We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
cover of episode Ep 50: Fireworks CEO Lin Qiao on Why There Won’t be a Single Model, Will Hyperscalers Win Inference & AI Use-cases with PMF

Ep 50: Fireworks CEO Lin Qiao on Why There Won’t be a Single Model, Will Hyperscalers Win Inference & AI Use-cases with PMF

2024/12/16
logo of podcast Unsupervised Learning

Unsupervised Learning
AI Deep Dive AI Chapters Transcript
People
L
Lin Qiao
Topics
Lin Qiao: 我认为Fireworks从一开始就考虑到了这一点。Fireworks是一个专注于AI推理的生成式AI平台，其首要目标是提供最佳质量、最低延迟和最低成本的推理服务。然而，推理并非简单的单一模型服务，它远比这复杂得多。未来AI推理系统将是一个复杂的系统，它将整合数百个小型专家模型，并具备逻辑推理能力，能够访问各种API和数据库。 单一模型由于其概率性本质和有限的知识，难以提供始终如一的准确结果并解决复杂的现实问题。控制模型的幻觉至关重要。此外，许多客户使用我们的平台来解决复杂的业务问题，这需要整合多个模型和多种模态。例如，我们现在进行的对话就涉及音频和视觉信息的处理，以提供良好的交互式体验。许多面向消费者的应用程序也需要处理多种模态的信息，甚至在同一模态内也需要使用多个专家模型，例如大型语言模型中用于分类、摘要、多轮对话和函数调用的不同专家模型。 单一模型的知识有限，其知识仅限于其训练数据，而训练数据是有限的，并非无限的。现实世界中大量信息存在于API（公共API或企业内部专有的私有API）的背后，如果没有直接与企业合作，就无法访问这些信息。因此，我们认为下一个挑战是如何超越单一模型服务。我们需要的是复合AI系统，它整合多个模型、多种模态以及各种API和数据库，以提供最佳的AI结果。

Deep Dive

Shownotes Transcript

Lin Qiao, the co-founder of Fireworks.ai, sits down for a deep dive into the future of AI. Lin ran the PyTorch team at Meta, which developed some of the most fundamental open-source AI software in use today. She’s got a riveting perspective on the AI landscape that is a must-listen.

 

[0:00] Intro

[1:06] Fireworks: Revolutionizing AI Inference

[2:12] Challenges in AI Model Development

[4:05] The Future of AI: Compound Systems

[4:32] Designing Effective AI Tools

[10:26] Customization and Fine-Tuning in AI

[14:06] Human-in-the-Loop Automation

[16:38] Evaluating AI Models

[19:18] Building Complex AI Systems

[21:18] Function Calling and AI Orchestration

[26:52] AI Infrastructure and Hardware

[31:08] Small Expert Models

[31:27] Hyperscalers and Resource Management

[32:14] Inference Systems and Scalability

[33:08] Running Models Locally: Cost and Privacy

[35:20] Open Source Models and Meta's Role

[36:41] The Evolution of AI Training and Inference

[38:04] Fireworks' Vision and Market Strategy

[40:46] The Impact of Generative AI

[45:18] AI Research and Future Trends

[46:58] Building for a Rapidly Changing AI Landscape

[49:36] Quickfire

 

With your co-hosts: 

@jacobeffron 

  • Partner at Redpoint, Former PM Flatiron Health

 

@patrickachase 

  • Partner at Redpoint, Former ML Engineer LinkedIn

 

@ericabrescia 

  • Former COO Github, Founder Bitnami (acq’d by VMWare)

 

@jordan_segall 

  • Partner at Redpoint