1
Linus Torvalds 称 AI 开发即便再强，也不代表你能少动脑
AI代码助手，基于上下文智能补全代码，支持多种语言，提升编程效率，但需开发者保持思考。
IT之家 5 月 23 日消息，在 Linux 基金会北美开源峰会上， Linus Torvalds 直言，AI 已开始改变 Linux 内核开发节奏。 Torvalds 指出自 Git 确立发布流程以来， 内核发布长期稳定了约 20 年，但过去 6 个月开始出现明显变化。 IT之家援引博文介绍，T…
2
HMD 在印度推出新机 Vibe 2 5G，预装支持 22 种印度语言的 AI 助手
HMD在印度推出Vibe 2 5G，预装支持22种印度语言的AI助手，主打本地化智能体验。
IT之家 5 月 22 日消息，据外媒 TechCrunch 今日报道，HMD 正式发布首款智能手机 Vibe 2 5G，并首次预装印度 AI 企业 Sarvam 开发的聊天机器人 Indus，希望借低价硬件推动本土 AI 助手在印度市场普及。 今年 2 月，双方在新德里举行的印度 AI 峰会上宣布…
3
Toward Robust Multilingual Adaptation of LLMs for Low-Resource Languages
被ICML 2026收录，提出针对低资源语言的LLM鲁棒多语言适配方法，增强跨语言泛化能力。
arXiv:2510.14466v3 Announce Type: replace Abstract: Large language models (LLMs) continue to struggle with low-resource languages, primarily due to li…
4
Multilingual jailbreaking of LLMs using low-resource languages
用非洲低资源语言玩多轮对话，成功绕过ChatGPT、Gemini等主流大模型的安全护栏，安全漏洞新发现。
arXiv:2605.18239v1 Announce Type: new Abstract: Large Language Models (LLMs) remain vulnerable to jailbreak attempts that circumvent safety guardrails…
5
Why Do Safety Guardrails Degrade Across Languages?
揭秘多语言下AI安全护栏失效的根本原因，从数据分布到tokenization的深度剖析。
arXiv:2605.17173v1 Announce Type: cross Abstract: Large language models exhibit safety degradation in non-English languages. Standard evaluation relie…
6
ML-Embed: Inclusive and Efficient Embeddings for a Multilingual World
多语言文本嵌入新方案ML-Embed，突破成本、语种和透明壁垒，迈向包容高效未来。
arXiv:2605.15081v1 Announce Type: cross Abstract: The development of high-quality text embeddings is increasingly drifting toward an exclusionary futu…
7
Multilingual and Multimodal LLMs in the Wild: Building for Low-Resource Languages
探讨多模态大模型在低资源语言环境下的实际构建挑战与方案
arXiv:2605.17152v1 Announce Type: new Abstract: Multimodal LLMs are evolving from vision-language to tri-modality that see, hear, and read, yet pipeli…
8
Multilingual OCR-Aware Fine-Tuning and Prompt-Guided Chain-of-Thought Reasoning for Multimodal Large Language Models
多模态大模型结合多语言OCR与提示引导的思维链推理，提升图文文字理解能力
arXiv:2605.16409v1 Announce Type: cross Abstract: Optical character recognition (OCR) and multilingual text understanding remain major failure modes o…
9
Does language matter for spoken word classification? A multilingual generative meta-learning approach
元学习在多语言口语单词分类中优于监督学习，新方法初探语言无关性
arXiv:2605.13084v2 Announce Type: replace-cross Abstract: Meta-learning has been shown to have better performance than supervised learning for few-sho…
10
A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM$\Delta$ Integration into Upcycled MoE
提出一种数据高效的多语言LLM扩展方法，通过后训练将PARAMΔ集成到升级MoE中，仅需少量目标语言文本即可显著提升性能。
arXiv:2605.18083v1 Announce Type: new Abstract: Expanding Large Language Models~(LLMs) to new languages is a costly endeavor, demanding extensive Cont…
11
Thinnest AI
用100+语言构建语音AI代理，每分钟仅1.5卢比，极致性价比。
Build Voice AI Agents in 100+ languages for ₹1.5/min Discussion | Link
12
Show HN: Bypassing the AWS Lambda 4KB limit to run polyglot AI agents
绕过AWS Lambda 4KB限制，在多语言AI代理中集成支付与安全模式。
Article URL: https://brewhubphl.com/engineering/parity-contracts-for-polyglot-llm-commerce-a-case-study Comments URL: https://news.ycombinator.com/ite…
13
Global news partnerships: Le Monde and Prisa Media
OpenAI携手两大国际媒体，ChatGPT新增法语西班牙语新闻内容，拓展信息生态。
We have partnered with international news organizations Le Monde and Prisa Media to bring French and Spanish news content to ChatGPT.
14
IndicSafe: A Benchmark for Evaluating Multilingual LLM Safety in South Asia
首个系统性评测大模型在南亚12种印度语言上安全性的基准，揭示低资源语言安全漏洞。
arXiv:2603.17915v2 Announce Type: replace-cross Abstract: As large language models (LLMs) are deployed in multilingual settings, their safety behavior…
15
DetectRL-X: Towards Reliable Multilingual and Real-World LLM-Generated Text Detection
多语言大模型文本检测新方案，针对真实场景打造更可靠识别系统
arXiv:2605.15518v1 Announce Type: new Abstract: The effective detection and governance of Large Language Model (LLM) generated content has become incr…
16
ForMaT: Dataset for Visually-Grounded Multilingual PDF Translation
PDF翻译新突破！ForMaT数据集保留原文布局，支持15种语言对，精准处理嵌套表格等复杂结构。
arXiv:2605.15794v1 Announce Type: new Abstract: We present ForMaT (Format-Preserving Multilingual Translation), a parallel corpus of 3,956 PDFs across…
17
Tell HN: Don't make your startup multilingual
初创公司应优先验证市场，不要过早铺开多语言支持，否则徒增维护成本
When launching a startup, don’t make it multilingual until you have stable revenue and a team to support it. You can’t know whether the startup will b…
18
Wiki Dumps to Training Corpora: South Slavic Case
从维基百科原始数据构建高质量南斯拉夫语语料库，详解七种语言的文本提取与清洗流程。
arXiv:2604.25384v2 Announce Type: replace Abstract: This paper presents a pipeline designed to transform raw Wikimedia dumps into quality textual corp…
19
Toward LLMs Beyond English-Centric Development
揭示大模型英语偏见真相，证明持续预训练成本优势不存在，语言专用投资或成必然。
arXiv:2605.15613v1 Announce Type: new Abstract: Through an analysis of sequences generated by open-weight large language models (LLMs), we demonstrate…
20
Do Chinese models speak Chinese languages?
探究中国开源大模型究竟支持哪些中国境内语言，揭示多语言能力背后的数据偏见。
arXiv:2504.00289v3 Announce Type: replace-cross Abstract: The release of top-performing open-weight LLMs has cemented China's role as a leading force …