AI为什么会“失语”？

时代少年团粉丝最近在社交媒体平台发了一组截图。他们让MiniMax写一句“时代少年团队长是谁”，模型支支吾吾，蹦出来的是“马嘉轩”“马丝祺”等完全错误的答案。但换个问法，问“时代少年团队长有什么经历”，模型对答如流。似乎AI知道马嘉祺是谁，但就是叫不 ...
来自MSN

How a 732-byte Python script exploited Linux

A 732-byte Python script has uncovered a significant vulnerability in the Linux kernel, affecting users worldwide. Explore the details of this exploit, its implications, and the urgent need for ...

AI 大模型的「中文税」：中文比英文更费 Token，为什么？

Opus 4.7 刚发布那几天，X 上怨声载道。有人说一次对话就把她的 session 额度用光了，有人说同一段代码跑完的成本比上周翻了一倍多；还有人晒出自己 200 美元 Max 订阅不到两小时就触顶的截图。 独立开发者 BridgeMind 承认 ...
IEEE

Controlling byte pair encoding for neural machine translation

Abstract: Byte pair encoding(BPE) is an approach that segments the corpus in such a way that frequent sequence of characters are combined; it results to having word surface forms divided into its' ...
Microsoft

BEAST: Efficient Tokenization of B-Splines Encoded Action Sequences for Imitation Learning

We present the B-spline Encoded Action Sequence Tokenizer (BEAST), a novel action tokenizer that encodes action sequences into compact discrete or continuous tokens using B-splines. In contrast to ...
IEEE

Optimizing Byte Pair Encoding Tokenization for South African Languages

Abstract: In this paper, we introduce an Optimized Byte Pair Encoding (OBPE) tokenizer where the algorithm is optimized for the South African languages, including Sesotho, Setswana, Xhosa, Xitsonga, ...