Tag
New research from Duke University indicates that AI models from various providers are becoming increasingly similar in their creative outputs over time, raising concerns about reduced diversity in AI-generated ideas.
This paper proposes Meta-Persona Anchoring and Filtered Temperature Scaling to reduce semantic convergence in LLMs, showing a drop in pairwise cosine similarity from ~0.85 to ~0.65 on the INFINITY-CHAT dataset with sub-20B open-weight models.
A factorial experiment reveals that persona detail does not monotonically increase LLM opinion diversity; interaction architectures explore non-overlapping opinion regions; low-cost interventions like temperature scaling have negligible effects.
This paper diagnoses the low diversity in LLM-generated stories, finding that 88.3% of sampled stories contain one of 11 common words (e.g., Elias, lighthouse) across models, and traces this homogeneity to post-training data and alignment rather than prevalence in pre-training data.
A Harvard research paper introduces Recoding-Decoding (RD), a novel decoding scheme that injects random priming phrases and diverting tokens to tap into an LLM's long-tail knowledge, significantly boosting output diversity without fine-tuning. The method maintains high relevance while mitigating response homogenization, with stronger models showing greater diversity gains.