An analysis of large language models suggests that probing them with specific queries can reveal insights into their training data and timelines. By using historical quizzes and analyzing error rates, researchers can estimate when a model's knowledge cutoff occurred, correlating it with pre-training completion dates. This method indicates that Anthropic's Opus 4.7 and later models likely stem from a single training run that concluded around late December 2025, sharing a similar knowledge cutoff. AI
IMPACT Provides a method to estimate LLM training cutoffs and dataset mixtures, offering insights into model development.
RANK_REASON The item analyzes AI models using a novel probing technique to infer training data and timelines, which is a research-oriented activity. [lever_c_demoted from research: ic=1 ai=1.0]
Read on HN — claude cli stories →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
