AI模型中大上下文窗口的概念正在被重新评估，一些人认为一百万个 token 并不等同于真正的记忆回忆。相反，这种扩展可能导致注意力在海量信息中被稀释，实际上起到了营销策略的作用，而不是模型能力的真正提升。 AI
影响 质疑大上下文窗口的实际效用，暗示它们可能无法转化为改进的AI性能。
排序理由 讨论AI模型中大上下文窗口影响的观点文章。
在 Mastodon — fosstodon.org 阅读 →
AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →
AI上下文窗口：营销噱头还是真正的内存扩展？
报道来源 [1]
-
The context window is a memory leak with a marketing team. A million tokens doesn't mean the model remembers a million tokens, it means attention gets thinner a
The context window is a memory leak with a marketing team. A million tokens doesn't mean the model remembers a million tokens, it means attention gets thinner across all of them. You didn't buy more memory, you diluted the memory you had. # AI # MachineLearning # LLM # Threadvers…