Humans&: Bridging IQ and EQ in Machine Learning with Eric Zelikman - No Priors: Artificial Intelligence | Technology | Startups Recap

Podcast: No Priors: Artificial Intelligence | Technology | Startups

Published: 2025-10-09

Duration: 37 min

Guests: Eric Zelikman

Summary

Eric Zelikman discusses the challenges of integrating human-centric goals in AI development, emphasizing the need for models that understand and empower people rather than replace them.

What Happened

Eric Zelikman reflects on his journey from Stanford to his current venture, Humansend. He explains his longstanding interest in unleashing human potential through technology, initially believing automation could free people to pursue their passions, but realizing the complexity of understanding human goals.

Zelikman discusses early AI models like chain-of-thought prompting and STAR, which aimed to improve model reasoning capabilities. He shares an experiment involving n-digit arithmetic, highlighting the surprising scalability of these models without an obvious performance plateau.

He elaborates on improvements made in the Quiet Star paper, which showcased the potential for scaling models using pre-training style data and introduced key enhancements to the STAR algorithm.

Touching on his tenure at XAI, Zelikman explains the varying intelligence of models, noting their proficiency in solving complex problems when posed correctly, but acknowledging their struggles with emotional intelligence and understanding human objectives.

Zelikman argues for the importance of keeping humans in the loop in AI processes, cautioning against fully automated systems that diminish human agency and understanding. He believes AI should empower people to innovate and solve problems collaboratively.

Discussing his new company, Humansend, he outlines a vision for AI that better understands human objectives over time, enhancing productivity and personal interactions by focusing on emotional and interactive capabilities.

Zelikman highlights the need for AI models to consider long-term implications of their actions, moving beyond task-centric paradigms to support deeper, positive integration into people's lives.

Finally, he invites talented individuals to join his team, emphasizing the importance of building AI systems that truly empower and collaborate with humans for greater innovation and problem-solving.

Key Insights