OpenAI has significantly strengthened safeguards against extracting internal reasoning from GPT-6.2 Astra and GPT-6.2 Sol. In adaptive red-teaming tests, the success rate for extracting protected reasoning fell from 54…
新模型会重置能力与性价比的前沿。每次发布改变「每美元能做什么」,团队就要重新评估该基于哪个模型构建。
摘要仅供参考,请点击来源链接查看全文。演示条目为示意。