What’s Ahead for the AI Industry in 2025

[00:00:01] Adobe Express is the quick and easy create anything app. It has taken my content to the next level, making it simple to create everything I need to promote my business. From Instagram and TikTok posts to logos and flyers, all in just a few clicks. Build your brand with Adobe Express. Search for Adobe Express to find out more and get the app for free. Adobe Express, the quick and easy create anything app.

[00:00:33] Welcome to Tech News Briefing. It's Thursday, January 2nd. Happy New Year. I'm Danny Lewis for The Wall Street Journal.

[00:00:41] We're hearing from our reporters and columnists about some of the biggest companies, trends and people in tech and what could be in store for 2025.

[00:00:49] And perhaps the biggest story in tech is what new developments the artificial intelligence industry has planned for the year ahead.

[00:00:57] 2024 saw the integration of AI into even more everyday tools for people and businesses.

[00:01:02] And while the AI industry could have an ally in power when President-elect Donald Trump returns to the White House in a few weeks,

[00:01:09] AI developers are also facing new challenges in the hunt for more data to train their models.

[00:01:14] What will this mean for the science and business of AI?

[00:01:18] WSJ reporter Deepa Sitaraman joins us to discuss.

[00:01:21] A quick note before we get into the conversation.

[00:01:24] News Corp, owner of The Wall Street Journal, has a content licensing partnership with OpenAI.

[00:01:31] Deepa, what do you think were the most important developments in AI in 2024?

[00:01:36] That is such a hard question.

[00:01:38] A lot happened.

[00:01:39] It is simultaneously a time of like incredible optimism, maybe even like peak AI bubble behavior,

[00:01:46] but also a moment where reality set in for other companies.

[00:01:52] So we saw a really big crop of new companies that were going to do amazing things, right?

[00:01:58] We had a company called Character AI that came out from one of the top research scientists in the world,

[00:02:03] a guy named Noam Shazir.

[00:02:05] We had Inflection come out.

[00:02:07] That was LinkedIn co-founder Reid Hoffman and former Google executive and DeepMind founder Mustafa Suleiman.

[00:02:13] And we had a bunch of these companies then that were trying to vie for that open AI space.

[00:02:18] And a lot of those companies simply stopped.

[00:02:22] Both of those companies were, they weren't bought, but they were absorbed.

[00:02:27] You know, in Character's case, Google, and in Inflection's case, Microsoft.

[00:02:31] But like that showed, they weren't able to show enough progress over a given period of time,

[00:02:37] but they were still talented enough that other companies wanted them.

[00:02:41] And so they were able to engineer these not acquisition kind of deals.

[00:02:47] And that's an indication that they couldn't make it out on their own either.

[00:02:50] So there is a little bit of reality set in there.

[00:02:52] The other thing that we've been seeing a lot of lately is this idea that you can't just scale,

[00:02:57] get bigger and bigger models and get more unexpected and incredible performances.

[00:03:03] And so there is a real range of talent, but nobody can actually crack that specific problem.

[00:03:13] What is it that we need to make people surprised and startled again,

[00:03:19] the way they did when ChatGPT came out?

[00:03:21] At the same time, more and more money is pouring into these companies.

[00:03:26] They're in startups.

[00:03:28] There's this Bernstein research report that came out several months ago

[00:03:31] that estimates that all the largest tech players will spend about a trillion dollars on AI projects.

[00:03:39] So that's infrastructure, data, scaling, GPUs, whatever, for over the next five years.

[00:03:47] That was 2024.

[00:03:49] Turning to 2025, what are some of the new developments that you're going to be watching for with AI?

[00:03:54] I want to see if the model developments are really possible and if we do get another generation of these models.

[00:04:04] The other thing that I'm going to be watching is agents.

[00:04:11] What is an AI agent?

[00:04:12] It's an AI tool that does stuff for you.

[00:04:16] It could look up the weather and the place that you're traveling to,

[00:04:20] or it could help book you a hotel or an appointment.

[00:04:23] It could just help you do stuff online.

[00:04:27] For that to work, you need these systems to be much more reliable than they appear to be now.

[00:04:34] But every major company is talking up the possibilities for agentic AI

[00:04:41] and actually having these things be useful to you in the real world.

[00:04:46] The other thing that people are talking about going forward is bringing AI into other forms of media.

[00:04:53] So you'll probably see more AI video creation, like you see with Sora, OpenAI's video model.

[00:04:59] And you'll probably see more realistic images.

[00:05:03] And then also AI voice.

[00:05:06] These voices could get more and more realistic.

[00:05:08] And you'll see these companies continue to spend so much money on developing AI.

[00:05:16] You'll see more data centers being built around the country.

[00:05:18] It's just going to continue that persistent march into every aspect of your life.

[00:05:25] That is the vision, anyway, for a lot of these companies.

[00:05:29] Coming up, AI has only been around in its current form for about two years.

[00:05:34] But researchers are already running out of training data for their models.

[00:05:39] What does that mean for the next generation of artificial intelligence?

[00:05:42] That's after the break.

[00:05:54] Deepa, we've talked about how the lack of training data has become a really big problem for AI companies.

[00:06:00] OpenAI delayed its next update to the model behind its chatbot, ChatGPT,

[00:06:05] in part because of a shortage of training data.

[00:06:07] If AI companies are already running out of this data just two years into the AI boom,

[00:06:13] what does this mean for the future of large language models?

[00:06:16] I mean, that's the question.

[00:06:19] And if you're asking me about another thing that I'm going to look for in 2025,

[00:06:23] it's how the different model builders are going to respond to just that problem.

[00:06:29] Because it's time for new ways of thinking about growing these models.

[00:06:34] And it is very unclear if that kind of growth and those advances can happen on the commercial clock.

[00:06:43] It can take decades for breakthroughs to appear in science.

[00:06:47] And AI is definitely a science.

[00:06:49] But these scientists are now working on sort of corporate America's timeline,

[00:06:54] which is like faster and moving forward and aggressive.

[00:06:57] And there's a lot of expectations.

[00:06:59] Can you actually get to a breakthrough, a research breakthrough of the magnitude you might need

[00:07:07] on that timeline with that kind of ticking clock behind you?

[00:07:10] That's another question that is going to be, it's not going to be fully answered,

[00:07:14] but we're going to have a lot more clarity as to what the answers might look like.

[00:07:18] It might be motivating, but right now the pressure is quite different from what

[00:07:22] an academic researcher would be experiencing in a lab at a university.

[00:07:27] Who's taking the lead in this next stage of AI development?

[00:07:30] That is open question, right?

[00:07:33] Open AI is at the dead center of the AI movement.

[00:07:38] They are still what people think of when you think of AI.

[00:07:44] So there's going to be a lot of scrutiny on them.

[00:07:47] That's 100% sure.

[00:07:49] There's going to be a lot of scrutiny on them and their relationship with Microsoft.

[00:07:53] They are going to be kind of ground zero for what is happening in the industry.

[00:07:59] But in terms of actual model capabilities, like who will build the best model?

[00:08:04] I don't know that we know that.

[00:08:07] Because a lot of the other companies, Google, Anthropic in particular, have caught up.

[00:08:15] Like they're building models that are about as good as what OpenAI has.

[00:08:18] And sometimes people view them as better.

[00:08:22] And so there is a real open question around who will be the most used model or most enjoyed

[00:08:28] or liked model.

[00:08:30] What's the outlook for AI under President-elect Donald Trump's incoming administration?

[00:08:35] Trump has already named an AI czar, David Sachs, who's a prominent venture capitalist, co-founder of PayPal.

[00:08:43] And he is going to be running point on AI policy.

[00:08:48] It's a little unclear what the role will entail and the kinds of powers that he might have.

[00:08:54] But the fact that we have that position at all already tells you a lot about the way,

[00:08:58] you know, the White House is studying and looking at AI applications.

[00:09:02] One of Trump's big allies is Elon Musk, who's building his own AI company called XAI,

[00:09:10] which is a competitor to OpenAI, a company that Elon is suing.

[00:09:17] So there are a lot of different alliances and feuds within the top of this industry.

[00:09:24] But the way a lot of these companies appear to be navigating the interpersonal issues or just to secure their future is by building relationships with the U.S. military and companies that do business with the U.S. military.

[00:09:40] So OpenAI just reached a deal with Andruil.

[00:09:43] It's a defensive deal.

[00:09:44] They are not working on offensive war weapons, but they are working on this defensive side.

[00:09:51] And they're making the argument that they are a national security asset.

[00:09:56] And you see a lot of these companies do that.

[00:09:58] Palantir and Anthropic just reached a big deal.

[00:10:01] Well, everyone's framing it in a very similar way, which is AI is something that the U.S. should cultivate and nourish and make sure it succeeds,

[00:10:10] because this is an important tool to have in our holster, especially against other countries, notably China.

[00:10:20] And so we're going to see a lot of more politics get into the AI world.

[00:10:26] So that's another thing I'll be watching for.

[00:10:29] That was our reporter Deepa Sita Raman.

[00:10:31] And that's it for Tech News Briefing.

[00:10:33] Today's show was produced by Julie Chang with supervising producer Catherine Millsop.

[00:10:38] I'm Danny Lewis for The Wall Street Journal.

[00:10:40] We'll be back this afternoon with TNB Tech Minute.

[00:10:43] Thanks for listening.