Why Ganesh Krishnan is Betting His AI Startup Against Today's LLMs

Every major AI company right now is building on the same foundation: the transformer architecture behind ChatGPT, Claude, and Gemini. Ganesh Krishnan, CEO of AI Hello, thinks that foundation is a dead end for solving AI's biggest unsolved problem — hallucination. He joins Mookie on this episode of Bald Ambition to make the case for something different entirely.The conversation opens as a real philosophical sparring match: is any of this actually intelligence, or just very fast, very sophisticated pattern-matching? Ganesh argues current LLMs are "glorified vector calculators" — zero consciousness, zero sentience, incapable of inventing anything genuinely new. Mookie pushes back hard, and they wrestle with the Turing test's collapse (people are now falling in love with chatbots and calling them their best friend), whether quantum theory or universal consciousness actually explains human sentience or is just a modern version of "turtles all the way down," and whether IQ has anything to do with consciousness at all.From there it turns pointed. Ganesh makes the case for local, on-device LLMs over frontier models — not just for cost, but for control: he describes losing 16 years of Facebook data overnight with zero recourse after another admin's post got his group banned. Mookie counters with a harder edge, questioning whether DeepSeek's ultra-cheap pricing is really altruism rather than a geopolitical play to dominate the AI race — and the conversation opens into ideological bias baked into model outputs, using Grok's "MechaHitler" incident and Chinese censorship of Tiananmen Square as two sides of the same coin.The back half gets genuinely technical: Ganesh's real bet is on reinforcement learning and diffusion models instead of transformers entirely — an architecture he believes can hit zero hallucination by grounding every response directly in real data rather than predicting the next likely word. His first real-world application, an internal tool called GraphInbox, reads a user's emails and meeting notes and drafts responses grounded entirely in their own historical data, with no context-window ceiling and no guessing.The GuestGanesh Krishnan is CEO of AI Hello, which builds AI-driven advertising optimization for e-commerce and Amazon sellers, managing over $1 billion in annual ad spend. He's currently developing a reinforcement-learning and diffusion-based architecture aimed at eliminating hallucination in AI-generated responses, built in collaboration with researchers at the University of Toronto.His Companyhttps://www.aihello.comSend the host a text! Let him know what you think Support the show