1
7名博士生仅用3个月从零训练7B大模型：代码+数据+训练日志全公开
几百个Agent参与研发，AI开始造AI
2
Recognition-Conditioned Reasoning: A Training-Free Multimodal-LLM Pipeline for Fine-Grained Micro-Action Understanding
无需训练的多模态大模型流水线，精准解析微动作语义，ACM Multimedia 2026亮点工作
arXiv:2608.21022v1 Announce Type: new Abstract: Micro-actions are subtle, short, low-amplitude body movements, such as a fidgeting hand or a slight he…
3
Show HN: I trained three LLMs from scratch and put them online to try them out
从零训练三个大语言模型，并放上线供你亲自试玩，体验AI训练成果。
Article URL: https://huggingface.co/spaces/JohnEnev/modern-llm-playground Comments URL: https://news.ycombinator.com/item?id=49299726 Points: 2 # Comm…
4
Show HN: Horus-runtime – Train your own tiny LLM from scratch
用TinyStories小数据集从零预训练微型LLM，适合资源有限时上手实战。
Article URL: https://github.com/temple-compute/pantheon/tree/main/workflows/ai/w02-tiny-llm Comments URL: https://news.ycombinator.com/item?id=4913270…
5
LLMAR: A Tuning-Free Recommendation Framework for Sparse and Text-Rich Industrial Domains
无需训练，基于LLM推理的工业推荐框架，创新性利用大模型捕获用户潜在动机，解决稀疏文本领域推荐难题。
arXiv:2604.16379v2 Announce Type: replace-cross Abstract: Industrial B2B applications (e.g., construction site risk prediction, material procurement) …
6
Scaling Native Multimodal Pre-Training From Scratch
从零开始的跨模态预训练规模扩展，揭示原生多模态模型的新范式与核心挑战。
arXiv:2607.22043v1 Announce Type: new Abstract: Although large language models (LLMs) exhibit remarkable reasoning capabilities, their reliance on tex…
7
Show HN: A signal engine with no training, no tokens, no neural net
管道工用AI搭档开发的无神经网络信号引擎，28秒处理真实脑数据，零训练零API密钥，笔记本即可运行。
Article URL: https://github.com/JJardine919/voodoo-aoi Comments URL: https://news.ycombinator.com/item?id=48832055 Points: 1 # Comments: 0
8
Show HN: Meadow Mind – a 7B diffusion LLM plays Gym games with zero training
7B扩散LLM零训练直接上手Gym游戏，展示语言模型在强化学习环境中的零样本决策能力
Article URL: https://github.com/Hey-Meadow/meadow-mind Comments URL: https://news.ycombinator.com/item?id=48479887 Points: 3 # Comments: 0
9
FreeRet: MLLMs as Training-Free Retrievers
多模态大模型无需微调即可作为检索器，FreeRet方法突破传统范式。
arXiv:2509.24621v3 Announce Type: replace Abstract: Multimodal large language models (MLLMs) are emerging as versatile foundations for mixed-modality …