Latent Space: The AI Engineer Podcast cover image

Latent Space: The AI Engineer Podcast

Why AI Agents Don't Work (yet) - with Kanjun Qiu of Imbue

Oct 14, 2023
Kanjun Qiu, co-founder of Imbue, shares insights on building capable AI agents and the challenges they face. He discusses the pitfalls of relying solely on reinforcement learning, revealing why it struggles with higher-level reasoning. The conversation covers innovative projects like Avalon, aimed at improving AI training environments, and the importance of high-quality data. Kanjun also emphasizes the need for creative company cultures and community-driven innovation to advance AI technology responsibly and effectively.
01:05:02

Episode guests

Podcast summary created with Snipd AI

Quick takeaways

  • imbue is focused on optimizing models for reasoning and developing non-leaky abstractions to improve the reliability of AI agents in real-world scenarios.
  • imbue is collaborating with other companies to co-build evaluation frameworks, aiming to contribute to the development of better benchmarks and evaluation metrics for agent research and development.

Deep dives

The focus on reasoning and inspectability

imbue is placing a strong emphasis on optimizing their models for reasoning and developing non-leaky abstractions to improve the reliability and trustworthiness of artificial agents. They aim to address these core issues in order to create agents that can be deployed effectively and reliably in real-world scenarios.

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
App store bannerPlay store banner