We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
cover of episode #190 - AI scaling struggles, OpenAI Agents, Super Weights

#190 - AI scaling struggles, OpenAI Agents, Super Weights

2024/11/28
logo of podcast Last Week in AI

Last Week in AI

AI Deep Dive AI Chapters Transcript
People
A
Andrey Kurenkov
J
Jeremie Harris
Topics
Andrey Kurenkov认为,当前AI发展面临瓶颈,单纯依靠扩大模型规模、增加数据和计算能力的策略,其改进效果正在递减。他认为,这并非意味着AI发展停滞,而是意味着单纯的规模化方法已不足以持续提升AI性能,需要探索新的方法。他同时指出,AI代理工具的出现和多模态模型的发展是AI领域的重要趋势。 Jeremie Harris补充指出,AI发展的瓶颈在于工业基础设施,例如能源供应和计算集群规模难以满足快速发展的需求。他认为,单纯的规模化方法已达到极限,需要从工业层面解决能源和算力问题。他强调,当前AI研究的重点已转向后训练阶段,包括强化学习和影响时间缩放定律,这些技术与训练时间缩放定律相结合,能够进一步提升AI性能。

Deep Dive

Chapters
Discussions around the potential slowdown in AI development, focusing on challenges faced by OpenAI, Google, and Anthropic in building more advanced AI models.
  • Next-generation models from OpenAI, Google, and Anthropic are not meeting performance expectations.
  • Pure scaling approaches are becoming challenging due to diminishing returns.
  • The community is divided on whether this signals a wall in AI improvement or just a temporary plateau.

Shownotes Transcript

Our 190th episode with a summary and discussion of last week's big AI news!

Hosted by Andrey Kurenkov) and Jeremie Harris).

Note from Andrey: this one is coming out a bit later than planned, apologies! Next one will be coming out sooner. Feel free to email us your questions and feedback at [email protected] )and/or [email protected])

Read out our text newsletter and comment on the podcast at https://lastweekin.ai/).

Sponsors:

  • The Generator -) An interdisciplinary AI lab empowering innovators from all fields to bring visionary ideas to life by harnessing the capabilities of artificial intelligence

In this episode:

  • OpenAI's pitch for a $100 billion data center and AI strategy plan outlines infrastructure and regulatory needs, emphasizing AI's foundational role akin to electricity. 
  • Google's Gemini model challenges OpenAI's dominance, showing strong performance in chatbot arenas alongside generative AI advancements. 
  • DeepMind's AlphaFold3 gets open-sourced for academic use, while new chips from NVIDIA and Google show significant performance boosts. 
  • Anthropic and TSMC updates highlight strategic funding, regulation influences, and the complex dynamics of AI hardware and international policy.

If you would like to become a sponsor for the newsletter, podcast, or both, please fill out this form).

Timestamps + Links:

(00:00:00) Intro / Banter

(00:02:44) News Preview

(00:03:34) Sponsor Break

Tools & Apps

(00:04:36) OpenAI, Google and Anthropic Are Struggling to Build More Advanced AI)

(00:16:22) OpenAI Nears Launch of AI Agent Tool to Automate Tasks for Users)

(00:19:14) Google drops new Gemini model and it goes straight to the top of the LLM leaderboard)

(00:19:14)  Chinese AI startup takes aim at OpenAI's Sora with image-to-video tool launch)

(00:20:04) Introducing the Forge Reasoning API Beta and Nous Chat: An Evolution in LLM Inference)

Applications & Business

(00:23:47) OpenAI Discusses AI Data Center That Could Cost $100 Billion)

(00:26:48) Elon Musk's massive AI data center gets unlocked — xAI gets approved for 150MW of power, enabling all 100,000 GPUs to run concurrently)

(00:29:34) Newest Google and Nvidia Chips Speed AI Training)

(00:34:45) Ex-OpenAI CTO Murati’s New Team Takes Shape)

(00:34:45) Amazon Discussing New Multibillion-Dollar Investment in Anthropic)

Projects & Open Source

(00:37:52) Google DeepMind open-sources AlphaFold 3, ushering in a new era for drug discovery and molecular biology)

(00:41:29) Near plans to build world’s largest 1.4T parameter open-source AI model)

Research & Advancements

(00:45:38) The Super Weight in Large Language Models)

(00:55:42) Compositional Abilities Emerge Multiplicatively: Exploring Diffusion Models on a Synthetic Task)

(01:03:47) Mixture-of-Transformers: A Sparse and Scalable Architecture for Multi-Modal Foundation Models)

(01:08:14) Contextualized Evaluations: Taking the Guesswork Out of Language Model Evaluations)

Policy & Safety

(01:11:14) The Code of Practice for general-purpose AI offers a unique opportunity for the EU)

(01:15:38) Three Sketches of ASL-4 Safety Case Components)

(01:23:05) U.S Department of Commerce finalizes $6.6 billion CHIPS Act funding for TSMC Fab 21 Arizona site) , TSMC cannot make 2nm chips abroad now: MOEA)

(01:26:21) OpenAI to present plans for U.S. AI strategy and an alliance to compete with China)

(01:30:42) OpenAI loses another lead safety researcher, Lilian Weng)

(01:33:00) Outro