目前模型之间是如何分类的

OpenAI o3-mini performs chain-of-thought reasoning in context,which leads to strong performance across both capabilities and safety benchmarks.These increased capabilities come with significantly improved performance on safety benchmarks,but also increase certain types of risk.We have identified our models as medium risk in Persuasion,CBRN,and Model Autonomy within the OpenAI Preparedness Framework.Overall,o3-mini,like OpenAI o1,has been classified as medium risk in the Preparedness Framework,and we have incorporated commensurate safeguards and safety mitigations to prepare for this new model family.Our deployment of these models reflects our belief that iterative realworld deployment is the most effective way to bring everyone who is affected by this technology into the AI safety conversation.32Authorship,credit attribution,and acknowledgmentsPlease cite this work as“OpenAI(2025)”.ResearchTrainingBrian Zhang,Eric Mitchell,Hongyu Ren,Kevin Lu,Max Schwarzer,Michelle Pokrass,Shengjia Zhao,Ted SandersEvalAdam Kalai,Alex Tachard Passos,Ben Sokolowsky,Elaine Ya Le,Erik Ritter,Hao Sheng,Hanson Wang,Ilya Kostrikov,James Lee,Johannes Ferstad,Michael Lampe,Prashanth Radhakrishnan,Sean Fitzgerald,Sebastien Bubeck,Yann Dubois,Yu BaiFrontier Evals and PreparednessAndy Applebaum,Elizabeth Proehl,Evan Mays,Joel Parish,Kevin Liu,Leon Maksin,Leyton Ho,Miles Wang,Michele Wang,Olivia Watkins,Patrick Chao,Sandhini Agarwal,Samuel Miserendino,Tejal PatwardhanProduct

Figure 1.3.4 illustrates the sectoral origin ofnotableAI releases by the year the models were released.Epoch categorizes models based on their source:Industry includes companies such as Google,Meta,and OpenAI;academia covers universities like Tsinghua,MIT,and Oxford;government refers to state-affiliated research institutes like the UK’s Alan Turing Institute for AI and Abu Dhabi’s Technology Innovation Institute;and research collectives encompass nonprofit AI research organizations such as the Allen Institute for AI and the Fraunhofer Institute.Artificial IntelligenceIndex Report 2025the AI model ecosystem.If readers believe that models from specific countries are missing,they are encouraged to contact the AI Index team,which will work to address the issue.Until 2014,academia led in terms of releasing machine learning models.Since then,industry has taken the lead.According to EpochAI,in 2024,industry produced 55 notable AI models.That same year,Epoch AI identified no notable AI models originating from academia(Figure 1.3.5).18 Over time,industry-academia collaborations have contributed to a growing number of models.The proportion of notable AI models originating from industry has steadily increased over the past decade,growing to 90.2% in 2024.18 This figure should be interpreted with caution.A count of zero academic models does not mean that no notable models were produced by academic institutions in 2023,but rather that Epoch AI has not identified any as notable.Additionally,academic publications often take longer to gain recognition,as highly cited papers introducing significant architectures may take years to achieve prominence.Table of ContentsChapter 1 Preview471.3 Notable AI Models Chapter 1:Research and DevelopmentNumber of notable AI models by sector,2003–24Source:Epoch AI,2025|Chart:2025 AI Index report6050403020100

首先为方便大家对大模型有一个整体的认知，我们先从大模型的整体架构着手，来看看大模型的组成是怎么样的。下面是我大致分的个层。从整体分层的角度来看，目前大模型整体架构可以分为以下几层：[heading3]1.基础层：为大模型提供硬件支撑，数据支持等[content]例如A100、数据服务器等等。[heading3]2.数据层[content]这里的数据层指的不是用于基层模型训练的数据基集，而是企业根据自己的特性，维护的垂域数据。分为静态的知识库，和动态的三方数据集[heading3]3.模型层：LLm或多模态模型[content]LLm这个大家应该都知道，large-language-model，也就是大语言模型，例如GPT，一般使用transformer算法来实现。多模态模型即市面上的文生图、图生图等的模型，训练所用的数据与llm不同，用的是图文或声音等多模态的数据集[heading3]4.平台层：模型与应用间的平台部分[content]比如大模型的评测体系，或者langchain平台等，提供模型与应用间的组成部分[heading3]5.表现层：也就是应用层，用户实际看到的地方[content]这个就很好理解了，就不用我多作解释了吧