More Latent Space episodes

Captaining IMO Gold, Deep Think, On-Policy RL, Feeling the AGI in Singapore  Yi Tay 2 thumbnail

Captaining IMO Gold, Deep Think, On-Policy RL, Feeling the AGI in Singapore Yi Tay 2

Published 23 Jan 2026

Duration: 5524

Researchers discuss AI advancements in areas such as spreadsheet automation, reinforcement learning, and large language models.

Episode Description

From shipping Gemini Deep Think and IMO Gold to launching the Reasoning and AGI team in Singapore, Yi Tay has spent the last 18 months living through...

Overview

The podcast explores the integration of AI models in handling spreadsheet tasks, highlighting their ability to generate summaries and reduce the need for manual effort. It also mentions the "Nano Banana" project, which successfully creates calming images and helps maintain system stability. The conversation reflects on returning to a familiar research environment at Google after a long absence, emphasizing the comfort of established infrastructure.

A major focus is placed on reinforcement learning (RL) as a critical research area, particularly in decision-making and model training. The discussion covers the evolution of AI models, the importance of self-awareness in learning algorithms, and the challenges of moving from imitation-based learning to independent decision-making. The potential of large language models (LLMs) in tackling complex problems, such as those found in the International Mathematical Olympiad, is examined, along with the development of systems like Gemini. Other topics include data efficiency, the difficulties of training models on large-scale data, and the future potential of artificial general intelligence (AGI).

Recent Episodes of Latent Space

20 Mar 2026 Dreamer: the Personal Agent OS David Singleton

Dreamer is an AI platform democratizing access to agentic tools for non-technical users via customizable AI assistants, community-built apps, cross-device integration, and privacy-focused features, with a beta emphasis on accessibility, real-world productivity use cases, and third-party developer opportunities.

More Latent Space episodes