This interview podcast discusses the state of AI in 2025. The conversation begins with a broad overview of AI advancements in 2024, highlighting significant cost reductions and increased capabilities of large language models (LLMs) without a major leap in intelligence. The discussion then shifts to the surprising decrease in LLM training costs, exemplified by DeepSeek's model trained for $5.5 million, and the implications for future development. Finally, the hosts and guest explore the challenges and potential of AI agents, focusing on their inherent gullibility and the need for improved user interfaces and more nuanced criticism of the technology. A specific example discussed is the vulnerability of AI agents to prompt injection attacks, even in sophisticated models like Anthropic's Claude.
