Meta Llama Drifts from Open-Source Roots—Ajinomoto Becomes AI Gold Mine
Audio in Mandarin Chinese · English transcript below
⚡ Meta seals code, Ajinomoto becomes AI gold; DeepSeek trains trillion-param model on smuggled chips; Chinese labs slash multimodal prices—global AI arms race surges!
The AI domain is undergoing a comprehensive reshaping, from LLM capabilities to application strategies and supply chains. DeepSeek V4's upcoming layered interaction model signals that AI applications are moving towards refinement and efficiency optimization, while its deep adaptation to Chinese hardware also underscores the strategic importance of cost control. Concurrently, Meta Llama models, despite their competitiveness, are gradually losing their open-source essence. This, coupled with the phenomenon of Ajinomoto being seen as a goldmine due to its monopoly on critical AI materials, jointly reveals that AI competition has permeated every layer of the technology stack, from high-level applications to underlying infrastructure, all fraught with strategic considerations.
Today's Top 3 Headlines
- 科技
传 DeepSeek V4 将于 4 月底发布
传闻称，DeepSeek 创始人梁文锋表示，公司新旗舰大模型 DeepSeek V4 预计 4 月底发布，被视为该 AI 企业的重要进展。
Source ↗
+12 more headlines
- DeepSeek V4：万亿参数AI模型
- AI 不再只是“更聪明”，而是更讲策略
- 中国初创公司DeepSeek被曝使用遭禁运的NVIDIA Blackwell芯片训练AI模型
- Stephen Thaler 早于 OpenAI 数十年为 AI 发明申请专利
- Meta（元宇宙）超智实验室发布多模态推理模型Muse Spark
- 阿里巴巴匿名发布新AI视频生成模型
- Anthropic 是在限制 Mythos 的发布以保护互联网，还是在保护 Anthropic（AI公司）？
- 70人AI图像初创公司挑战硅谷巨头
- 谷歌、Meta（元宇宙）、Snap（社交应用）、微软谴责欧盟儿童性剥削扫描法失效
- DeepMind CEO称英国AI人才领先硅谷
- AI在总结复杂癌症病理报告方面优于医生——西北医学研究
- DeepMind首席执行官称ChatGPT用户数破亿为“战时”，Google全力迎战——AOL