It was highly used mid 2024 and trended again (though not as highly around the start of 2025. LLMs have documented (intended) recency bias thus weighing more heavily than older data in the set. Perhaps you could elaborate on why you think it was fast given that known intended data ingestion bias?
