The AI Arms Race Begins | Raghav Toshniwal on Sovereign AI, China, India & Frontier Models

About this episode

Timestamps00:00 Claude Fable gets blocked03:26 Does the ban actually help Anthropic?08:49 AI becomes a national security issue09:28 Why countries want sovereign AI13:30 NVIDIA, Taiwan, and the AI supply chain17:04 Who should fund AI infrastructure?18:28 Are AI models actually profitable?21:27 What went wrong with GPT-4.5?24:20 Can India build its own frontier model?28:27 RL environments and the new data bottleneck30:02 Is this good or bad news for India?32:41 What this means for frontier AI labs33:36 Recursive self-improvement begins35:23 Do AI products have switching costs?40:46 Are we already in AI takeoff?42:32 Why hasn’t AI caused runaway growth yet?44:42 Where do AI use cases stop?50:16 What does AI safety actually mean?52:51 Why would AI become dangerous?56:08 What safety teams are working on59:32 Does RLHF solve alignment?1:00:26 Reward hacking and Goodhart’s law1:03:40 Why safety research needs compute1:07:38 What should normal AI users do?1:08:03 Gradual disempowerment1:11:27 Are we already seeing warning signs?1:13:48 Are LLMs the path to AGI?1:15:23 Timelines for automated AI researchers1:18:20 Can independent researchers contribute?1:22:45 ClosingIn this episode, I sit down with Raghav Toshniwal to unpack why AI is no longer just a product race, but increasingly a geopolitical one.We discuss the Claude Fable controversy, why countries are thinking about sovereign AI, what the US-China AI race means for India, and how compute, chips, NVIDIA, Taiwan, and frontier labs all fit into the bigger picture.We also get into recursive self-improvement, whether we are already in an AI takeoff, and why AI safety researchers worry about alignment, reward hacking, and gradual disempowerment.Topics covered: The AI arms race between the US and China Sovereign AI and why countries want their own models India’s position in the frontier model race NVIDIA, Taiwan, TSMC and the compute bottleneck OpenAI, Anthropic and recursive self-improvement AI safety, alignment and existential risk Whether automated AI researchers change everythingMore on RaghavPodcast Socials :Instagram : https://www.instagram.com/decentmakeoverTwitter : https://twitter.com/decentmakeovrLinkedin : https://www.linkedin.com/in/ryan-dsouza-74542b295/PODCAST INFO:Podcast website: https://anchor.fm/ryandsouzaApple Podcasts: https://apple.co/3NQhg6SSpotify: https://spoti.fi/3qJ3tWJAmazon Music: https://amzn.to/3P66j2BGoogle Podcasts: https://bit.ly/3am7rQcGaana: https://bit.ly/3ANS4v1RSS: https://anchor.fm/s/609210d4/podcast/rss