Eric Zeltman

Founder of Human's End

Claims

As speaker

Mentioned in

Key positions and views

The primary goal of AI should be to empower people and grow the overall economy, not to replace humans and reallocate existing segments of GDP.

The most critical missing capability in current AI models is an understanding of the long-term implications of their actions and statements, a core component of emotional intelligence (EQ).

The AI industry's over-reliance on task-centric training paradigms is a fundamental flaw that prevents models from developing deeper user understanding and discourages investment in crucial features like memory.

Most AI labs view the human as an intermediate step to be automated away, rather than as a permanent and essential collaborator for AI systems.

Advanced reasoning algorithms like Star demonstrate that models can learn to solve progressively harder problems through iterative self-improvement, indicating that current reasoning capabilities have not yet plateaued.

Podcast consensus on Zeltman

Points of consensus

▶Eric Zeltman consistently argues that the prevailing AI development paradigm is flawed because it is overly task-centric, which hinders the development of crucial capabilities like memory and a deeper understanding of users.Mar 2026

▶He repeatedly emphasizes a philosophical distinction between AI for empowerment versus AI for replacement, believing the former grows the economic pie while the latter merely reallocates it.

▶Multiple claims highlight his belief that current AI models are fundamentally limited by their lack of emotional intelligence (EQ) and an inability to grasp what humans value or the long-term implications of their actions.Mar 2026

▶Zeltman's work and commentary on algorithms like Star and QuietStar consistently point to his belief that there are still significant, unexplored avenues for improving AI reasoning and problem-solving capabilities.Mar 2026

Points of debate

▶While Zeltman is founding a company, Human's End, to focus on AI's 'EQ', he also states there are still 'meaningful, unexplored dimensions for scaling the IQ,' suggesting a potential tension between his commercial focus and his broader research interests.Mar 2026

▶He critiques the industry's focus on removing humans from the loop to simplify scaling, yet his own work on reasoning algorithms like Star focuses on enabling models to solve problems autonomously with less human intervention.Mar 2026

▶Zeltman identifies the product trade-off between performance and latency as a key reason for AI failures in coding, which presents a potential internal conflict for his own company that will need to build products with both high EQ and practical performance.Mar 2026

▶He notes that task-centric benchmarks persist for organizational reasons like 'credit assignment' in large labs, which contrasts with his technical critique that these benchmarks are a key reason models lack deeper user understanding, highlighting a conflict between organizational reality and technical ideals.Mar 2026

Key themes

▶AI for Empowerment vs. ReplacementMar 2026

Zeltman articulates a core philosophical divide in AI development. He contrasts models designed to replace human labor, which he believes only reallocate existing GDP, with models designed to empower people, which he argues can grow the entire economic pie by augmenting human capabilities.

Investors should note that Zeltman's new venture, Human's End, is founded on this 'empowerment' thesis, suggesting its product strategy will likely focus on collaborative tools rather than fully automated solutions.

▶Critique of the Task-Centric ParadigmMar 2026

Zeltman argues that the AI industry's reliance on task-centric training and benchmarks is a primary obstacle to progress. He claims this approach leads to a neglect of crucial features like memory and prevents models from being more deeply integrated into users' lives, as labs optimize for metrics that simplify credit assignment rather than genuine user understanding.

This critique signals a potential market opportunity for companies that can develop and successfully evangelize new, more holistic evaluation metrics for AI that capture long-term utility and user collaboration.

▶The Next Frontier: AI Emotional Intelligence (EQ)Mar 2026

A central theme is Zeltman's belief that the most significant missing capability in AI is not raw intelligence (IQ) but emotional and interactive intelligence (EQ). He defines this as understanding human values, what people care about, and the long-term implications of actions, which is the stated focus of his new company, Human's End.

Zeltman is betting that 'EQ' will be a key differentiator and value driver for the next wave of AI applications, moving beyond pure task completion to become more integrated personal and professional partners.

▶Advancing AI ReasoningMar 2026

Drawing on his work with algorithms like Star and QuietStar, Zeltman discusses the mechanics of improving AI reasoning. He explains how iterative learning from successful attempts allows models to solve progressively harder problems, demonstrating that there are still significant gains to be made in the core 'IQ' capabilities of models, even without obvious plateaus.

Analysts should track developments in these self-improvement and reasoning algorithms, as they represent a key vector for capability scaling that is complementary to simply increasing model size or data volume.

Source episodes

Sentiment over time

Not enough data for timeline

Changes over time

Prior to XAI

Eric Zeltman was affiliated with Stanford University.

During XAI Tenure

Worked at XAI, contributing to pre-training data for Grok 2, the reasoning recipe for Grok 3, and tool use/agentic infrastructure for Grok 4.

Recent

Departed from XAI and founded a new company, Human's End, to focus on developing the emotional and interactive capabilities ('EQ') of AI models.

Present

Human's End is actively hiring for key roles, including infrastructure engineers, researchers focused on memory, and product-focused individuals.

Suggested prompts

How does Zeltman's focus on 'EQ' at Human's End reconcile with his stated belief that there are still meaningful, unexplored dimensions for scaling AI's 'IQ'? &nearr;What specific product trade-offs might Human's End face in developing AI with high emotional intelligence, similar to the performance vs. latency trade-off Zeltman identified in coding models? &nearr;If task-centric benchmarks are entrenched due to 'credit assignment' in large labs, what alternative metrics could Zeltman propose to measure progress in AI's 'EQ' and long-term understanding? &nearr;How might the 'Star' algorithm's iterative learning process be adapted to teach a model about subjective human values and long-term implications, which often lack verifiably correct answers? &nearr;

Key concepts

AI Model Limitations 1 ep AI Reasoning Algorithms (Star/QuietStar) 1 ep Human-AI Collaboration 1 ep Task-Centric Benchmarking 1 ep AI Emotional Intelligence (EQ) 1 ep Human's End 1 ep AI Scaling 1 ep XAI / Grok 1 ep

Notable quotes

“The most fundamental thing is that the models kind of don't understand the long-term implications of the things that they do and say.”

ERIC ZELKMANN · No Priors Ep. 135 | With Humans& Founder Eric Zelikman

“But I actually think if you kind of empower people, if you have models that really understand what people are trying to accomplish and really support them in accomplishing those things, you have the potential to actually grow that pie instead of basically replacing all of those segments.”

ERIC ZELKMANN · No Priors Ep. 135 | With Humans& Founder Eric Zelikman

“for most labs, the human is kind of the intermediate until you have this fully automated system.”

ERIC ZELKMANN · No Priors Ep. 135 | With Humans& Founder Eric Zelikman

Report last updated: Apr 21, 2026

Get started free

Back to Entities Intelligence Report