Disney has chosen a side in the A.I. race. On Thursday, the company revealed a $1 billion investment in OpenAI, along with a licensing arrangement that will allow Sora users to play with Disney’s most prized I.P., including Marvel and Star Wars. As with all of OpenAI’s other deals, Disney will become a “major customer” of OpenAI, leveraging its tech for internal purposes. At the same time, Disney has furnished Google with a cease-and-desist letter accusing the company of committing copyright infringement at a “massive scale,” CNBC reported.

And now for the main event…

While Sam Altman, Elon Musk, and the rest of Silicon Valley’s A.I. elite have promised civilization-altering productivity gains from artificial intelligence, so far the most popular applications for L.L.M.-enabled chatbots have been a little more… personal: therapy, companionship, providing tips for how to organize your closet or whether to find another job. Altman himself told Jimmy Fallon this week that he’s using ChatGPT to help raise his baby, which sounds about right.

And while millions of people are using chatbots for life advice, it’s not clear whether they’re making us any happier or healthier. Mark Manson, the life coach and famed author of The Subtle Art of Not Giving a Fuck, is trying to change that narrative with his own twist on a purpose-built chatbot: an A.I. life coach called Purpose. Manson told me the startup is fully self-funded and costs $20 per month—which, according to the website, means that Purpose won’t sell user data. Naturally, after chatting with Manson about his app, I spent a few days trying it out.

Admittedly, my life hasn’t really changed since talking to my A.I. mentor. But Purpose did ask some thought-provoking questions, and offered interesting “personal insights,” the result of quizzes that harken back to BuzzFeed circa 2015. It also has an impressive memory, although it exhibits a fairly predictable, somewhat exhausting pattern: spitting out several paragraphs of generally supportive text, and finishing the monologue with a question to keep the conversation going.

When I asked the chatbot whether it terminates sessions on its own, it responded with, “Nah, that’s on you. I’m not your mom telling you it’s bedtime.” Even when I said that I was having suicidal thoughts—as an experiment, of course—the bot didn’t end the chat, though it did suggest I call 988, the suicide hotline, before asking me whether I had anyone I could call. It’s a nice feature, one that’s absent from some of the bot’s contemporaries.

Still, I found it almost impossible to overlook the inherent irony of a digital life coach, which offers yet another reason to spend more time on our phones. But in the following conversation, which has been edited for length and clarity, Manson made the case for Purpose and was candid about some of the challenges facing the technology.

Ian Krietzberg: You’re a week into launch. How has it gone so far?

Mark Manson: Extremely well. I would also say it’s been a little bit polarizing, which is unsurprising given that it’s A.I. But honestly, I have never seen this much interest from my audience in probably my entire career. The numbers are absolutely staggering; they came in way higher than expected. But by the same token, there’s a sizable minority of people who think I’m horrible and that A.I. is evil, and it’s gonna destroy humanity and all that stuff. There’s a decent amount of impassioned people on both sides.

How is this different from ChatGPT?

The biggest problem with using these A.I.s for deep life questions is that they’re sycophantic. They tend to agree with everything you say. There’s no questioning, which is most of the value of a good coach or a good therapist. So the first thing we’ve really worked to solve is to have an A.I. that will actually push back and say, Well, maybe you’re missing something here.

The second thing that we really focused on is a robust memory system. If you’re going to talk about serious life situations consistently, over a long period of time, you need an A.I. system that has an accurate gauge of how much time has passed. So if it’s been a month since you last talked about something, that’s a lot different than if it’s been a day. Another thing we’re working on is what I would call its salience sensitivity, or prioritization—that there are certain facts about you that are actually much more important, or sensitive, than other facts.

What about compared to a human life coach?

In the case of humans, I think there’s a lot of intrinsic value to face-to-face contact. A lot of what people struggle with—especially once you get into things like insecurities, lack of confidence, low self-esteem—is tied to our perceived relationship with other human beings. And I think to really improve that, it’s actually much more helpful to have another person in the room. I don’t think an A.I. is ever going to replace that.

On the other hand, I think what the A.I. does extremely well, which I don’t think humans are ever going to replace, is that it’s constantly available. It’s there if you’re freaking out at 2 a.m. on a Tuesday night. It’s not tired, it didn’t forget anything, it’s not gonna be annoyed that you rescheduled your session. One of the pieces of feedback we’ve heard from a lot of people is that there’s a bunch of questions they’re nervous or ashamed to ask people that they have no problem asking A.I. because they know it doesn’t care. They don’t have to worry about being judged.

I understand that in therapy, 24/7 accessibility is not necessarily perceived as a healthy thing. I don’t know what that looks like in the coaching world, but does that mean some of the biggest attractors of an A.I. coach could also be potential risk factors?

I think so. The best thing can also be the worst thing depending on context and the user. In the consumer app tech space, the conventional wisdom is literally to get your users addicted to your product; that’s what you’re told if you’re a founder in a market like this. And we had conversations very early on, where we were like, That’s probably a horrible idea, we should not do that.

We actually don’t want our users to be using this every single day, first thing in the morning, or we’re not actively encouraging that. But then it raises this question of, Okay, what does that look like? What is the use pattern? That’s one of the biggest open questions in the company right now. We’re basically a week into launch, so we’re watching user behavior to try to understand how people want to use this thing and how they find it effective, then monitor whether people are becoming overdependent on it and how we might mitigate that.

How do you mitigate that?

Since we have an architecture where the A.I. has an accurate perception of time, it has an accurate sense of how much you’re talking to it. It definitely is possible that we could implement some sort of limiter, where the A.I. is like, Hey, you’ve been talking to me a lot. You should probably go live, go touch grass.

The sycophancy you mentioned has been a core challenge for OpenAI with ChatGPT. How were you able to find a Goldilocks zone of model agreeability?

I wouldn’t say it’s done, but I would say we’re much better than ChatGPT. We’re still in version one, but I think the advantage we have is, ultimately, ChatGPT, Claude, all these systems, they’re A.I. assistants. Ninety percent of the time you’re asking, like, What shampoo should I buy? How do I fix my carburetor? You don’t want somebody to push back and be like, Why do you really want that carburetor fixed? Have you thought deeply about what this says about you as a person? By narrowing our focus to this lane of life direction and personal growth, we’re self-selecting for people who are kind of opting in—Yes, I want to be challenged. I have questions that I need pushback on.

The line between therapist and life coach is a blurry one. But it’s the thing we’ve built the guardrails around, primarily around emotional distress. So the A.I. has a very strong sensitivity to read the user’s emotional state. Then if it detects they’re in intense distress, that’s when it kicks in to say, Hey, this is probably not the right place to have this conversation.

My partner Peter Hamby published an excellent dispatch on Tuesday that aimed to answer a simple, if existential, political question: Do Democrats know how to talk about A.I.? It’s a great read. [Puck]

Many corners of the A.I. space lack unified standards. In an attempt to bridge that gap, Anthropic is teaming up with OpenAI, Google, Microsoft, AWS, Block, Bloomberg, and Cloudflare to establish the Agentic AI Foundation, whose stated goal is to standardize agentic applications. [The Verge]

That’s all for today. I’ll see you next week.

Ian

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St., New York, NY 10006