Nvidia 研究人員認為超大 AI 上下文窗口是條死路
Nvidia researchers argue huge AI context windows are a dead end
Nvidia 研究團隊認為,超大上下文窗口往往成本高昂且過度;在多項測試中,定期摘要和針對性查找的效果大致相若。GPT-4 在 2023 年 3 月推出時可處理 8,192 個 token,GPT 5.6 Sol 現可處理約 100 萬個 token,約為 120 倍。

Nano Banana Pro prompted by THE DECODER
Language models can now process entire novels at once. An Nvidia research team says that's often overkill and expensive. In many tests, periodic summaries and targeted lookups worked about as well.
A language model's context window is its working memory, covering everything it holds in view when generating a reply. Over the past few years, that window has grown rapidly. GPT-4 launched in March 2023 with 8,192 tokens. Today, GPT 5.6 Sol handles around one million tokens, about 120 times as much and enough for Tolstoy's "War and Peace."
來源:The Decoder · the-decoder.com