How agents will change banking forever | E2260 - This Week in Startups Recap

Podcast: This Week in Startups

Published: 2026-03-10

Duration: 1 hr 1 min

Summary

The episode explores how AI models, particularly through Andre Karpathy's Auto Research, are evolving to improve themselves autonomously, marking a significant step towards democratizing AI development. It highlights the contrasting perceptions of AI in the U.S. versus China, showcasing both excitement and skepticism in the respective regions.

What Happened

The episode kicks off with Jason and his co-host discussing the latest advancements in AI, particularly focusing on Andre Karpathy, the former AI head at Tesla. Karpathy has released a tool called Auto Research on GitHub, which allows users to create a simple training loop for AI models that can improve their own code in five-minute increments. Jason highlights that this tool represents a significant development in the AI space, demonstrating that self-improvement in AI is not only possible but also accessible to a broader audience. The implications of such tools could lead to a surge in interest and experimentation among non-experts, as seen with Toby Lutke from Shopify, who achieved impressive results using the Auto Research tool over a weekend.

The conversation then shifts to the overall landscape of AI, noting the growing number of individuals who might not be traditional AI researchers but are nonetheless able to engage with and improve AI models. Jason emphasizes the potential for exponential growth in the number of people who understand AI, drawing parallels to past technological shifts where once only a few could develop applications. This democratization, coupled with the increasing pace of innovation in labs like OpenAI and Anthropic, suggests a bright future for AI development. However, the episode also touches on the stark contrast in public perception of AI between the U.S. and China, with a recent poll indicating a significant skepticism toward AI in the U.S., where only 26% of people viewed it positively compared to 46% who opposed it.

Key Insights

Key Questions Answered

What is the Auto Research tool released by Andre Karpathy?

Auto Research is a tool that allows users to create a simplified training loop for AI models, enabling them to improve their own code in five-minute increments. This tool showcases the possibility of AI models enhancing their performance autonomously, thus marking a significant milestone in the journey towards self-improving AI. The simplicity and accessibility of this tool make it an exciting development for both AI experts and novices alike.

How does Toby Lutke's experimentation with Auto Research demonstrate the potential of AI?

Toby Lutke, the CEO of Shopify, tested the Auto Research tool over a weekend and achieved a remarkable 19% improvement in results using a model with 800 million parameters. His ability to successfully experiment with AI, despite not being an ML researcher, underscores the democratization of AI technology. This kind of experimentation among business leaders could lead to a broader understanding and utilization of AI tools across various industries.

What are the implications of AI democratization for non-experts?

The democratization of AI means that a larger number of individuals, including non-experts, can engage with and improve AI technologies. Jason notes that this could lead to an exponential increase in the number of people who understand how large language models work, similar to how app development became accessible to many after the introduction of smartphones. This shift could result in a more innovative and diverse array of AI applications and solutions.

What is the current public sentiment towards AI in the U.S. compared to China?

Recent polling indicates that public sentiment towards AI in the U.S. is quite negative, with only 26% of people viewing it positively and 46% expressing opposition. In contrast, there is a burgeoning enthusiasm for AI in China, where communities are actively engaging with AI tools like OpenClaw. This dichotomy highlights the challenges faced in promoting AI technology in the U.S., where skepticism may hinder progress.

What future developments in AI can we expect based on current trends?

The episode suggests that the pace of AI development is likely to accelerate, particularly with tools like Auto Research making self-improvement accessible. As more individuals experiment with AI, and as traditional barriers to entry diminish, we can expect a surge in innovation. The conversation also implies that advanced AI research is likely progressing at an even faster rate behind the scenes in major labs, which could lead to groundbreaking developments in the near future.