#490 – State of AI in 2026: LLMs, Coding, Scaling Laws, China, Agents, GPUs, AGI
Sunday, 1 February 2026 · 2 min read · Listen to the episode ↗
The podcast explores the future of AI by examining the development of large language models (LLMs) and their applications in various sectors, including coding and customer service. It discusses the competitive dynamics between U.S. and Chinese tech firms, emphasizing the need for high-quality data and human oversight to balance AI's capabilities and challenges. Additionally, it addresses the societal impacts of AI, including job displacement, trust in AI-generated content, and the importance of meaningful implementation over mere marketing hype.
The podcast delves into the state of artificial intelligence, focusing on advancements and future trends. Host Lex Fridman engages with guests Sebastian Raschka and Nathan Lambert, who discuss the significance of building models for understanding AI, particularly large language models (LLMs) and reinforcement learning from human feedback. Fridman critiques companies that exploit AI for marketing rather than meaningful implementation, emphasizing the need for genuine applications.
The conversation highlights AI agents, such as Finn for customer service, and the role of large language models in companies like Shopify. The timeline for automating human programming is examined, stressing the importance of review and debugging in AI-generated code. The podcast notes the competitive landscape in AI research, with significant advancements like the DeepSeq moment in January 2025, and discusses the impact of Chinese tech companies on the AI market.
The speakers explore the hype surrounding AI models, including Gemini 3 and Claude Opus 4.5, and the competition between US and Chinese firms. They predict a competitive landscape for 2025, with OpenAI as a key player. The discussion also addresses the trade-offs between intelligence and speed in AI models, user loyalty to language models, and the challenges of keeping up with rapid advancements.
The podcast emphasizes the importance of high-quality data in training models and the evolving landscape of open language models. Participants discuss the benefits of using multiple tools for programming and the significance of building models from scratch for learning. The conversation touches on the complexities of model architecture, including the mixture of experts concept and the importance of scaling laws in AI.
Concerns about the financial viability of scaling models are raised, alongside discussions on the implications of reinforcement learning and the challenges of training at scale. The speakers advocate for a balanced approach to using AI in coding, emphasizing the need for human oversight and the potential for AI to enhance productivity while preserving the joy of problem-solving.
The podcast also addresses the societal implications of AI, including the potential for job displacement and the importance of trust in distinguishing between AI-generated and human content. The speakers express optimism about the future of AI, highlighting the need for ongoing research and collaboration to foster innovation and address challenges in the field.
This summary was generated from the episode transcript and can contain mistakes.