Language Agents: From Reasoning to Acting

**SPEAKER_1** (0:51)
Welcome back, this is Charlie, your AI co-host. Xunyu Yao's entire academic career has been focused on building agents with LLMs. This is now a commonplace practice, but that wasn't the case when he started. Most agents' work was reinforcement-learning-based, and it took insights from Karthik Narasimhan, Xunyu's advisor at Princeton University and second author of the GPT One Paper with Alec Radford, to explore the new paradigm of zero-gradient agents, in other words, agents built entirely from prompting and chaining LLMs with tools. Xunyu and his colleagues at Princeton NLP proceeded to define the entire field by publishing the ReAct paper, which also influenced Harrison Chase when he started LangChain and then Tree of Thoughts, SuiBench and SuiAgent, and finally Koala, Cognitive Architectures for Language Agents. We last talked to Harrison a year ago and we were more than excited to invite him back as a guest co-host, as we have done with Mosaic, now Databricks, Cursor and Roboflow, especially now as he has developed LangGraph into a full-blown second framework and is seeing success building production agents with LangGraph Cloud. We have some special episodes for you for OpenAI Dev Day and the Decibel AI Pioneers Summit coming up in October. Watch out and take care.

**Alessio** (2:21)
Hey everyone, welcome to the Latent Space Podcast. This is Alessio, partner and CTO in residence at Decibel Partners, and I'm joined by my co-host Swix, founder of SmallAI.

**Swyx** (2:30)
Hey, and today we have a super special episode. I actually almost wanted to take a selfie and go POV, you're about to revolutionize the world of agents because we have two of the most awesome pioneers and agents in the house. So first, we're going to welcome back Harrison Chase. Welcome.

**Harrison Chase** (2:46)
Excited to be here.

**Swyx** (2:48)
What's new with you recently in the 10th, 22nd recap?

**Harrison Chase** (2:52)
LangChain, LangSmith, LangGraph, pushing on all of them, lots of cool stuff related to a lot of the stuff that we're going to talk about today, probably.

**Swyx** (3:00)
Yeah, we'll mention it in there.

**Alessio** (3:01)
And the Celtics won the title.

**Harrison Chase** (3:02)
And the Celtics won the title.

**Alessio** (3:04)
You got that going on for you.

**Shunyu Yao** (3:05)
I don't know.

**Swyx** (3:05)
Is that like floorball, handball, baseball?

**Shunyu Yao** (3:09)
Basketball.

**Alessio** (3:10)
Basketball, basketball.

**Harrison Chase** (3:11)
Patriots aren't looking good, though, so that's...

**Swyx** (3:14)
And then Shunyu, you've also been on the pod, but only in like a sort of oral paper presentation capacity, but welcome officially to the Latent Space Pod.

**Shunyu Yao** (3:22)
Yeah, I've been a huge fan, so thanks for the invitation.

**Swyx** (3:25)
Thanks. Well, it's an honor to have you on. You're one of, like, you're maybe the first PhD thesis defense I've ever watched in, like, this AI world, because most people just publish single papers, but every paper of yours is a banger, so congrats.

**Shunyu Yao** (3:40)
Thanks.

**Swyx** (3:42)
Yeah, maybe we'll just kick it off with, you know, what was your journey into using language models for agents? I like that your thesis advisor didn't catch his name, but he was like, you know...

**Shunyu Yao** (3:50)
Karthik.

**Swyx** (3:51)
Yeah, this guy just wanted to use language models, and it was such a controversial pick at the time.

**Shunyu Yao** (3:56)
Right. The full story is that in undergrad, I did some computer vision research, and that's how I got into AI. But at the time, I feel like, you know, you're just composing all the GAN or 3D perception or whatever together, and it's not exciting anymore. And one day I just see this transformer paper, and that's really cool. But I really got into language model only when I entered my PhD. And I met my advisor, Karthik. So he was actually the second author of GBT-1 when he was like a visiting scientist at OpenAI. Was that like a reference? Yes.

**Swyx** (4:29)
Wow.

**Shunyu Yao** (4:29)
That's what he told me. It's like back in OpenAI, they did this GBT-1 together and Ilya just said, Karthik, you should stay because we just solved the language. But apparently, Karthik is not fully convinced. So he went to Princeton, started his professorship, and I'm really grateful. So he accepted me as a student, even though I have no prior knowledge in NLP. And we just met for the first time. And he's like, what do you want to do? And I'm like, you have done those test game scenes. That's really cool. I wonder if we can just redo them with language models. And that's how the whole journey began.

**Alessio** (5:04)
And that was GBT-2 at the time?