Evo 2 Drops: 8.8T-Base LLM Maps All Life Domains—People's Park Talks AI
Audio in Mandarin Chinese · English transcript below
⚡ Trillion-gene AI, film notes, Microsoft multimodal, local news fightback, Apple tags AI tunes—morning overload!
Today's core trend in the AI domain lies in the simultaneous leap in its specialization and multimodal capabilities. The open-source Evo 2 model has made significant progress in genomic analysis, while Microsoft Phi-4, with its multimodal inference model, has expanded the boundaries of visual understanding. This not only demonstrates enhanced technological depth but also, with Reflection AI's valuation doubling, mirrors the market's strong optimism about AI's future potential.
Today's Top 3 Headlines
- 科技
开源AI模型Evo 2经万亿碱基训练，可识别基因与剪接位点
Evo 2是一款开源AI，基于卷积网络StripedHyena 2，使用含8.8万亿碱基的OpenGenome2数据集训练，涵盖细菌、古菌、真核生物及噬菌体。训练分两阶段：先以约8000碱基片段识别关键特征，再以百万碱基片段捕捉大尺度特征。该模型可在复杂基因组中识别调控DNA与剪接位点。
Source ↗ - 科技
NotebookLM推出电影级视频概览功能
Google NotebookLM新增电影式视频概述功能，可将用户的研究与笔记生成完全动画化的视频。该功能采用Gemini 3、Nano Banana Pro、Veo 3等AI模型，由Gemini负责叙事、视觉风格与格式的决定。功能今日上线，仅限18岁以上Google AI Ultra订阅用户，每日最多生成20部，目前仅支持英文。
Source ↗
+6 more headlines
- LangWatch 开源 AI 代理评估层
- 微软发布Phi-4-reasoning-vision-15B多模态推理模型
- 观点：AI颠覆将带来增长
- AI能否拯救地方新闻
- 单负振幅扩展至引力子
- Apple Music将引入AI音乐透明标签