Personal Agents Will Unlock Cheap Drug Discovery | OpenAI

Houda Nait El Barj (0:00)

The interesting thing with agents in interpersonal relationship is that you can bring it into a space where both of you are interacting with it. I use Instinct a lot with some of my best friends to try and plan things. And it's really interesting to have this third person who's mediating your life together. There are some hard questions around what context is okay to be shared in those spaces. If you put it in a group thread with, I don't know, your partner, it shouldn't just reveal all the secrets, you've been telling it in your personal thread.

But then over time, I think we're going to get much better at handling this multi-user scenarios. But I don't think we're there yet.

Theo (0:43)

We're live with Houda Nait El Barj from OpenAI, who leads research on emerging AI systems and new forms of human AI interaction, including some agent stuff. Personal agents have been a big topic over the last month. We've seen Instinct raising at $10 billion pre-revenue. We've seen Muse increasing Meta's market cap. I think our stock went up 12% yesterday, mostly just from Muse.

So clearly, personal agents are getting quite valuable. Certainly the labs are working on things in this vein as well. Houda, welcome to MTS.

Houda Nait El Barj (1:16)

Thank you for having me, Theo.

Theo (1:18)

So tell us about why you're so excited about personal agents. What capabilities do you think they'll unlock that people's normal chat GBT and Codex can do right now?

Houda Nait El Barj (1:30)

I think there are three different axes along which they can really unlock new capabilities and functionalities for people in their life. The first one that we think about right away is just, it frees up your time to do things that are more meaningful to you. Like, I don't think I should be clicking for hours on a website to buy an item or to do some groceries and that was always the vision like a really long time ago. But instead, you can use that time to be with your friends or to do things that matter to you like take pictures or whatever.

I think the second axis is also about the fact that they help you do things that you couldn't do in the past. So for example, I love to take photographs. But I don't necessarily know how to put together a website or find the right people who would be interested in buying my photography. And today with an agent, I just don't need to think about that anymore. I can just talk with it and say, hey, I would love to figure out how to sell these pictures that I've been taking. Can you just figure out how to do this for me? And all of a sudden, it really just expands the set of things that people can do without them having to go through the logistics or needing to have a team to do that for them or spending even money on this.

So it just expands all the things you could have done in the past. And I think one final thing that I'm really excited about is that it can really help you become who you want to be because it can keep track of all your context, of the things that really matter to you, of your goals. And you can figure out a path for you to get there. Like ChatGBT and most chatbots and assistants today are turn-based. But as soon as you start having more persistent agents that can keep track of long-horizon task and even be proactive, you move into a very different paradigm where it starts to be a sort of entity in your space.

Theo (3:35)

Right.

Do you think agents should be primarily aligned to the user or primarily aligned to a model spec? To what extent should they do what the user asks them to, even if it's not something that most people would want?

Houda Nait El Barj (3:52)

I think that's a good question. I think there's no denying that the agent should be aligned with a model spec originally. But I think the really hard thing for researchers and product manager is to make sure that at the end of the day, the model is aligned with what the person wants and never infringes on that. And it's a really hard personalization boundary to keep because what matters to you and what matters to me are very different. And part of building responsible AI is about ensuring that the agent figures this out. And it becomes even trickier when you move into proactive environment because, like, a very simple example.