[2206.07682] Emergent Abilities of Large Language Models
6
公开标注数
6
参与人数
2026-07-26 09:17:41
首次 Whisper
本页的公开 Whisper
划选高亮2026-07-26 15:41:41
原文高亮摘录
“Transactions on Machine Learning Research (TMLR), 2022”
Whisper 随想笔记
Ah, the classic arXiv citation with a journal name, but I heard TMLR is not even indexed yet.
划选高亮2026-07-26 12:38:41
原文高亮摘录
“unpredictable phenomenon that we refer to as emergent abilities”
Whisper 随想笔记
But is it really unpredictable or are we just bad at measuring small gains?
划选高亮2026-07-26 12:29:41
原文高亮摘录
“unpredictable phenomenon that we refer to as emergent abilities”
Whisper 随想笔记
It's wild how one day the model just gets it, no warning at all.
划选高亮2026-07-26 09:35:41
原文高亮摘录
“Scaling up language models has been shown to predictably improve performance”
Whisper 随想笔记
Not convinced it's truly unpredictable, maybe our extrapolation methods are just too naive.
划选高亮2026-07-26 09:26:41
原文高亮摘录
“Scaling up language models has been shown to predictably improve performance”
Whisper 随想笔记
I've seen this in practice—our small model couldn't reason at all, but the big one just... could.
划选高亮2026-07-26 09:17:41
原文高亮摘录
“Scaling up language models has been shown to predictably improve performance”
Whisper 随想笔记
So scaling is basically a lottery ticket? Sometimes you win big, sometimes you just get better at the same stuff.
分享本页 Whisper
短链接
https://domwhisper.com/s/2d76bc16fcc7嵌入代码
<iframe src="https://domwhisper.com/embed/2d76bc16fcc7" width="100%" height="480" style="border:0;border-radius:16px" loading="lazy"></iframe>