Transcription
Hello everyone. Today we are diving into the world of artificial intelligence forecasts. The topic, well, you know, is changing at a cosmic speed. And we have an interesting source – it's YouTube blogger Nate B. Jones, who at the beginning of the year made 17 predictions for 2025, and then very recently, in August, after 8 months, he himself evaluated them. Do you know what verdict he delivered to himself? 41% hits. Not that much, it would seem. >> Well, 41% in a field like AI, I would say, is even not bad. Everything here changes not by the day, but by the hour. Predicting here is an almost thankless task. And our goal today is not just to put checkmarks, did it come true, did it not come true, in his opinion, we want to dig deeper, understand why exactly so, which trends really shot up, and which, well, got stuck. And what it generally says about where AI is moving and how it, well, affects our lives. So let's figure it out. >> Yes, absolutely right. The most valuable thing in such analyses is not so much the accuracy of the numbers, but the analysis of the reasons why the forecast worked or why it didn't, what factors played a role: technology, market, people, expectations. All of this is like an autopsy, only for trends. >> Excellently said. Well, shall we start then with the hits, as the author himself called them. He counted seven of them. And the very first block is about media, about content. Very relevant. Prediction number one. AI creators are coming out of the shadows. He predicted that AI characters, well, virtual influencers, models would start seriously competing with real people. >> Yes, and this seems to be really happening. >> He gives examples, well, quite vivid ones, let's say, AI models on platforms like OnlyFans. There, he says, some are already earning $10,000 a month, and audience growth is crazy. 300% over some period. >> $10,000 a month. A virtual model. Impressive. >> And this is not an isolated case. Entire creative agencies are emerging that specialize precisely in the creation and promotion of such AI content. And what's important, large platforms, social networks, they sort of encourage this or, at least, don't hinder it. Why? Because such content, as he writes, is very sticky. People get hooked on it, spend more time. >> This is perhaps the most interesting thing. Not the fact that you can generate a picture or a character, we've already seen that, but precisely this speed of commercialization and platform adoption. That is, if before it was somewhere on the periphery, well, such an exotic thing, a gimmick, now it's a full-fledged business with real money, with agencies, with integration into the mainstream. And this shows that there is demand, real market demand for such synthetic content, synthetic personalities. And this perfectly leads to his second hit. They are directly related. >> Well, yes, logically. >> Prediction number two. The feed is becoming synthetic. If we have more and more AI creators, then the content they produce, or which is heavily processed by AI, is simply flooding our feeds, social networks. >> Absolutely. One follows from the other. The author refers to estimates that almost up to 40% of content, say, on Instagram already has a strong AI influence, meaning it's either generated, improved, or somehow modified. And giants like Meta, they are not sleeping, they are embedding AI tools directly into their platforms so that users can easily generate and improve things themselves. >> Uh-huh. They are making it accessible to everyone. And the key problem he talks about, and which we all probably notice, is that it's becoming increasingly difficult to distinguish such content from human-created content. Pictures, videos, texts, the line is blurring. >> Yes, especially with image and video generation. The progress over the last year and a half is simply incredible. Sometimes you look and really can't tell if it's a real photo or a neural network's work. >> He also mentions Google's Genny 3 technology as an example of where we are heading. This is not just static image generation. It's the generation of entire immersive worlds. And, as I understood it, almost in real-time, as you move, interact. >> And there was another detail about Gny 3 that particularly caught my attention. It was mentioned somewhere that within this world generated by a large model, navigation or actions of some agent, well, a character, are controlled by another, more compact model, a small LLM. That is, AI within AI. >> Well, yes, one system builds the scenery, and the other plays a role in these decorations. You understand? This is no longer just a content generator. This is like the birth of complex simulations, virtual realities created and populated by AI. >> It sounds a bit creepy, even, but also exciting. And here we come to another prediction that the author also considers a hit, albeit with reservations, and it is directly related to this synthetic reality. >> You mean the great authentication crisis, number seven on his list of hits. >> Exactly. The prediction was that problems with deepfakes, identity theft, verification would only grow. >> Well, it's hard to argue with that, in my opinion. The author writes: "Creating perfect deepfakes has become frighteningly easy. Fraud using them is growing, and identity verification procedures are consequently becoming more complex. All these multi-factor authentications, biometrics, >> yes, the trend is absolutely obvious, although he also made a specific sub-prediction that there would be cases of fraud using a CEO's deepfake voice to deceive employees." And he says that there is no public confirmation of this yet. >> Interesting. So the trend of the authentication crisis itself, yes? But the specific example hasn't taken off yet. Or >> or as he himself asks: "And if something like that happened, would they talk about it publicly? Companies don't like to admit such blunders." >> Exactly. Reputational risks are huge. >> So, perhaps such cases already exist. We just don't know about them. The technologies for this definitely exist. Creating a voice deepfake today is not such a difficult task. >> So, look at the connection. The flourishing of creators and the synthetic feed and predictions 1 and 2 directly lead to the authentication crisis. Prediction seven. One feeds the other. >> Absolutely. The more synthetic stuff around, the harder it is to understand what's real. The higher the value of trust and authenticity. And the more acute the question becomes: how do we verify information, identity, content in this new world? This is no longer a technical question, it's, well, a fundamental challenge. >> Yes. Okay, let's move from media to other areas. For example, work. There were interesting predictions here too. Here's number four. A new AI colleague for non-technical specialists. Initially, he predicted that AI would become a common work tool not only for IT professionals but also for, so to speak, ordinary office employees, people from various fields. >> And how did he evaluate it? >> He evaluated it as almost a hit. That is, the trend exists, but perhaps not as fast or as widespread as he thought at the beginning of the year. He cites data. In the US, about 45% of workers already use AI tools in one way or another. That's quite a lot. And it's indeed penetrating traditional industries: marketing, law, finance, education. >> Uh-huh. >> And integration is accelerating. For example, in many companies, it's already normal to tag AI in a work chat, like Slack, so that it can, I don't know, summarize meetings or draft a letter. >> Yes, such functions are becoming more and more common. It's convenient. >> But the author himself admits that he might have underestimated one important aspect. He calls it the jaggedness or unevenness of implementation. >> Oh, that's a very accurate word. Jaggedness, I would say, is the key word for understanding what's happening with AI in 2025. >> Tell me more, what do you mean? >> Well, look, the technologies seem to be available, the models are becoming more powerful, the interfaces are simpler. It would seem, just take and use, but in practice, implementation is very uneven, like the teeth of a saw. In one company, especially a tech company or a startup, it's already part of daily routine, while in another, perhaps more conservative or in another industry, they are just starting to think about it or consider it fantasy at all. It depends on a lot of factors. Company size, budgets, task specifics, corporate culture, employees' willingness to learn new things, the availability of internal champions who promote these technologies. You understand? There is no unified front. >> Yes, I understand. So, the average temperature in the hospital, these 45%, doesn't reflect the real picture in each specific company or industry. >> Exactly, somewhere it's already 80%, and somewhere it's 5%. That's the jaggedness for you. >> And this is directly related to another of his predictions, which he rated as partially correct, number two in this category. The ultimatum on AI skills. Initially, he thought that proficiency in AI tools, well, AI literacy, would become almost a mandatory requirement in job postings almost everywhere. Such an ultimatum from the labor market. >> And it turned out, >> and it turned out to be the same jaggedness again. Yes, the requirement is growing, yes, training programs for employees are appearing, but this is mainly happening in certain sectors, for certain roles. There isn't yet a situation where everyone absolutely needs to become a prompt engineering guru. >> Absolutely. Instead of a universal ultimatum, we see more of, I would say, polarization or specialization. >> Polarization in what sense? >> Well, in the sense that for some professions, it's becoming such an important tool that you can't do without it. Programmers, data analysts, marketers, designers, they need to understand it deeply. And for others, the impact is still minimal or it's used for some very simple auxiliary tasks. And this division arises. Plus, new roles appear at the intersection – people who help implement and adapt it to the company's needs, train colleagues. >> So the conclusion is: not everyone needs to become an AI expert, but everyone needs to understand how AI can or cannot affect their work, their industries. I would put it this way: yes, you need to stay informed, assess the risks and opportunities for yourself, for your career, follow the trends in your niche, because this jaggedness is not forever. What seems distant today might become reality tomorrow. >> Logical. Let's talk about the technology itself, about AI capabilities. There was a prediction that the author considers one of the main hits of 2025. Number six. Memory changes everything. >> Oh, yes. Memory is fundamental. >> He says that key players like OpenAI with GPT, Anthropic with Claude, are actively investing in developing memory functions for their models. And this is not just about remembering within a single dialogue; inter-conversational memory is also developing. That is, AI remembers what you talked about yesterday or a week ago, and memory for AI agents so they can perform long-term tasks. >> And this completely changes the game, you understand? Without memory, it's like, you know, an eternal beginner. Every time you start communicating with it, it's like for the first time. You have to re-explain the context, your preferences, goals. >> Tedious. >> Very. And with memory, it can build, well, a user model, remember important information, adapt to your communication style, learn from previous interactions. This makes communication much more natural, productive, and, well, meaningful. >> In terms of work, that's understandable. An assistant who remembers your projects, deadlines, who you work with, that's a dream. >> Absolutely. But it's not just important for work. Remember the AI companions we talked about. Memory is what allows you to build a semblance of relationships with them. And AI that remembers your birthday, your shared jokes, important events in your life. This is a completely different level of interaction. >> And he also had a prediction about companions, hit number five. Visual AI unicorn, the companion phenomenon. He predicted a surge in the popularity of digital friends, partners. >> And he hit it. He considers it a clear hit. He refers to the fact that the AI companion market has already exceeded $120 million a year and to the mass deployment of visual companions by Grok. This is Elon Musk's chatbot from XAI, which has now apparently received some visual embodiment. >> $120 million is already a serious market. It shows that people are willing to pay for such communication. >> Willing. And popularity, according to him, is growing explosively. But there's an interesting nuance here. At the same time, another of his predictions related to companions, he rated only as partially correct. >> And which one? >> AI joins the party. The idea was that AI companions would start appearing not only in personal chats but also in real social situations. Well, I don't know, at meetings with friends, on dates, maybe even at some events. Interesting thought. And what didn't work? >> The evaluation is not very positive yet. That is, technologically, it's already possible to bring an AI interlocutor to a tablet screen or even in the form of a hologram. But the main problem is not in the technologies, >> but in what? >> In social norms, in etiquette, in how society will look at it. The author believes that a breakthrough use case is needed to make it acceptable. It's not there yet, but he believes it will happen later and even doubles down on this prediction for the future. >> And this is a very telling moment, in my opinion. It perfectly illustrates the gap between technological capabilities and social adaptation. Look, the technologies for creating convincing, personalized, memory-enabled AI companions exist, they are developing. The market is growing. People clearly find something for themselves in this: communication, support, entertainment. But as soon as we try to bring this companion out of the private sphere into the public, into social interaction, everything becomes much more complicated. Are we ready to sit in a cafe where someone at the next table is talking to their AI avatar? Are we ready to bring a digital friend to a family dinner? >> Sounds strange, doesn't it? >> For now, yes. A lot of questions arise, how will this affect live human relationships? Won't it lead to even greater isolation? What etiquette rules should apply here? Technology is clearly running ahead of our understanding and our social norms. >> And this, by the way, leads us to one of the three things that the author himself admitted he missed in his predictions. He believes he should have predicted it, but didn't. >> Oh, that's always the most interesting, the most critical part. What did he miss? First is LM attachment, i.e., a strong emotional reaction of people to changes. He himself is surprised why he didn't see this trend, although there were prerequisites. >> And he gives an example, >> yes, very telling. When OpenAI updated one of the ChatGPT voices, users raised a real storm on social media. They demanded the old voice back because they had, well, gotten used to it. They liked it. Well >> this is exactly what we were talking about in the context of companions and memory. People are starting to form not just a user habit, but a real emotional connection, and it ceases to be just a tool, a soulless program. It acquires, in the user's eyes, well, a certain personality, character, voice, communication style, even mistakes. All of this becomes part of the image to which one can become attached. >> like a character in a book or movie. Or even more. >> Perhaps even more, because you interact directly with AI. It reacts to you, it remembers you. If there's a memory function, it creates an illusion, or not an illusion, of reciprocity. And this has huge ethical and psychological consequences. How should developers account for this attachment? How to manage user expectations? What to do if the model needs to be updated or shut down, and people have become attached to it? We are just beginning to think about this. >> Yes, this is indeed an important point that was easy to miss by focusing only on technology. And what else did he miss? >> The second thing he should have predicted is the year of code. He believes he underestimated how important the application of AI in writing, analyzing, and debugging software code would become. >> But this has been talked about before. Copilot and other tools >> yes, it has. But the author, apparently, did not expect such a scale and speed of AI integration into software development processes in 2025. And importantly, he didn't see the close connection of this trend with the development of AI agents and the releases of new powerful LLMs that are well-tuned for code. >> And what's the connection with agents? Well, look, for an AI agent to perform complex tasks in the digital world, for example, book tickets for you, find information, manage other programs, it often needs to generate and execute code, call APIs, write scripts. And the better language models handle code, the smarter and more capable agents become. These are two sides of the same coin. And AI for code is one of the most powerful catalysts for the development of autonomous systems. >> Understood. So a breakthrough in AI for code is not only about speeding up programmers' work but also the foundation for the next generation of AI agents. >> Exactly. And the author admits that he underestimated this connection and the scale of impact. >> Okay. And the third thing he missed, >> the third is the fall of Llama, the rise of open source. He says he underestimated the speed and impact of the spread of powerful language models through open source. He specifically mentions Chinese models. >> And Llama is the model from Meta that they released into the public domain. Yes, Llama 2, then Llama 3. And it was an important step, but the author believes that he did not foresee how quickly other strong open-source models would appear, including from China, which can be run locally on one's own hardware and which, in terms of quality, will start to catch up with closed commercial models from giants like OpenAI or Google. >> And why is this so important? Well, open source is open source. >> Oh, it changes a lot. Firstly, it's democratization of access. You no longer need to pay huge amounts for APIs or depend on the whims of a single corporation. You can take a powerful model and use it yourself, yes, train it for your needs, integrate it into your products. Secondly, it accelerates innovation. A huge community of developers worldwide starts experimenting, improving, finding new applications. The pace of development in open source often outpaces closed developments. Thirdly, it reduces the risks of market monopolization by a few giants. A real alternative appears, and this is especially important for countries and companies that want more technological independence. >> So the rise of open-source LLMs is such a powerful trend that goes, as it were, against the dominance of big tech and which the author overlooked, >> yes, or at least underestimated its power and speed in 2025. This is indeed one of the key trends changing the landscape right now. It's very interesting, even an experienced analyst who is deeply in the subject can miss such important things. This once again emphasizes how complex and unpredictable everything is. >> Absolutely. >> Let's get back to his predictions. We've covered the hits and partially correct ones. There are still completely wrong ones. There are four of them. >> The most interesting part begins. Where expectations completely diverged from reality? >> Exactly. So, the first miss. AI consciousness activism. He assumed that in 2025, there might be real activists who would, I don't know, picket IT company offices or speak at shareholder meetings, demanding recognition of rights or even consciousness for AI. >> Hmm, a bold prediction, >> yes? And reality? Nothing of the sort. All discussions on this topic have remained on the internet, in blogs, at conferences, in academia. There have been no real protests or lobbying. >> Well, that's probably not so surprising. After all, the gap between philosophical discussions about machine consciousness and some practical actions to protect them is still huge. We haven't reached the point where AI demonstrates something that would convince a significant portion of society of its consciousness or feelings. And the very definition of consciousness – that's an eternal problem. Agreed. Apparently, the author got too far ahead here, succumbing to the hype around AGI. >> Quite possibly. Or overestimated people's readiness to transfer human categories to machines at this stage. >> Second miss. Smart mirror. He expected a boom, investment, and popularity of smart mirrors using visual AI, well, you know, for virtual clothing try-ons, for analyzing fitness exercises, makeup tips, and so on. >> And what happened in reality? In reality, silence. There were some niche products, mainly for fitness or health, but no boom, no mass demand happened. It didn't become the next big thing. >> There could be several reasons for this. First, the price. Such devices are likely still quite expensive. Second, the usefulness. How necessary are these functions in everyday life? Maybe it's easier to try on clothes in a store or watch exercise videos on a phone? Perhaps the killer use cases that would justify buying such a mirror haven't been found yet. The technologies are there, but a convincing application for the mass market is not yet. >> It seems so. Third miss, and quite a loud one, the iPhone moment for humanoid robots. The author was very optimistic about human-like robots. He expected a breakthrough already in 2025. He even suggested that an advertisement for some humanoid might appear at the Super Bowl in 2026. >> Yes, I remember the hype around robots from Tesla, Figure AI, and others. There were many impressive videos. >> Exactly. And in reality, as the author himself admits, he rushed. The real iPhone moment, that is, the appearance of a mass-market, affordable, and truly useful home or work robot, is at least a year or two away, or even more. Well, humanoid robots are perhaps one of the most complex engineering tasks, and it requires combining so many things: complex mechanics, a lot of sensors, advanced navigation in unstructured environments, safe interaction with people and objects, the ability to learn new tasks, and all this must be reliable, autonomous, and cost acceptably. Progress is certainly being made. We see robots that walk, carry objects, even try to perform tasks like folding laundry. >> Yes, I saw such a video. A robot folding laundry, very slowly. >> Exactly, slowly, not always perfectly. And such a robot costs as much as a premium car. Yes, the iPhone moment that will change our lives is still very far away. Expectations have again significantly outpaced reality. Fortunately, I'm not sure I'm ready for a robot butler at home. >> For now, you can sleep soundly. >> And the last, fourth miss, according to the author, AI weddings. He predicted that in 2025, some practice or even attempts at legalizing marriages between people and AI might appear. >> Seriously? AI weddings. >> Yes, apparently as a logical continuation of the AI companion trend. But there is no movement in this direction at all, not even close. There is no legal basis, nor public demand. The author fully admits his mistake. >> But this is already extreme exoticism. The transition from an AI companion, even a very advanced one, to a formal marriage with legal consequences is a giant leap, both social and legal. Society is, to put it mildly, completely unprepared for this. And the very idea raises many ethical questions. Agreed. This prediction seems the most, well, far-fetched, perhaps. >> Perhaps it was needed to illustrate some extreme point of development in human-AI relationships, but reality is much more prosaic for now. >> So, we have analyzed all 17 predictions: seven hits, six partial hits, four misses, and three more things that the author believes he missed. The picture is quite mixed. >> Yes, very mixed. And this, in my opinion, is the most important thing. It shows that AI development is not a direct and smooth upward path. It's a complex, non-linear, jagged process. >> Let's try to draw some conclusions. What does this analysis of predictions tell us about the current state and future of AI? >> Well, first, it confirms the incredible speed of technological progress in some areas. Media generation, language models, code processing. Here, breakthroughs are truly impressive and happen very quickly. This concerns the synthetic feed, the code code. >> Yes, the technological part is flying forward. >> Second, it vividly shows the gap between technological capabilities and their actual implementation and adoption. That same jaggedness. The existence of technology does not mean its immediate and widespread use. Economics, corporate culture, social norms, psychology come into play here. This is evident in the examples of AI colleagues, the skills ultimatum, AI companions in society, AI at parties, smart mirrors, humanoid robots. The technologies are seemingly there, but mass adoption or acceptance is not yet. >> So, social and economic factors slow down or direct technological development. Not so much slow down, rather, they create the real context in which technologies must find their place. Sometimes expectations for technologies are overestimated, as with agents or robots, and time is needed for the trough of disillusionment and the search for real niches. The classic Gartner hype cycle, which the author mentioned in the context of AI agents. >> By the way, he rated the prediction about agents as partially correct. That is, the fiasco is not complete, but it's not a takeoff either. He predicted the burst hype bubble with significant losses. There is no direct evidence of billions in losses, or they are hidden, but a rollback from the initial enthusiasm is evident. Why? The complexity of implementation turned out to be higher than expected. The real return on investment (ROI) is often unclear or deferred. And a huge shortage of specialists capable of creating and implementing these agents. The author says that demand exceeds supply by 10 times. 10 to one. What a deficit. >> And as a consequence, many AI projects related to agents are expected to be canceled. Public failures are also possible. Gartner predicts that same trough of disillusionment, but the author still believes in the long-term potential of agents. It's just that the path to it will be longer and more difficult. >> Understood. So the hype is subsiding. The time for sober assessment and hard work is coming. >> Absolutely. This is a normal stage for any breakthrough technology. >> And what else does this analysis show us? >> The third important conclusion, in my opinion, is the growing importance of the human factor. AI is penetrating deeper into our personal lives, our communication, even our emotions. We see this in the example of the companion phenomenon, in the example of LM attachment, which the author missed. We see it in the prediction about the revenge of audio, which we didn't analyze in detail, but the author considers it a hit. >> Yes, about audio. He talked about OpenAI's voice assistants, compared them to the movie Her, about emotional AI companions becoming mainstream, and about complaints about ChatGPT's voice change as confirmation of attachment. So voice, emotions, memory, attachment, and it ceases to be just a tool, it becomes, well, a kind of social actor. And this raises completely new questions: ethical, psychological, social. How do we build relationships with AI, what boundaries do we set? How does it affect us? >> And this echoes his prediction about AI safety becoming personal, which he rated as partially correct but believes was ahead of its time. >> Exactly. He mentioned AI offline spaces, digital detox programs, discussions about personal boundaries with AI companions, and the impact on mental health. That is, people are starting to think not only about data security but also about psychological safety, about the impact of AI on their personal well-being. >> But it's not a mass trend yet. >> Not yet, apparently. The author believes that this is more of a 2026-2027 trend, but it's embryonic. The more AI is in our lives, the more we will reflect on how it affects us and how we live with it. >> So, let's sum up. Technologies are flying, society and economics are trying to adapt. Jaggedness, hype cycles. And at the intersection, new deep questions arise about our identity, relationships, trust, and well-being in a world where the line between real and synthetic, between human and machine is becoming increasingly blurred. >> I would say that's an excellent summary. Predictions are just an attempt to find a vector, and reality always turns out to be more complex, multifaceted, and interesting. And analyzing these predictions, their successes and failures, helps us better understand this complexity. Not only what is happening with AI, but also what is happening with us. >> And now, concluding our conversation today, I would like to offer a thought for our listeners to consider. We've talked a lot about quite obvious trends. Synthetic media, AI at work, companions, authentication crisis. But given this complexity, jaggedness, unpredictability, and the emergence of phenomena like emotional attachment to AI, what non-obvious, perhaps hidden now, ethical or social dilemmas might suddenly become acutely relevant in the next year or two, about which we are almost not talking today, but which could explode just as unexpectedly as, for example, the mass demand to return ChatGPT's old voice. That's what I think is worth thinking about. M.