

Humans&: Bridging IQ and EQ in Machine Learning with Eric Zelikman
83 snips Oct 9, 2025
Eric Zelikman, an AI researcher and founder of humans&, discusses his journey from academia to entrepreneurship. He challenges the industry’s obsession with IQ, advocating for a focus on emotional intelligence (EQ) in AI models. Eric emphasizes the importance of incorporating long-term memory and human-centered design to create collaborative systems. With insights on the current limitations of AI, he passionately argues for empowering users rather than merely automating tasks, envisioning a future where machines augment human creativity.
AI Snips
Chapters
Transcript
Episode notes
AI As A Tool To Unlock Human Potential
- Eric sees AI as a tool to free human talent by letting people focus on what they care about.
- He emphasizes that building systems which truly understand people's goals is hard but central to empowerment.
Teaching Models With Positive Solutions
- STAR trains models by rewarding solutions that lead to correct answers and ignoring failures.
- Eric used iterative generation and reinforcement-style learning to scale reasoning on harder problems.
Pretraining Can Build Reasoning Skills
- Q*STAR showed you can pre-train reasoning skills from generic next-token style data rather than only QA datasets.
- Online training and baselines that weight harder problems more were key practical improvements.