我以两种方式测试了 Sonnet 4.6,其中一种变得个人化
📄 中文摘要
Sonnet 4.6 发布,具备 100 万个令牌的上下文窗口,改进了编码功能,价格与 4.5 相同。作者进行了两项实验,第一项实验将六个月的未发表草稿(约 24,000 字)加载到模型中,并询问:“我真正想表达什么?”模型意外地识别出潜在的焦虑情绪,揭示了在所有写作中存在的寻找赋权与不确定性的矛盾感。该实验展示了 AI 在理解和分析个人写作方面的潜力。
📄 English Summary
I Tested Sonnet 4.6 Two Ways. One of Them Got Personal
Sonnet 4.6 was released with a 1 million token context window, improved coding capabilities, and the same price as 4.5. The author conducted two experiments, the first of which involved loading six months of unpublished drafts (approximately 24,000 words) into the model and asking, 'What am I really trying to say?' The model unexpectedly identified an underlying anxiety present throughout the writing, revealing a tension between seeking empowerment and experiencing uncertainty. This experiment highlights the potential of AI in understanding and analyzing personal writing.
Powered by Cloudflare Workers + Payload CMS + Claude 3.5
数据源: OpenAI, Google AI, DeepMind, AWS ML Blog, HuggingFace 等