Claude Sonnet 5性能逼近Opus 4.8，价格更低，新分词器让同等文本token增加约30%，值得模型使用者关注。
What's new in Claude Sonnet 5 Claude Sonnet 5 came out this morning . I always head straight for the "what's new" developer docs because they ten…
揭秘LLM无法真正阅读，而是靠BPE分词器将文本切成词块，频繁内容整体化，罕见内容碎片化。
Here's a fact that breaks people's mental model of large language models the first time they really sit with it: A language model never sees your word…
突破不同模型家族间的分词器壁垒，提出在策略蒸馏新方法，提升跨模型知识迁移效果。
arXiv:2606.09456v1 Announce Type: new Abstract: On-Policy Distillation (OPD) has become a core technique in the post-training of Large Language Models…
一种针对Brahmic文字优化的131K词汇BPE分词器，可无损替换OpenAI的o200k_base，显著提升印地语/孟加拉语等文字压缩率，同时保留英文和代码性能。
arXiv:2605.29379v1 Announce Type: cross Abstract: We present BrahmicTokenizer-131K, a 131,072-vocabulary byte-level BPE tokenizer that closes the Brah…
大模型分词器也能泄露训练隐私？新研究揭示针对LLM分词器的成员推理攻击
arXiv:2510.05699v4 Announce Type: replace-cross Abstract: Membership inference attacks (MIAs) are widely used to assess the privacy risks associated w…
无需辅助组件的投影引导跨分词器知识蒸馏，有效解决词汇不兼容问题。
arXiv:2605.21699v1 Announce Type: cross Abstract: Cross-tokenizer knowledge distillation allows a student model to learn from teachers with incompatib…
SpeakerLLM 将说话人理解与验证推理整合到自然语言界面，不仅区分‘是谁’，还能解释声音轮廓、录音条件等证据，为可解释的说话人认知铺平道路——这比单纯打分有用得多。
arXiv:2605.15044v1 Announce Type: cross Abstract: As audio-first agents become increasingly common in physical AI, conversational robots, and screenle…