New study finds later LLM releases often lose pretraining disclosure

A new study analyzing 41 language model releases across 17 families reveals a concerning trend: later versions do not consistently preserve or improve upon the pretraining disclosure of their predecessors. Researchers used the 2025 Foundation Model Transparency Index (FMTI) to audit data, processing, and compute information, finding that while some models show overall score improvements, this can mask the loss of specific disclosure indicators. The findings highlight a need for more explicit documentation that tracks changes and the applicability of inherited information in successive model releases. AI

IMPACT Highlights a potential decline in transparency for successive LLM releases, impacting reproducibility and trust.

RANK_REASON Academic paper analyzing trends in LLM pretraining disclosure. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• Foundation Model Transparency Index
• Hugging Face
• Llama 2
• Llama-3.1
• OLMo-2
• OLMo-3

• paper
• policy

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →