Why AI’s Next Leap Forward Is ‘Long Thinking’
14:19 Share

You want a straightforward path to your goals, but at Merrill, we know things may get in the way.

Or new opportunities can put you at a crossroads. With the bull at your back, you get a personalized plan and a clear path forward. Go to ml.com slash bullish to learn more. Merrill, a Bank of America company. What would you like the power to do? Investing involves risk. Merrill Lynch Pierce Fenner & Smith Incorporated. Registered broker dealer. Registered investment advisor. Member SIPC. A wholly owned subsidiary of Bank of America Corp.

Welcome to Tech News Briefing. It's Friday, December 13th. I'm Belle Lynn for The Wall Street Journal. Google's 15-year-old bet that cars of the future will drive themselves might finally be paying off. We'll find out why Waymo, the driverless car company owned by tech giant Alphabet, has even more at stake as one of its biggest competitors shuts down. And then it's time to get ready for AI that can...

For a long time, AI models that are designed to take more time to think over the results they generate for us are coming. Our Enterprise Technology Bureau Chief and columnist Stephen Rosenbusch tells us what that means for the chatbots we talked to.

But first, General Motors this week announced that it is shutting down Cruise, its robotaxi program, citing competition, time, and costs needed to scale the business. But that's good news for Waymo, the driverless car unit under Google parent company Alphabet. Our own Dani Lewis has been reporting on this as part of a special WSJ podcast series called Driverless, Waymo and the Robotaxi Race. You can listen to part one right here in the tech news briefing feed.

Danny, General Motors announced that it is slamming the brakes on its driverless car service, Cruise. What does GM's move mean for Waymo? This is a really big opportunity for Waymo, and it widens their lead in the robo-taxi industry even more than they already had been.

Which was also, in a lot of ways, thanks to Cruise. Cruise was one of Waymo's biggest competitors for a very long time. But Cruise ran into some pretty serious trouble last year. In October 2023, there was an incident where a person who had been crossing the street in San Francisco was hit by a car being driven by a human.

But she ended up in the path of one of Cruise's driverless cars. It struck her and she ended up being stuck underneath the vehicle. And Cruise ended up having their permits to run a robo-taxi business in California pulled. They've been working really hard over the last year or so to try and restart paid robo-taxi services in San Francisco and a few other cities around the country.

But it seems like it just got too expensive for General Motors and they've decided to pull the plug. So Waymo kind of already had a pretty immense lead, as you said. But now with Cruise dropping out and their other competitors operating in other parts of the space, the road is clear for Waymo. It's just...

Accelerator straight ahead. Let's back up a little bit, though. Waymo had been seeing success in its ridership increasing in San Francisco. Do you think it can repeat its success elsewhere in the country?

That's Waymo's big question right now. They've expanded their operations to paid rides through all of Los Angeles. They've announced a partnership with Uber where they're going to be launching Waymo services on Uber's app in Austin, Texas and Atlanta, Georgia. And recently, Waymo announced that it's going to be bringing its Waymo One robo-taxi service to Miami, Florida as well in 2026. And it's going to be a great opportunity for them to be able to launch their services on Uber.

And it seems like one thing that's notable is that all those metropolitan areas also have large groups of techies, right? Just like San Francisco, the home of Silicon Valley and the Bay Area. So we'll see if more techies in other places want to ride in robo-taxis, right? They definitely seem to be hoping that going to places where maybe there are more people who are curious about riding in a driverless car and may be inclined to take one. I spoke to an industry analyst. Her name is Shweta Kajuria. She's from Wolf Research.

And she told me that there's basically no one else who can do this kind of service right now. There really isn't any other competitor in autonomous. What else are consumers today going to be testing? There isn't anything else. And so that also allows Waymo the first mover advantage. Yeah. But surely they have some roadblocks on this path to crazy success. One of the big ones is, can they make money?

Waymo and a lot of these companies say that it's cheaper to operate because you're not paying a driver. But driverless cars have a lot of extra costs associated with them as well. There's the cost of the technology. There's the cost of the sensors that they need to see the world around them. There's the cost of the computers that they need to run the programs that drive the cars.

You know, and that's in addition to all the normal things of maintaining the cars, making sure they're clean, making sure they're in good condition. At the moment, analysts have said that most Waymo rides cost a little bit more than a regular cab or ride hail might cost a person.

There's also questions about safety, whether they can maintain their pace of expansion and avoid bad accidents down the line. That was WSJ's Dani Lewis. And you can check out part two of Dani's special series, Driverless, Waymo and the Robotaxi Race, this Sunday, right here in the Tech News Briefing feed. Coming up, it turns out that AI that can literally think for a long time might produce better answers than AI which doesn't.

what the ability for long thinking actually means and how it could affect our use of AI after the break. Say this is your financial life. Over time, things can get more complex with a personalized plan.

Merrill can help you navigate it all. Learn more at ml.com slash bullish. Merrill, a Bank of America company. What would you like the power to do? Investing involves risk. Merrill Lynch, Pierce, Fenner & Smith, Inc., register broker, dealer, register investment advisor, member SIPC, a wholly owned subsidiary of Bank of America Corp.

Artificial intelligence is just like us. Sort of. Scientists and researchers are working on developing AI's long-thinking capabilities. Quite literally, the ability for the technology to think for a longer period of time. That may just be AI's next leap forward as AI companies and researchers race to build up more computing firepower and improve the intelligence of their AI models.

While this advancement is still in its early stages, some experts say it's on track to improve significantly very soon. For more on why we need AI to think for longer and why that's probably a good thing, we're joined by WSJ Enterprise Tech Bureau Chief and Columnist Stephen Rosenbush.

Stephen, our listeners might be familiar with the idea that AIs can reason, but what's the idea behind long thinking? Well, we've all been blown away over the last few years by the incredible speed with which AI tackles all sorts of problems. But the next stage of AI may be even more powerful, and as it's allowed to take even more time to solve problems,

more complex problems. And it draws on the system of thought or inspired by the system of thought that psychologists refer to as system two. System one is quick, instantaneous, doesn't have to work very hard. It's almost instinctive. And that's sort of where most generative AI is located right now. But there's this shift

underway at OpenAI and underway at NVIDIA and other tech companies to embrace this sort of system two model of thinking. And we've seen this with what's known as the O1 series of models at OpenAI, but NVIDIA is advancing this research, Salesforce is advancing this research. So it's

beginning to enter the market in force. Okay, Stephen. So it sounds like system two thinking allocates attention to these effortful mental activities like complex computations. But what about AGI, the idea of artificial general intelligence? Does that move us into system three thinking?

I don't know that we have a system three yet, but when I spoke to Srinivas Narayanan, the engineer at OpenAI about the O1 series models, he did say that the O1 series models are going to lead to the release of what OpenAI refers to as AI agents. The agents are in turn a step along the path toward some sort of

AGI or think human-like broad-based reasoning. Let's talk about the benefits of long thinking. What does that really mean for these AI models or chat GPT like chatbots for them to be able to think in the long term or long think? You may have noticed that the AI that we now have and presumably use is

will make mistakes. It will hallucinate. And also, if you're really, really maxing out AI models today, you do hit a wall. Like there are really complex processes

problems in science, math, that AI that we're working with right now isn't so great at. And the O1 models developed by OpenAI can take more time to solve more complex problems in areas such as math, coding, and science that OpenAI says its earlier models weren't as adept at.

So in those areas, we could see a noticeable improvement in the output from these models. That is the idea. The AI that we're working with right now tends to double down on a strategy and produce an answer because that's what it must do. And that's one reason why you end up hallucinations.

The 01 models have the capacity to step back and say, maybe this isn't the best approach. Let me try another approach to solve the problem. And that's why it's able to take more time and it's able to reason a little bit more. There's a certain self-critical element that's been introduced.

What happens when these AI models think for a long time? How much longer do we have to wait for a response? That's an interesting problem. As NVIDIA CEO Jensen Huang said earlier this year, people in general are applying AI to problems that can take 100 days to solve.

to address or solve. And you think about what certain models can do in the blink of an eye. Think about all of that computing power trained on a problem for 100 days. You're able to attack much more complicated problems

such as predicting the weather everywhere in the world all the time, all at once, within a square kilometer. You're able to attack problems in genetics, as Katherine Brownstein, a researcher at Harvard Medical School, told me. You're able to maybe make more advances in personalized medicine.

Are there any concerns around long thinking for AI models? As in every discussion of AI, there's a need to think about how the technology is put to use, what guardrails are in place. Yes, you can apply this technology to solve all sorts of important societal problems, but you could also, in theory, use it to create all sorts of societal problems. And we need to think about

How we cope with that and at what level? Is this something that we really want to continue to happen at the company level or the developer level? Do we want greater public oversight? All those questions need to be thought through and we should probably be decisive, intentional. That was our Enterprise Tech Bureau Chief and Columnist, Stephen Rosenbush.

And that's it for Tech News Briefing. Today's show was produced by Julie Chang. Logging off for the weekend, I'm your host, Belle Lin. Additional support this week from Pierre, Bionna Mae, and Danny Lewis. Jessica Fenton and Michael LaValle wrote our theme music.

Our supervising producer is Catherine Milsop. Our development producer is Aisha Al-Muslim. Scott Salloway and Chris Sinsley are the deputy editors. And Falana Patterson is the Wall Street Journal's head of news audio. We'll sign back in this afternoon with TNB Tech Minute. Thanks for listening. Say this is your financial life. Over time, things can get more complex. With a personalized plan,

Merrill can help you navigate it all. Learn more at ml.com slash bullish. Merrill, a Bank of America company. What would you like the power to do? Investing involves risk. Merrill Lynch, Pierce, Fenner & Smith, Inc., Registered Broker, Dealer, Registered Investment Advisor, Member SIPC, a wholly owned subsidiary of Bank of America Corp.