@cjziems: We're going live in 30 minutes, and we'd love to have you join Joined by @dorazhao9 and @Diyi_Yang, I'll be talking abo…

We’re going live in 30 minutes, and we’d love to have you join Joined by @dorazhao9 and @Diyi_Yang, I’ll be talking about our new paper and discussing promising future directions for more human-centered LLM design! Stream it here: https://youtube.com/live/2d49pMXiJOA?si=xlrvDMVn4LaGoW8c…

TL;DR: Augmented Mind podcast live stream invites Dora and Caleb to discuss their co-led human-centered LLM report, emphasizing that AI should go beyond capability benchmarks and focus on human flourishing and long-term well-being.

Live Stream Background & Podcast Introduction

This live stream is the first attempt of the Augmented Mind podcast, adopting a real-time format to share work beyond regular episodes with the audience. The hosts include PhD students from Stanford and MIT: Michael, Shannon, and Eugene (Eugene leads the questioning in the subsequent conversation). The podcast launched last October, with the first episode released earlier this year, aiming to explore AI from two perspectives: the misconception within the community that human-centered work is non-technical, and external concerns that AI will lead to mass unemployment. The team hopes this platform will demonstrate that human-centered AI can be highly technical and positively impact society.

The Philosophy of Human-Centered AI

In the opening video, the team emphasized that while AI has leapt from simple chat to controlling computers and writing code over the past three years, the real challenge lies in building machines that collaborate with humans, rather than just pursuing stronger models. The human-centered North Star is benevolence toward individual humans, with core questions: What does ideal collaboration look like? How can we develop AI that augments minds rather than replaces them? These inquiries drive every episode of the podcast.

Guest Introductions & Personal Work

Caleb Ziems (Stanford NLP Group)

Caleb is about to start his final year of his PhD, advised by Diyi Yang (? but not explicitly stated in the transcript). His research focuses on evaluating social intelligence and cultural competence in large language models, with a particular emphasis on human-AI collaborative annotation. He mentions a recent project “Culture Mapping”: to address the limitations of past evaluations that only used static questions to compare human and LLM responses, he proposes a mixed-initiative approach—using LLMs to identify long-tail knowledge with low confidence, then having human experts supplement salient knowledge, thereby more reliably assessing the model’s cultural capabilities.

Dora Zhao (Stanford HCI & NLP Intersection)

Dora is a third-year PhD student advised by Diyi Yang and Michael Bernstein. Her research focuses on designing AI systems for long-term human benefit, operating on two levels: first, measuring the actual impact of AI on individual psychosocial indicators (e.g., AI companionship and well-being) and societal dimensions (e.g., labor market changes); second, building new systems that promote long-term user benefits. She introduces her recent work “Behavior Profiling,” a method that infers user motivations from unstructured interaction data, thereby creating agents that more proactively address users’ latent needs.

Paper Project: Reflections & New Directions for Human-Centered LLMs

Project Origin & Scale

The paper originated from a course project, initially conceived by advisor Diyi Yang, with Dora and Caleb serving as co-first authors for organization and editing. The project brought together over 50 student authors, culminating in a comprehensive report of over 60 pages, titled “Reflections and New Directions for Human-Centered Large Language Models.” Caleb emphasizes that this is not a mere literature survey but an active roadmap looking to the future.

Core Idea: From Capability to Human Flourishing

Caleb notes that current AI development is primarily driven by technical benchmarks and capability milestones, such as increasingly powerful models each year (e.g., showing improvements in loss functions). But from a human-centered perspective, the question is not just “What can AI do?” but “What kind of world can we build with AI?” The paper aims to help understand whether AI creates pathways for human flourishing, enhancing both productivity and human well-being. Dora adds that they hope to shift the conversation from a technical orientation to a social one, making AI truly serve humanity.

Future Directions & Q&A Preview

Host Eugene indicates that the second half of the live stream will adopt a more open format to answer audience questions, with advisor Diyi Yang joining for a Q&A in the final 20 minutes. Dora and Caleb encourage viewers to ask questions in the chat about the report or their personal research. Eugene also teases a special year-end episode and the next episode featuring Lean founder Jeremy Avigad discussing AI’s impact on mathematics.

Source: https://www.youtube.com/live/2d49pMXiJOA

Augmented Mind Podcast (@augmind_fm): Join us this Thursday May 28th 6-7pm PST for our first ever AM Podcast Live Stream! 🎉

We are hosting @cjziems, @dorazhao9, and @Diyi_Yang for a discussion on their new paper “Reflections and New Directions for Human-Centered Large Language Models”!

RSVP 🔗⬇️