions. The authors describe this as a shrinking generative horizon caused by alignment procedures. The work is a research analysis and does not announce a model or product.
A new arXiv study addresses a limitation of post-training LLM agents with trajectory-level outcome labels: such supervision offers little signal for keeping multiple distinct successful strategies that branch from the same decision state. The authors frame this as a successful trajectory diversity problem and introduce Direct Diversity Optimization, a method intended to preserve varied winning paths during preference-based post-training.
A newly updated arXiv paper presents a reinforcement learning approach that moves beyond the usual objective of a single deterministic, reward-maximizing policy by incorporating uncertainty over rewards to generate varied behaviour. The authors argue this diversity is essential for applications like fine-tuning language models and accelerating scientific discovery, where multiple distinct solutions are more useful than one optimized output. The v2 release is cross-listed in both the cs.AI and cs.LG categories.
OpenAI is accepting applications for the second round of its Scholars program, which supports people from underrepresented groups in AI research. Selected participants receive stipends and mentorship to study deep learning full-time for three months. Each scholar is expected to release an open-source project at the end of the program.
WHY IT MATTERS ↘Funded, full-time research stints are cheap relative to industry lab headcount, making this a low-cost way for OpenAI to widen the talent funnel and generate open-source outputs it can point to. It also pressures rival labs to match visible diversity and early-career pipelines, since stipend-level programs are a comparatively inexpensive recruiting and reputational lever.