Everyday AI Podcast – An AI and ChatGPT Podcast - EP 444: AI Market Forces Collide: Global Powers, Prices, and Problems (2025 AI Predictions - Vol. 2)

Starting point is 00:00:00 This is the Everyday AI Show, the Everyday Podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live in Adobe Firefly, the All In One Creative AI Studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. Thousands of hours of studying AI, hundreds of conversations for a couple,

Starting point is 00:00:55 dozen AI predictions for you. I get it. You don't have three, four, ten hours a day to keep up with what's going in AI. There's probably a lot of uneasy feelings and saying like, hey, this is. is going way too fast if only I had something to help me keep up and to cut through the nonsense. Well, that's what you have in everyday AI. And today, we are bringing you a special edition, Volume 2 of our 25 biggest and boldest AI predictions for 2025.

Starting point is 00:01:38 And unlike last year, why I jammed them all into one episode. that was entirely too long. This week, we are bringing them to you five at a time, five days, like I said, hundreds of conversations that I had about generative AI in 2024. And, you know, going back to literally, I remember in March when I partnered up with Nvidia at their GTC conference, I was already writing notes for this very show, for this very series so you can get ahead and grow your company. and grow your career. That sounds like what you're trying to do. This series and this show is for you.

Starting point is 00:02:18 Welcome. This is Everyday AI. My name is Jordan Wilson and this thing's for you. It is a daily live stream podcast, free daily newsletter helping us all learn and leverage generative AI to grow our companies and career. This is unscripted, unedited. I'm just bringing it to you real. All right, everyone else is trying to tell you what you need to hear or what you think you want to hear. I am telling you what you need to hear when it comes to AI. All right. Something else you need to hear, our website. You got to go there.

Starting point is 00:02:47 It's your everyday AI.com. Sign up for the free daily newsletter. If you're looking for the daily news, technically this one's pre-recorded because it's a lot of work. You know, it's technically like nighttime, but you're probably hearing it in the daytime. But live stream audience, I still want to hear from you. And let me just answer this. Like, okay, Jordan, why are you doing this? Why are you doing predictions, number one?

Starting point is 00:03:08 Number two, why are you drawing it out? Well, a couple of reasons. Number one, why am I doing these predictions? People ask me all the time. Jordan, what's next? Jordan, you talk to, you know, people at Microsoft and IBM and Google and, right? I have great, great relationships and conversations with people. Those, you know, I've interviewed hundreds of people in 2024, and I've had hundreds of

Starting point is 00:03:31 conversations that you didn't hear, right? Get things off the record, on the record, NDAs, non-disclosures, right? But now I can finally give it to you all in my predictions. And I did this last year. And I still had people reaching out to me like 10 months later being like, how did you know all that, right? You were correct. I didn't say those things, but you can go listen to the old 2024 episode out there and let me know for yourself.

Starting point is 00:03:57 But without further ado, let's jump back into it. Like I said, this is volume two of a five volume series. Happy Tuesday, if you're listening live. but you can go listen to yesterday's episode and join us in the coming days as we go through volumes three, four, and five. But to quickly set today's stage, today we're going to be covering a lot of different things, but I'm calling this group of five predictions, AI market forces collide, global powers, prices, and problems.

Starting point is 00:04:28 Yeah, you like that alliteration, don't you? All right. So let's jump into it. But I'll give you a quick recap of yesterday's five predictions. If you want to know more, go back, read, listen, whatever you want to do. We recap them in our newsletter yesterday as well. So 25, agent orchestrators will be a growing position. 24, public companies will post job for AI agents.

Starting point is 00:04:52 23, company reasoning data collection. 22, high-end professional services will go through a pricing crisis. And 21, UBI will become a household. conversation. All right. So today's volume two, here's our five predictions. We're going to dive in deep, but go pretty quick. I'm challenging myself to do these episodes in 25 minutes. You know, branding, 25 predictions across five days, five predictions each for a total of 25, trying to do these in 25 minutes. All right, here we go. Here's today's, and we're going to dive into them. Ready, 20. Open source surges. Open LLMs will temporarily overtake proprietary models.

Starting point is 00:05:33 19. Chinese AI will dominate and cause confusion. 18, perplexity will either pivot, get acquired, or get squashed. Yeah, hot takes. It's Tuesday, y'all. 17 API prices are going to drop like they're hot. And 16, embodied AI will be an exploding sector. All right. Live stream audience, like yesterday, I want to know which one of these are the most likely and which one is the least likely, right?

Starting point is 00:06:07 I'm going to revisit this in six months, but I want to see, do you know your stuff? Do you have a crystal ball, right? Do you know what's going on? Let me know. You can even just put what the number in which one you think is most likely, the number, what you think is least likely. Please go ahead and do that right now. I always love hearing from our live stream audience.

Starting point is 00:06:28 All right. Enough of the chit chat. Let's get into it. and talk about our first one, open source models. Here's the reality. Open source models haven't been fully tapped into yet. They started to really trend in the latter part of 2024. All right.

Starting point is 00:06:51 So there are fully open source models. Let me just get this piece out of the way right now, right? There's fully open source models that you can essentially, there's an open source standard. You can download them. You can fork them. you can do anything. And then there's also META's Lama models,

Starting point is 00:07:07 which are not technically open source. META calls them open source. They aren't open source by the technical open source standard, although what META is trying to do is they're just trying to redefine what open source is. But for a lot of intent and purposes, it is an open model, although not truly open source. But meta has been a huge driver in the movement nonetheless,

Starting point is 00:07:30 because y'all, you have big companies. like Nvidia forking and building off Meta-Lama's or sorry, meta-lama models. And they are topping, not topping, but they are in the top tier of rankings and of these charts that we'll be talking about. So why? Why the heck are open-source models going to temporarily overtake proprietary? So I kind of told you what open-source models are.

Starting point is 00:08:01 you can, for the most part, download them if you have a powerful computer. You know, there's smaller, you know, small language models that you can download. There's a lot of reasons why they're becoming very popular. Well, one is they're more secure, right? You don't have to send any of your information. You don't have to send any of your data to the cloud. If you're working with a smaller, smaller model, right? Maybe something that's only a billion or two parameters, it's very fast, right?

Starting point is 00:08:31 because you don't have that kind of extra inference time that you're waiting on a cloud provider to, you know, go and crunch your prompt. So security, it's fast. It's better for the environment too, right? So there's a lot of, you know, good, I guess, kind of under the radar reasons that I think is good to root for open source models, right? proprietary models like OpenAIs, GPT, Claude Anthropic, you know, there's plenty of others, right? Google Gemini, they're big models, right? These big companies, they're trying to, obviously, right, they want to profit off their models.

Starting point is 00:09:11 Everyone does, right? So you can't download them and fork them and build off them and distill other models with them. You can't. You have to log on to their website or use their API on the back end and pay for it, right? So not everything is open source. But I think open source in 2025, this is the year. I don't think an open source model is going to top the LM arena or the chatbot arena like we say. I don't think necessarily that we are going to see an open source model with the highest as an example, MMLU benchmark.

Starting point is 00:09:49 So, you know, I like to say the, and a lot of people say, but the MMLU. is a benchmark. It's like the ACT or the SAT for large language models. So I don't think that it's, but like when I say it's going to temporarily overtake, I'm not saying it's going to be the number one model for a certain benchmark or for ELO scores on the chat botterina. What I will say, what I will say, right? I'm just going to go ahead and put these out, put these out there, right? I do think that more than half of the top 20 models on the chatbot arena leaderboard, at some point in 2025, I'm going to say more than half of them are going to be open source. And you might be saying, okay, well, it sounds like there's only, yeah, there's more than a million

Starting point is 00:10:42 different open source models because people can download them, especially smaller ones. It's not terribly time consuming, right, to download one of these, fork them, do some fine-tuning, you know, kind of make a smaller model of your own. And that's what a lot of people do. And then they put it out there for everyone else to use. proprietary models, not so much. But there are like, I don't know, like 10 different versions of Google Gemini, right? There's, there's, you know, 01 preview, 01, mini, 01 full, 01 Pro. So there are dozens of very strong, very powerful proprietary models.

Starting point is 00:11:20 So it's not crazy to think that open source could take half of those. Also, recent study said 41% of enterprise companies plan to increase open source model usage. Companies know that there are huge benefits. I don't think that they were really paying attention in 2023 or the first part of 2024 until I do think meta made that first big open splash and said, yo, look at us because their models at the time, 3.1 and 3.2, I mean, these were top three, top five models. They were competing with proprietary models. So that is something, I think, when you crowdsource these fixes, right, because if you put your model out there for everyone

Starting point is 00:12:03 to look at, for everyone to download, for everyone to improve upon, they're going to find ways to optimize the models, to make them better, right, to improve the output. So it's essentially when you put your work out there and crowdsource people to build it, to pick it apart, to tear it apart, it's going to get better. And that compounding, kind of crowd sourcing of compute, right? We'll say that, triple C, right? The compounding crowd sourcing of compute. It's the rising sea, lift all lifts all ships or whatever the saying is, right?

Starting point is 00:12:39 All right. Number 19. Chinese AI will dominate and cause confusion. So speaking of, you know, models on the leaderboard, let's look at the top 10. I do think at some point in 2025. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI Assistant, now live in the Adobe Firefly app, the all-in-one creative AI

Starting point is 00:13:12 studio. Powered by Adobe's Creative Agent, Firefly AI Assistant lets you start with your vision, just describe what you want. and shape the outcome as it takes form with the assistant. The assistant orchestrates multi-step workflows, drawing on 60-plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations.

Starting point is 00:13:49 Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adobie.com. Five of those top 10 models are going to be from China. All right. So reportedly, Alibaba's Quen models have already been deployed by 90,000 enterprise customers. All right. Also, something that really started to cause a scene at the end of 2024, Deep Seeks v3 model.

Starting point is 00:14:33 All right. So, so, so, so cheap. And here's the thing. I don't think it's just large language models when it comes to, you know, Chinese AI will dominate and cause confusion more on that part here in a second. But I think robotics, China is their humanoid and their robotic production. It's scary. We've shared some of the videos, not like, oh, this is so cool. Look at this. It's nuttier than a squirrel on keto. No, it's scary, right? It's warehouses of what looks like thousands of humanoid robots that are embodied AI, right?

Starting point is 00:15:17 Yeah, it's kind of scary. So I think it's not just large language models. It's not just prices. It's not just benchmarks. It's robotics. I think we're going to see other things. I do think China has competed very well in the AI video sector with Kling and some others. But I also think that we're going to see an AI audio competitor as well, right?

Starting point is 00:15:44 So China kind of silently, right? So Deepseek v3, one of the Chinese. one of the cheapest models to use via the API, cling to AI video models, one of the best out there. I would say definitely top three, you know, the Quinn models and, you know, Quinn has a reasoner model. They're great. They're competing all across the board. Also, here's why it might cause some confusion. There's some people rumbling about this on the internet, and I'm not a conspiracy theorist.

Starting point is 00:16:17 don't think that about me, but I do think that there is an element of a kind of Trojan horse here. Right. So even with, you know, some of these models that are dirt cheap, you know, companies and individuals, they just kind of close their eyes, right? And they just assume like, oh, okay, it's like using Google. It's like using, you know, it's like using OpenAI. Oranthropic, it's not. You should probably start reading everyone out there, business owners, if you're a decision maker, making a big choice on what AI system to use, I'm not going to name names. Read the terms of services, read the privacy policies, understand what happens with your data if you're uploading it to these companies. I'll tell you this. I've read them all.

Starting point is 00:17:04 Yeah, I'm a dork. I actually read them. I read them. And I also have different AI systems help me read these terms of services. you don't want to know. It's a Trojan horse. That's what it is, right? It's everything's free, everything's cheap, everything's fast, everything's great,

Starting point is 00:17:19 and it's like at what cost? Well, we might find out. And I do think that there's going to be, I won't say an event, but I think there is going to be some mainstream media attention to this very fact as AI, I think is also going to get very political in 2025. And this is one of those things when we talk about geo, political tensions, yeah, AI is going to be a big part of that, right? A lot of restrictions on, you know, exporting, bands on exporting GPU chips to certain countries,

Starting point is 00:17:53 you know, potential tariffs, all of these things. The AI scene is going to get very political and keep an eye out for stories. I would say probably midway through the year once the rest of the world kind of looks and understands what's going on because I think these models, they're very capable, They're very popular. They're very cheap. They're very fast. People are going to use them.

Starting point is 00:18:13 Businesses are going to use them. And eventually, someone's going to be like, wait, should we have looked at this a little more closely? All right. Number 18. Here's a spicy hot take for Hot Take Tuesday, volume two of our 2025 AI predictions. Perplexity, everyone's favorite answer engine will either pivot, get acquired, or get squashed. Here's why. I think perplexity has lost its focus as a company. Right? They're going in so many different directions. They started making a daily podcast and then the CEO started talking about

Starting point is 00:18:59 making hardware, right? It started going in a lot of directions, but that's fine because all other companies are exploring other verticals. Yes, perplexity is a little smaller than, you know, Open AI, smaller than obviously, you know, Google, right? But here's the issue. When companies like OpenAI, they start competing in different spaces, Google, Claude, you know, Anthropic Claude, their core product improves first, right? That's one thing, especially the two startups, you know, that are building these models, Anthropic and Open AI, right? If they start going in another direction, that means their core product, their core USP has improved. I'm not seeing that with perplexity. Sorry, perplexity at its core, it's not a large language model, right? You choose your model,

Starting point is 00:19:52 although they have their sonar model. I don't really know anyone that uses that. You know, most people will use, you know, GPT40, Claude Sonnet 3-5 Nu or 351, whatever you want to call it, 36. You know, that's not, perplexity is not a large language model. It is an answers engine. It is a great product. However, the core functionality, I hate to say this. It's actually gotten worse. It's gotten worse.

Starting point is 00:20:20 I've showed it on the show before. What it's supposed to do is it's supposed to crawl out there, look at 10 to 20 different websites, and give you answers. It's not getting better at that. In my experience, it's actually gotten a little worse. And the Answers Engine Space has very recently got very crowded. All right, very crowded.

Starting point is 00:20:43 So chat GPT search, which has its flaws too, right? I personally liked browse with being better. ChatGPD search has some problems. I've talked about that on the show many times. But then you also have these large language models are just getting better internet connectivity. You also have Google deep research. Google is getting an AI mode.

Starting point is 00:21:04 I mean, speaking of deep research, it's one of the most impressive AI tools I've ever used, right so chat gpd search might look at you know eight to ten websites and a single prompt perplexity usually does more maybe 12 to 20 y'all google deep research it takes a little while right uh perplexity you know you might ask it something it might take you know five to 15 20 seconds same thing with chat gpt maybe a little faster uh because it doesn't go to to as many sources uh deep research yeah might take one two three minutes but i've had certain uh deep research um deep research that it looks at 1,300 websites, but I routinely get it to look at usually 60 to 200 websites. That is an agentic researcher that works for you.

Starting point is 00:21:52 Okay. So perplexity's got to pivot because what they were doing, they didn't get better. And it's too crowded. They're not that special anymore, right? 18 months ago, a year ago, they were still special. Not special anymore. Their core product is not getting better. the space is getting more crowded and the interest, I don't think is there.

Starting point is 00:22:13 As an example, y'all know I always bring receipts. Search volume, right? That talks about demand. Right now there's, you might think, I think we live in an AI bubble. So many of us out there, we might think, oh, you know, sure, chat GPT is maybe a little bit more popular than perplexity, maybe twice as popular three times. No, 66 times is popular. At least when we talk about search demand, right?

Starting point is 00:22:36 There's about 450,000 people a month searching for perplexity and 30 million people searching for chat GPT. So perplexity, I think, has kind of lost its way. I don't use it as much as I have. I still subscribe to it. I paid the paid plan for every single model out there because I'm pushing them every single day so I can tell you what's worth your time to grow your company in your career. Pnexity might not be worth it very much anymore, right?

Starting point is 00:23:04 I have paid Google Gemini, so I get deep research. Chat GPT search for very certain things is very good. Perplexities doesn't have any more. They could get acquired. I don't know. You know, you might see something like an Amazon. I don't know. They might get aqua hired.

Starting point is 00:23:21 They might pivot or they might fade into oblivion. All right, 17. I've got to go fast now. I'm trying to keep these under 25 minutes. So large language model API prices are going to drop like they're hot. This is not a surprise to anyone. But when I say drop, I'm saying they are going to drop. And one of the reasons why, well, a couple of reasons why.

Starting point is 00:23:41 Number one is open source, right? If there's free alternatives, the paid alternatives, if they want to stay relevant and make billions of dollars, they have to keep their prices low. Obviously, the hardware, the GPUs, the NPUs, the TPUs, right? All these AI and data center chips are getting more powerful, more plentiful. I think more and more things are going to go edge AI, which takes a little of that some of those resources off of the cloud. But I think hardware is getting better. The AI chips are getting better.

Starting point is 00:24:15 The open models are getting better. All those things drive costs down, right? Inference time goes down, all these things. But I do see. And it's crazy to think about when GPT4 was first release in the API. So I'm talking about backend. These are for developers, not, you know, if you're paying $20 a month to use a product on the front end, that's all you're paying, right?

Starting point is 00:24:37 Or $200 a month if you're, you know, on like chat GPT Pro. But otherwise you're paying an API price, right? You're paying, you know, a price per token. So when GPT4 was first made available in the API. So for developer, for I'm sure there's hundreds of thousands, if not millions, I don't know, maybe millions of businesses, these companies don't say. but it was $60 per million tokens for output. All right.

Starting point is 00:25:06 So you get chat GPT to spit out 60 million tokens, or sorry, a million tokens, 60 bucks. Now with GPT40, a wildly more capable model, $2.50. Went from $60 to $2.50 in about 19 months.

Starting point is 00:25:26 All right. I see in 2025, leading frontier models going even cheaper. I think, you know, for easy comparison, I think we're going to break that 50 cent mark, less than 50 cents per million tokens for a leading state-of-the-art model. All right. Two more here.

Starting point is 00:25:47 Number 16, embodied AI is going to be an exploding sector. All right. So what is embodied AI? Well, it's artificial. intelligence systems that interact with the physical world, whether that's, you know, robots, autonomous vehicles, drones, wearable AI, robotaxies, right? It is, you know, we've, we've had these things for a long time, right? Those things I just mentioned, none of those are quote unquote new, you know, humanoid robotics, autonomous vehicles, drones, wearable AI, etc.

Starting point is 00:26:24 none of that's new. What's new is having generative AI in large language models on board. That changes everything. That's how now you have startups as an example like figure that can go from obscurity. And in two years, turn into a multiple billion dollar company in like two years. You couldn't do that pre-generative AI because you had to build all that artificial intelligence. Now you can just use these companies API, use the technology that's already there, and essentially start stuffing it in humanoid robots, in robotaxies, in drones, in wearables, and it changes

Starting point is 00:27:07 how we interact with the world. I think I'm going to blame Apple, right? Apple came out with the Apple Vision Pro and everyone's like, oh, this is going to change the world. Literally before it came out, I'm like, this thing's going to suck. No one wants it. Right. But I think some of these early examples of embodied AI, especially wearable tech, they were terrible. And just so you guys know, go back and listen.

Starting point is 00:27:32 I was never bullish on any of these early wearables. I'm like, nope, nope, nope, that'll flop. No one wants it. I'm very bullish in 2025 on embodied AI. We're going to see AI out and about, just about everywhere. right. It's going to start in some of those areas that I just talked about. So even in example, in 2024, VC funding in robotics and drones soared by more than 40%. Another additional kind of hot take on top of this prediction.

Starting point is 00:28:08 I do think in some major cities where the law allows it, I think probably more than 10 to 15 percent of ride share rides will happen with autonomous vehicles, right? I don't think we're going to get the Tesla, you know, whatever they're calling it, the robo taxi AIs. You know, they're always, you know, they're like, oh, this is coming out, you know, in three months and it's going to be $30,000. Then it comes out eight years later for $100,000 like a cyber truck. So I don't expect Tesla to have anything in 2025.

Starting point is 00:28:41 But I do think other companies with this embodied AI bringing essentially large language model technology, vision technology, I think it's going to be fairly common. I do think in big cities, one in every 10 rides, potentially by the end of the year, will be in these type of vehicles. So embodied AI is going to be huge. All right. I made it. I made it.

Starting point is 00:29:10 I didn't accidentally talk for a couple of hours. So quick recap of the five bold AI predictions in this. volume of our series. So open source searches. I think open source models will temporarily overtake proprietary models. I think Chinese AI will dominate and cause some confusion. I think perplexity will either pivot, get maybe acquired or aqua hired, or just get squashed. I think API prices are going to get ridiculously cheap. I think we're going to see like less than 50 cents per million token for big models like through Open AI and Google Gemini. And then I think embodied AI is going to be an exploding sector.

Starting point is 00:30:01 All right. I hope that was helpful, volume two in the books. If this is helpful for you, let me know. But probably you should go listen to yesterday's volume one. It's going to be in your show notes. Make sure to check it out. We talked about a lot of different things, but a lot of things on agents and kind of the next steps in the workplace. So make sure you join us tomorrow for volume three. We're going to be talking

Starting point is 00:30:23 a lot about the kind of contents, AI revolution, videos, streams, and stars. Hope to see you back tomorrow and every day for more everyday AI. Thanks y'all. Meet Firefly AI assistant. Now live in Adobe Firefly, the Allman One creative AI studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows, across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution.

Starting point is 00:31:04 Stand control with the ability to step in and refine at any time. See it today at firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, Visit your everyday AI.com and sign up to our daily newsletter so you don't get left behind.

Starting point is 00:31:35 Go break some barriers and we'll see you next time.

Everyday AI Podcast – An AI and ChatGPT Podcast - EP 444: AI Market Forces Collide: Global Powers, Prices, and Problems (2025 AI Predictions - Vol. 2)

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.