Pierre Bienname (3:23)

What kind of user base is this chatbot enjoying so far and what happens to users data?

Heidi Mitchell (3:28)

So it started out with just a few hundred young people ages 18 to 30 and that was just in America. And now it's being used by 60,000 people in 190 countries. Users can opt out of having their data used by the application, but it is anonymized, it's not used to train the model. So the way that the model works is it's pre trained on a large language model and then you fill out this quite lengthy survey and form and with specifics they're open ended questions and the more information you give it then the more accurate and the more specific the answers would be.

Pierre Bienname (4:03)

And is there a business model here for futureyou? Does it cost money the way some other chatbots do?

Heidi Mitchell (4:07)

There's no cost to using this. It's just a project right now. There's no plan to monetize it.

Pierre Bienname (4:14)

Heidi Mitchell is a Wall Street Journal contributor. Heidi, thank you so much for joining me in the present.

Heidi Mitchell (4:19)

Thank you. Future Heidi thanks you. And present Heidi thanks you too.

Pierre Bienname (4:23)

Coming up, WSJ senior personal tech columnist Joanna Stern is basically a pro when it comes to getting the most out of AI. She tells us which models lead the pack. That's after the break.

ADP Representative (4:38)

ADP imagines a world of work where smart machines become too smart.

Joanna Stern (4:43)

Copier.

Janet (4:43)

I need 15 copies of this print, by the way. Irregardless. Not a word, Janet.

Heidi Mitchell (4:50)

Yeah, I know.

Janet (4:52)

Page 6 should be regardless of or irrespective of, just print them please. If it were a word, Janet, it would mean without irregard, which is copier.

Heidi Mitchell (5:00)

Switch to silent mode.

Janet (5:02)

Let's put a pin in it.

ADP Representative (5:03)

Anything can change the world of work. From HR to payroll, ADP helps businesses take on the next anything.

Pierre Bienname (5:14)

The competition among AI bots is heating up more than two years after the release of OpenAI's ChatGPT. The alternatives abound and some of them require a subscription for the full suite of features. So which should you consider to help you write research and do digital errands? The Wall Street Journal's senior personal technology columnist Joanna Stern looked at three models, ChatGPT by OpenAI, but also Claude by Anthropic and DeepThink, R1 by the Chinese company DeepSeek. And she's here to tell us which of these came out on top. Joanna, what do you need an AI assistant for?

Joanna Stern (5:47)

First of all, why do you need one? What don't you need one for?

Pierre Bienname (5:50)

What are you using it for?

Joanna Stern (5:51)

All the things that you kind of know at work you're spending too much time doing, and it's not that intriguing, I'm starting to think can be given to an AI assistant. Creating spreadsheets, helping with research, creating calendars, all the types of things you might spend time on. But you're like, I don't think this is the best use of my time.

Pierre Bienname (6:12)

So what AI bots do you use typically?

Joanna Stern (6:14)

So over the Last few months, ChatGPT and Claude have become my go tos, and I'm paying both of those companies, OpenAI and Anthropic, $20 a month for their pro plans. But of course, over the last couple of weeks, everyone wants to talk about deepseek, the Chinese open source model. And so I added that in. I will say it is not as impressive in terms of features. It is a smart model and of course that is something to really consider. But what this piece really was about is that smarts are important, but these features, the ones that let you take advantage of those smarts, are a lot more important when you're considering which of these assistants or bots you want to pay for.

Pierre Bienname (6:52)

Okay, so you had those three contenders. How do they compare?

Joanna Stern (6:54)

You want to think about, am I getting the best model? And of course that is the race that is happening in the industry. And lots of them right now are talking about reasoning models. This is the idea that the model takes a little bit more time to think about its answer. And in the case of Deep seq, it even gives you kind of a play by play of how the model is thinking. It's kind of a like self aware, quirky bot that's like, maybe I should answer with this, maybe I should answer with that. And you see that. But actually OpenAI has had a reasoning model and it doesn't do that, but it just says thinking. And that's where I think these all are really going. To become more about features, because you're only going to require so much of that smarts if you're just a basic worker. We're not all trying to find the cure to cancer or solve the world's biggest issues at work. We're all trying to just kind of get some of our basic stuff done, whether that be writing or research or creating spreadsheets, managing email. And so that's where I think the features now come into play in a big way. And you want to look at what those features are. And that's what I tried to do in this piece, is really try to compare, which is the leading one, in terms of features and in terms of how you can actually utilize the smarts.

Pierre Bienname (8:03)

Can you give me some examples of what you used them for and which one maybe shined the most?

Joanna Stern (8:07)

So the reason I'm paying for two right now is that one of the models, Claude, is really good at writing, it's really good at research and projects. But then ChatGPT is decent at writing, but is really good at web access, memory and other types of features. So let's break down Claude for a second. One of the things I love about anthropics, Claude, is they have this projects feature where you can go in. I'm working on a book right now, so I have a big project that I'm working on and you can go in and you can upload lots of documents and information about a certain topic or project and then the bot really starts to know those things. So when you prompt it to say, oh, remind me about who I wanted to interview about that healthcare AI startup, it knows. It pulls that from the document and says, oh yeah, that person is blah, blah, blah. Here's some more information and background information about them. OpenAI with ChatGPT has added a similar feature, but I got really sucked into Claude on this.

Pierre Bienname (9:02)

So Claude, like any good assistant, is going to have some memory of, like, what you're doing, what you're working on, what you want to achieve.

Joanna Stern (9:07)

Well, memory is really interesting. So Claude has memory within that project feature, but if you're using it just in the normal interface, it doesn't have much memory. And that's a place that ChatGPT actually wins. They've given you the ability to put in information about yourself through something called custom instructions. You can tell it. For instance, ChatGPT knows I'm allergic to avocado. It knows I have two kids, it knows what I do, it knows I'm a technology journalist. But also along the way, as you're chatting with it, it picks up information and stores it in its mem. ChatGPT over time is starting to remember things about you, which makes its answers more helpful. Anthropic is working on that. But again, you do have that memory in the projects area.

Pierre Bienname (9:50)

And tell us more about this third one, Deepseek. It's kind of the AI model that's been really turning heads lately.

Joanna Stern (9:56)

Deepseek really I think you have to think about as the early days of ChatGPT in terms of features. It is really just a chatbot right now. What's special about it is that it's quite smart and it has this R1 reasoning model. But of course the big buzz about that in the industry hasn't necessarily been how smart it is. It's been has this company been able to create smarts with a fraction of the power and the costs that the American AI companies have. And so I have been using it to just ask questions and see the results. I should say with Deepseek, I do have some concerns about privacy and security of the Chinese ownership. So I've actually been using it via perplexity. And that company is hosting the model on US servers which don't have t to China.

Pierre Bienname (10:38)

That was our senior personal tech columnist Joanna Stern. And that's it for Tech News Briefing. Today's show is produced by Julie Chang. I'm your host Pierre Bierme. Additional support this week from Bell Lynn. Jessica Fenton and Michael Lavalle wrote our theme music. Our supervising producer is Kathryn Milsop. Our development producer is Ayesha Al Muslim. Scott Salloway and Chris Sinceley are our deputy editors. And Falana Patterson is the Wall Street Journal's head of news audio. We'll be back this afternoon with TNB Tech Minute. Thanks for listening.

ADP Representative (11:13)

ADP knows any big thing, any small thing, any trendy thing, even a trendy thing that everyone knows isn't a great idea but management just wants us to give it a try for a bit. Can change the world of work. From HR to payroll, ADP designs forward thinking solutions to take on the next anything.