In this episode, Grigorij Dudnik (Linkedin) shares his journey to becoming a pioneer in generative AI automation. He spoke at the Big Data Conference 2025 with his presentation How we created a thinking robot (Check out the presentation). The conversation centers on his philosophy that AI should liberate humans from "digital drudgery" rather than increasing screen time. Check out his YouTube channel where he shares his learnings and his Github.
The conversation follows Grigorij’s journey from building digital automation tools to creating physical robots that "think":
From Engineering to AI Automation: Grigorij begins by explaining his background in mechanical and electrical engineering, which eventually led him to the world of generative AI. He shares his core philosophy: AI should be used to reduce "digital drudgery" and free people from repetitive screen-based work.
The "Tinder GPT" Experiment: The discussion dives into one of Grigorij's viral early projects—an AI agent that automated dating app interactions. He explains how he used LLMs to handle the "bureaucracy" of initial small talk and scheduling, highlighting the potential for AI to manage social and administrative overhead.
Building the "Thinking Robot": Grigorij details the technical shift from digital agents to physical robotics. He describes the process of running 7B parameter LLMs locally on accessible hardware like Raspberry Pi, allowing robots to process information without relying on the cloud.
Solving the Trolley Dilemma: A standout moment in the podcast is the discussion of his toy train project. By connecting an LLM to a physical train set, he forced the AI to solve the classic ethical "Trolley Problem" in real-time, demonstrating that modern AI can handle nuanced reasoning that traditional programming cannot.
Overcoming Hallucinations in Robotics: The episode concludes with a look at the future of AI in hardware. Grigorij explains how integrating LLMs with physical sensors helps mitigate "hallucinations," as the physical world provides a "ground truth" that forces the AI to remain accurate and safe.
