User asked for research on perplexity about the latest daily AI news (Claude, OpenAI). Collected tasks: – Summarize what ‘perplexity’ measures in language models. – Discuss limitations of perplexity for news/real-time evaluation. – Propose additional metrics and methods for assessing model behavior on daily AI news (hallucination rate, factuality, model drift, robustness, safety signals). – Suggest experimental setups and datasets to measure these metrics in practice (time-stamped news corpus, human eval, claim verification, calibration tests). – Provide quick tooling pointers (fact-check APIs, embedding search, model eval frameworks). – Deliver in concise bullet points and recommended next steps.
Output type: research brief for tech-savvy audience.