MaChan retweetledi
“Grau, teurer Freund, ist alle Theorie, Und grün des Lebens goldner Baum.”
- 灰色的，亲爱的朋友，是一切理论；
而生命的金树常青 -
— Faust, Faust I, Studierzimmer —
#BanGDream
#AveMujica #三角初華
中文
MaChan
61 posts
@reu_kastel @Rococo90933671 我感觉threads比小红书还要呃呃很多，小红书至少还确实有很多有用的小技巧/信息分享，threads的算法设计出来就是为了激怒每一个用户，timeline上全是最纯粹的ragebait
中文
MaChan retweetledi
> "but natural text sits in between, partly structured and partly random"
想到一个很有趣的研究，如果你问一个模型一个喜欢黄色的人最可能做什么工作，其大概率可能回复你是校车司机。因为"there is a correlation between words that cluster with yellow and words that cluster with school buses"。大模型可能会错误地学到错误的/不准确的/本不存在的词汇与词汇（而非概念与概念）之间的怪异的 n 阶相关性。
garymarcus.substack.com/p/new-ways-to-…
arxiv.org/abs/2512.09742
Rohan Paul@rohanpaul_ai
Terence Tao says the math behind today’s LLMs is actually simple. Training and running them mostly uses linear algebra, matrix multiplication, and a bit of calculus, material an undergraduate can handle. We understand how to build and operate these models. The real mystery is why they work so well on some tasks and fail on others, and why we cannot predict that in advance. We lack good rules for forecasting performance across tasks, so progress is largely empirical. A key reason is the nature of real-world data. Pure noise is well understood, perfectly structured data is well understood, but natural text sits in between, partly structured and partly random. Mathematics for that middle regime is thin, similar to how physics struggles at meso-scales between atoms and continua. Because of this gap, we can describe the mechanisms but cannot yet explain capability jumps or give reliable task-level predictions. That mismatch, simple machinery versus hard-to-predict behavior, is the core puzzle. ---- Video from 'Dr Brian Keating' YT Channel (Link in comment)
中文
