This week on the show, the hosts cover the intense competition at the frontier, starting with the report that OpenAI has declared a "code red," halting non-essential projects as rivals close the gap. Shifting to infrastructure, AWS announced the Nova 2 model, powerful Trainium 3 chips, and "frontier agents" for autonomous work. The CEO of Turing then warns that the simple data labeling era is over, replaced by the need for highly skilled human experts. This high cost is being challenged by DeepSeek, which released two powerful, free open-source models that rival GPT-5. The hosts then dive into new development trends, covering Google's partnership with Replit for "vibe-coding", followed by a technical discussion on the need for "reinforcement learning environments"—simulated digital worlds where AI agents can learn from failure. This leads to new applications, such as MIT’s "speech-to-reality" system that uses AI and robotics to build custom objects from verbal commands. The conversation shifts to the social impact, noting Anthropic's study found a mix of professional optimism and anxiety, with data showing automation is replacing tasks. Finally, they cover the global strategic split, with Huawei’s founder arguing China should prioritize industrial automation over the US's pursuit of AGI, while the Trump administration is criticized for undermining US progress by gutting federal research funding. Hosted on Acast. See acast.com/privacy for more information.
