We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
cover of episode OpenAI Chief Research Officer Mark Chen: GPT 4.5 is Live and Scaling Isn’t Dead

OpenAI Chief Research Officer Mark Chen: GPT 4.5 is Live and Scaling Isn’t Dead

2025/2/27
logo of podcast Big Technology Podcast

Big Technology Podcast

AI Deep Dive AI Chapters Transcript
People
M
Mark Chen
Topics
我:GPT-4.5是我们在可预测的扩展范例中的最新里程碑。它代表着比以往模型数量级的提升,这种提升与GPT-3.5到GPT-4的飞跃相当。 至于为什么不是GPT-5,我们的命名决策是基于可预测的扩展趋势。GPT-4.5的性能符合我们对该名称的预期,它反映了模型在计算量、效率等方面的改进。 我们现在有两个可以扩展的维度:无监督学习和推理。GPT-4.5是无监督学习扩展的最新成果,但我们也在大力发展推理模型。GPT-5可能是这两个方向的集大成者,将融合两者优势。 关于扩展的局限性,我的观点是:无监督学习的扩展可以通过增加计算量、算法效率和数据来实现,GPT-4.5就是证明。无监督学习和推理是互补的,需要知识来构建推理。GPT-4.5在日常使用和知识工作方面优于GPT-4和O1,因为它拥有更多世界知识。 GPT-4.5和推理模型(如O1)在响应速度和思考深度上有所不同。GPT-4.5响应迅速,但思考较少;O1思考时间长,但答案更佳。在创意写作、部分编码和特定科学领域,GPT-4.5的表现优于推理模型。 GPT-4.5的规模是我们目前发布过的最大模型,我们观察到在该规模下,增加计算量、数据量仍然能获得与之前相同的回报。开发过程中,我们确实会中途停止、分析和重新启动模型训练,但这并非GPT-4.5独有的情况。 模型的效率提升与核心能力的开发是相对独立的。我们一直在努力提高推理效率,降低服务成本。混合专家技术等架构改进也适用于GPT-4.5以提高效率。 关于大型通用模型与小型专用模型的关系,我们既开发大型基础模型,也提供更小、更经济高效的模型。我们的目标是推动智能前沿,并使这些能力更经济高效地服务于所有人。大型模型的改进会内在地提升产品的性能,例如深度研究。 GPT-4.5在传统基准测试中取得了数量级的提升,同时在情感智能方面也有改进。这并非目标的转移,而是模型新能力的体现。我们希望用户能发现更多GPT-4.5的有趣用例。 最后,关于OpenAI的人才状况,我认为我们仍然是世界一流的AI组织,人才流动是AI领域发展的自然现象。我们内部人才济济,有很多人愿意承担责任。

Deep Dive

Chapters
This chapter explores the release of GPT-4.5, highlighting its significant improvements over previous models and addressing the question of why it's not called GPT-5. The discussion also touches upon the complementary nature of unsupervised learning and reasoning in AI scaling.
  • GPT-4.5 represents a significant advancement in OpenAI's predictable scaling paradigm.
  • The model demonstrates an order of magnitude improvement compared to previous versions.
  • OpenAI's research program explores both unsupervised learning and reasoning as parallel approaches to scaling AI models.
  • GPT-4.5 shows a 60% preference rate in everyday use cases and a 70% preference rate for productivity and knowledge work compared to GPT-4 or O1.

Shownotes Transcript

Mark Chen is the chief research officer at OpenAI. Chen joins Big Technology Podcast to discuss the debut of GPT 4.5, the company's largest model, which is going live today. In this bonus episode, Chen speaks about what the new model says about the AI scaling wall, how scaling traditional GPT models compares to reasoning models like OpenAI's o1, how important EQ is for AI models today, whether product matters more than models, and how OpenAI's talent bench looks after last year's departures. Tune in for an inside look at what it took to build OpenAI's newest and biggest large language model, and why OpenAI is committed to pushing the frontier forward.