Learning Library

← Back to Library

2025 AI Prediction Scorecard

Key Points

  • The speaker reviewed 17 tech predictions made in January 2025, using a self‑created grading rubric to assess which were accurate (“hits”) and which missed the mark.
  • Seven predictions were deemed “home runs,” with the strongest being the rise of AI‑only creators who are now earning six‑figure incomes and prompting the emergence of AI‑native creative agencies.
  • A major shift in social media content was confirmed: roughly 40 % of Instagram feeds now contain AI‑generated material, making it increasingly hard for users to differentiate synthetic from human‑created posts.
  • AI integration is deepening across platforms, exemplified by tools like Google’s Genie 3 and Meta’s internal AI features, even leading to recursive AI systems where one model navigates worlds generated by another.

Full Transcript

# 2025 AI Prediction Scorecard **Source:** [https://www.youtube.com/watch?v=VN8_LWeYlHk](https://www.youtube.com/watch?v=VN8_LWeYlHk) **Duration:** 00:17:12 ## Summary - The speaker reviewed 17 tech predictions made in January 2025, using a self‑created grading rubric to assess which were accurate (“hits”) and which missed the mark. - Seven predictions were deemed “home runs,” with the strongest being the rise of AI‑only creators who are now earning six‑figure incomes and prompting the emergence of AI‑native creative agencies. - A major shift in social media content was confirmed: roughly 40 % of Instagram feeds now contain AI‑generated material, making it increasingly hard for users to differentiate synthetic from human‑created posts. - AI integration is deepening across platforms, exemplified by tools like Google’s Genie 3 and Meta’s internal AI features, even leading to recursive AI systems where one model navigates worlds generated by another. ## Sections - [00:00:00](https://www.youtube.com/watch?v=VN8_LWeYlHk&t=0s) **2025 Prediction Review: Hits & Misses** - The speaker revisits their 17 2025 forecasts, explains their grading rubric, and highlights seven predictions that came true, notably the rise of openly AI‑generated creators. - [00:03:16](https://www.youtube.com/watch?v=VN8_LWeYlHk&t=196s) **AI Voice Companions Go Mainstream** - The discussion highlights emotional AI voice companions becoming mainstream, backlash over voice changes, rapid adoption of AI tools across non‑tech jobs and workplace platforms like Slack, and the emergence of visual AI companions driving a growing multi‑hundred‑million‑dollar market. - [00:06:22](https://www.youtube.com/watch?v=VN8_LWeYlHk&t=382s) **AI Dating Trends Self‑Assessment** - The speaker reviews their recent hits and partial successes, then predicts that AI‑powered virtual companions will expand in dating platforms despite social etiquette and ethical challenges. - [00:10:50](https://www.youtube.com/watch?v=VN8_LWeYlHk&t=650s) **Reassessing AGI Timelines and Debate** - The speaker reflects on delayed AGI expectations, the increasing oddities in AI performance, and a shift toward personal AI safety concerns. - [00:14:01](https://www.youtube.com/watch?v=VN8_LWeYlHk&t=841s) **Review of Missed AI Forecasts** - The speaker evaluates a series of optimistic AI predictions—home visual feedback tools, a robot “iPhone moment,” Super Bowl robot ads, laundry‑folding bots, and AI weddings—admitting most missed their mark, noting only companionship remains a viable prospect. - [00:17:07](https://www.youtube.com/watch?v=VN8_LWeYlHk&t=1027s) **Request for Grading Feedback** - The speaker asks viewers to identify any missed points or grading mistakes and share their thoughts in the comments. ## Full Transcript
0:00Back in January of 2025, at the 0:02beginning of the year, I put my money 0:05where my mouth is or I put my 0:07predictions out on the table. I listed 0:0917 predictions that I thought would come 0:12true in 2025. And just about 8 months 0:16have come off the clock at this point. 0:18It is up to me to come back and be 0:20honest with you about what's working and 0:21what's not with my predictions because I 0:24believe in laying the cards on the 0:25table. All right, so we're going to go 0:26through all 17 predictions. I'm going to 0:28share with you my grading rubric first 0:31and why I'm grading things the way they 0:33are for each and you can push back and 0:35tell me where you think I'm wrong or 0:38where you think I missed something that 0:39might shift the grade. So, first we're 0:41going to go through what I call the 0:44hits, the things that I got right. So, 0:46out of 17 predictions, seven, we'll call 0:48them home runs. I knocked them out of 0:50the park. They absolutely were. Number 0:52one, 0:53AI creators come out of the closet. I 0:56predicted that we would see a world 0:58where AI characters and creators compete 1:01with human creators. Yes, I'm really a 1:03human. I know people sometimes think I'm 1:04an AI. I'm an actual human. The beard is 1:06real. But AI creators are coming out of 1:08the closet. There are AI only fans 1:10creators that are earning $10,000 a 1:12month or so with apparently 300% revenue 1:16growth. So, I'm sure it's much larger 1:17now. There are AI native creative 1:19agencies out there now. Major platforms 1:21are allowing AI content because frankly, 1:23it's sticky. I'm going to give myself 1:26yes. AI creators did come out of the 1:28closet. This is the year of AI creators. 1:30It's going to get bigger from here. 1:32Maybe they're going to have an AI Nate 1:33by this time next year. Who knows? 1:35Number two, your feed goes synthetic. 1:39So, this is sort of a correlary to 1:40number one, right? If you have more AI 1:42creators, you have more synthetic 1:43content in your social feeds. Well, that 1:46one is predictably a hit. Apparently, we 1:49are nearing 40% of Instagram feeds being 1:52heavily AI influenced. So creative 1:54elements with AI, AI creators 1:56themselves, etc. AI content generation 1:58tools are being integrated deeply into 2:00meta so Zuck can get people making more 2:03stuff with AI. And increasingly users 2:06cannot distinguish AI from human 2:08content, which comes down to advances 2:09we've seen this year in image and video 2:12generation. If you haven't seen Genie 3 2:16from Google, I encourage you to look at 2:18some of the videos. It can generate 2:19immersive worlds as you navigate them, 2:22right? It's one of those things that's 2:23really eerie because the world literally 2:25doesn't exist until you go there and 2:26then it's just it appears. And one of 2:29the things we learned from Google in 2:31August is that they now have so much 2:34confidence in Genie3 that they've 2:36actually got a tiny little LLM 2:38navigating the world inside the mind of 2:40that other LLM. In other words, they 2:42have a tiny LLM steering an agent 2:44through the world created by the AI 2:47Gen3. There's an AI living in the mind 2:50of another AI. I did not predict that 2:52one, but boy would have been cool if I 2:54did. Next hit, the revenge of audio. So, 2:57I'm calling out sort of the importance 2:59of her like voices and how that's going 3:02to be a big part of 2025. We have 3:05literally open AI's voice assistant 3:07comparisons to her in 2025. I did not 3:10make that one up. We have contextually 3:12aware AI voice systems being deployed. 3:14We have emotional AI voice companions 3:16becoming mainstream. It is a big year 3:19for voice. Next prediction. I don't 3:21think that one's going to be too 3:22controversial to say that's a hit. In 3:24fact, it's such a hit that when Chad 3:25GBT5 came out and the voice changed, the 3:28character changed, the way the content 3:30comeback changed, people complained. 3:32They complain because they liked the 3:33voice of the old version of 40 so much. 3:36That's how you know it's a hip. Next, 3:38your new colleague is an AI for non- 3:41tech. 45% of US workers report using AI 3:44tools at work. AI adoption is spreading 3:46beyond tech into traditional industries. 3:48Yeah, workplace integration is 3:49accelerating. I'm giving myself mostly a 3:52hit for this. I think by the end of the 3:54year, it's going to be more true as a 3:56hit. It's turning up that way. I will 3:58fully agree. We are we are not in every 4:02company at the point where you can tag 4:04the AI in Slack. In many companies, you 4:07can. And I think that it's worth calling 4:09out that if that feels strange and 4:10science fiction to you, that's more a 4:12function of your company context than it 4:15is of where the tech is actually at. 4:17Next up, a visual AI unicorn or the 4:20companionship phenomenon. This idea that 4:21we would have digital girlfriends, 4:23digital boyfriends, that's got to be a 4:25hit. Grock's virtual visual companions 4:27are rolling out on a massive scale. 4:29People have lots of feelings and 4:31thoughts about that, but they're there. 4:32The AI companion market as a whole is 4:34hitting $120 million plus per year with 4:37widespread adoption. It's just exploded. 4:39And so, I think that one's a hit. It's 4:41going to be even more of a hit in four 4:42months at the end of the year. Memory 4:44changes everything. Chat GPT is 4:47continuing to invest in memory. We now 4:49have Claude investing in memory. Cross 4:52conversation memory is happening. AI 4:55agentic memory is happening with 4:56multiple AI agents having different 4:58memory strategies. We have tremendous 5:01investment in the ability to even 5:03partially remember things. By the end of 5:06the year, we're going to have even 5:07better memory features across more LLMs 5:09and more agents. Memory is going to be a 5:11really, really big deal. And so I'm 5:14going to give myself a hit on that one. 5:15I think that's one of the defining AI 5:17transformations in 2025. And then the 5:20great authentication crisis. Essentially 5:22what I suggested is that we are going to 5:24have more and more issues with these 5:26voices creating deep fakes causing fraud 5:29issues etc. It's certainly true that 5:31perfect def deep fakes are becoming 5:33trivially easy to create. It's true that 5:36identity verification is becoming more 5:38and more of an issue. we are having to 5:40go to greater and greater lengths to 5:42prove our identity. And the last piece, 5:45the piece that I don't know if it's 5:46actually going to come true or not. This 5:48was a very specific prediction. I want 5:49to be honest about the fact that I don't 5:51know that we have this yet and we'll 5:52have to just flip the coin for four 5:54months to see if it really lands. 5:55Despite the overall trend being in my 5:57favor here, I did predict that a CEO 6:01would have an issue because someone 6:03would voice print that CEO from a public 6:06earnings statement and then use that 6:08CEO's voice to defraud a corporation. 6:12Maybe sounds a little sci-fi. I think we 6:14have all the building blocks on the 6:15table in August of 2025. No one has yet 6:18gotten that to happen that I have seen. 6:20On the other hand, the devil in me wants 6:22to point out if it did happen, would 6:24they report it? I don't know. Anyway, I 6:27think the trend is in my favor on this 6:28one. Now, we're going to move from the 6:30seven things that are broadly hits to 6:32things that I want to grade myself down 6:33on. These are partials. I think that we 6:35have we have partial hits on all of 6:37these. Uh there's six of them, and I 6:39want to go through them. By the way, if 6:41you're if you're tracking at home, seven 6:43hits out of 17 plus six partials. Yeah, 6:46we're not doing too bad. That's that's 6:4813 out of 17. that that are going okay. 6:50Let's be honest, no one's perfect, 6:52right? All right. Uh, strong partials, 6:54AI joins the party. So, AI, human dating 6:57platforms launching virtual companions 7:00and social settings. This is starting to 7:02happen. I think that the blocker here is 7:04actually social and etiquette, and we 7:06will just have to see how long that 7:08takes. I don't think the blocker is 7:09necessarily technical, although we may 7:11need a breakthrough app that makes it 7:13socially acceptable for people to bring 7:16their digital companion to lunch with 7:19the parents or whatever. That may sound 7:21weird and it's going to bring up all 7:22kinds of ethics and norms issues. It's 7:24going to happen. And so, I also want to 7:26call out areas where like on these ones 7:28that I'm not entirely like really seeing 7:30this year, I want to call out like am I 7:32doubling down here or am I walking away 7:34and saying I was wrong. I think I'm 7:36doubling down. I think we're going to 7:37see more here. AI skills ultimatum. So 7:40AI literacy is becoming a job 7:42requirement for some roles, not for all. 7:45Workplace de workplaces are starting to 7:48divide over AI adoption. I see that 7:50anecdotally with teams training programs 7:52are proliferating. I think what makes 7:55this a partial is that this is happening 7:57in particular sectors of the economy. 7:59It's not necessarily happening 8:00economywide. And I think that's one of 8:02the really interesting lessons of 2025 8:05is that AI transformation is as jagged 8:08as the intelligence of AI itself. And I 8:11don't think I fully laid that out and 8:13appreciated that in January of 2025. So 8:15I'm giving myself a partial here. And 8:17I'm going to say I am betting that the 8:19jaggedness will continue. I am betting 8:21that we will see these AI skills 8:23ultimatums and AI transformations in 8:25pockets. And I am betting that there 8:27will be other pockets that are much less 8:30changed over the next year or two. So I 8:32think that I am revising my thesis a bit 8:34here. Number three, big tech agent 8:37fiasco. I predicted there would be a bit 8:39of a bubble pop here. And I think that 8:42the reason that this is a partial is 8:44because I specifically predicted 8:47somebody would get this so wrong there 8:48would be billions of dollars in market 8:50capitalization losses. And I can't prove 8:52that there has been. Again, people are 8:54probably incentivized not to say that 8:56too much. So maybe there has been and I 8:57just got over optimistic about the 8:59economic consequences or overly 9:01pessimistic as it were. That being said, 9:03the larger trend, big tech did lean 9:06really heavily on the Asian hype train 9:08and we are seeing a lot of pullback on 9:11AI agents because the knowhow to 9:13successfully implement an agent that 9:15delivers ROI is way behind the hype. The 9:18number of people that can do a great AI 9:21agent, even one as simple as an NADN 9:23agent, is a tenth a tenth 9:26of the demand at least. And so 40%ish of 9:31AI projects are likely to be canceled in 9:33the next uh year or two uh according to 9:35widespread reporting um from AI newsshub 9:38and other sources. Uh you have had very 9:41public failures of AI agents and Gartner 9:44itself has predicted widespread agent 9:47shakeout. And so to me, it's the classic 9:49Gartner hype cycle of the trough trough 9:51of disillusionment and then we get to 9:53the plateau of productivity. AI agents 9:55are headed for the trough of 9:56disillusionment. And I say that as 9:58someone who's very bullish on agents 9:59long term. You're just going to have to 10:01go through the fact that a lot of people 10:03got hyped up on it, weren't given the 10:05tools, could not find people who could 10:07actually build and successfully 10:08implement them, and got very 10:09discouraged. We need more people that 10:11can build successful agents in 10:14production. So that one's a partial. I 10:16don't think that we got the billions in 10:17market cap losses, but I think that we 10:18did get a little bit of an agent 10:20implosion. So, hardware, I am giving 10:22myself a partial here because this is 10:25behind schedule. I am doubling down on 10:27this. I am predicting that this will 10:29happen. I'm predicting a new specific 10:31hardware device from OpenAI or from 10:34others as well. Hardware with AI native 10:37is coming. It is going to be an iPhone 10:39challenger of some sort. perhaps not 10:40immediately positioned to cannibalize 10:42the iPhone, but certainly intended to be 10:45another major device, a third device 10:47along with your laptop and your phone. 10:49This is coming. So, I think I'm giving 10:50myself credit in doubling down. It is 10:52delayed. I don't think that we are going 10:54to see something until lateish in 2026 10:57at this point. And so, I can't give 11:00myself full credit here, even if I'm 11:01bullish long term. Next one. The AGI 11:03debate gets weird. I think that this is 11:06partial only because of the definition 11:08of weirdness and also there's no 11:11consensus on AGI definition which I 11:13probably should have emphasized more. 11:14You do see AI systems achieving high IQ 11:17scores and yet they also fail some basic 11:19tests. Famously Andre Carpathy called 11:22out at YC school that you sometimes have 11:24AI still not realizing that 9.11 is a 11:27smaller number than 9.89 89 because it 11:31just doesn't look that way if you're not 11:33reading correctly and you're not doing 11:35the actual math. So, the AGI debate is 11:38absolutely getting weirder and weirder 11:41the more releases we have. And so, I 11:43think I'm going to double down on 11:45continued weirdness. I think that maybe 11:48I will maybe I will back away a little 11:51bit from the AGI debate being as central 11:54as I think I positioned it in January. I 11:56don't think AGI is going to matter as 11:58much as I thought. Even if the debate is 12:01going to continue to get weird. Okay, 12:03last one that's a partial and then we'll 12:05get to my misses. I know you guys stayed 12:06around for the misses. We'll get to 12:08that. AI safety gets personal. So, 12:12there's a lot of like early motion 12:14toward this. I just think I was too too 12:16ahead of the curve. This maybe giving 12:18myself too much credit, but like I don't 12:20see this fully realized yet. So, there's 12:22starting to be some AI offspaces and 12:24detox programs emerging. We're starting 12:26to see some public health implications 12:28of overaffirming AI getting reported 12:30widely. Personal AI boundaries are 12:32becoming something people talk about. 12:34Speaking of companions, it's becoming an 12:35issue in human romantic relationships. 12:37Do you have a digital companion? Do we 12:39have boundaries around that, etc. Uh, 12:41wellness industry is pivoting toward AI 12:43concerns as well, but it's not really 12:45fully there. To me, this feels like I 12:47want to double down, but it feels like 12:48it's running a little bit behind where I 12:50thought it would be. And so, I think 12:51it's more of a 2026, 2027 thing to get 12:54widespread. Okay. Okay. I know you were 12:56waiting for this. The misses. What has 12:59been too early and too wide? Sort of 13:02just not on the table. We have four 13:05things cuz remember we're doing seven. 13:06We did seven that were hit, six that 13:08were partials. Now, we have to do the 13:10fours that were terrible that I missed. 13:12I predicted that there would be AI 13:14consciousness activism at board meetings 13:16from people who believe in AI 13:18consciousness. There are certainly 13:20people who believe in AI consciousness, 13:22but it's been mostly an academic 13:24discussion. It's been mostly a chat room 13:26discussion. It has not yet disrupted any 13:28board meetings. It seems somewhat low 13:30probability that it will. It's I'm 13:32giving myself a miss on this one. I 13:34think I just did not correctly predict 13:36this. Even though I did predict that 13:38there would be some degree of 13:40consciousness conversation, that's not 13:42the same thing as saying that there 13:43would be a sort of disruptor that's like 13:45allin for AI labor rights at the board 13:47meeting. I just that didn't I think the 13:49smart mirror that I predicted was a miss 13:51too. I predicted that there would be a 13:52lot of investments in smart mirrors 13:54because people want to get into visual 13:58AI, right? They want to try on outfits. 14:00They want to exercise things that they 14:01can do at home that need feedback that 14:03are visual from the AI. We've had a few 14:05fitness and health things. It's not It's 14:08not a major breakthrough. It's not on 14:10track to become a household product. I 14:12just missed this one. I'm not going to 14:14double down. Third miss, humanoid robots 14:17will have their iPhone moment. I 14:19predicted specifically that there would 14:20be a robot 2026 Super Bowl commercial, 14:23which I still have time on the clock 14:24for, so I'm not giving up on that. But I 14:26think largely speaking, we still don't 14:29have the iPhone moment for robots. We 14:32are we are still a year or two away and 14:35it's been that way for a bit. I know 14:37there are now production robots on sale 14:40that can solve household problems. I saw 14:42one that folds your laundry. Looks way 14:44too expensive to fold my laundry. I 14:46think I was too early on this one. I'm 14:47not going to double down. Last one. AI 14:50weddings. I predicted that an AI 14:52relationship moment would turn into a 14:53wedding. As much as we've talked a lot 14:55about companionship and the social 14:56implications of AI, I've seen no 14:59traction on AI weddings anywhere in the 15:01world, no legal framework, nobody 15:03stepping up to say, "Come to our 15:04country. We will do AI weddings for 15:06you." It might still happen, but I think 15:08I just jumped in the water too early on 15:10this one. So, I'm not going to double 15:12down there. I think the companionship 15:13thing is sticking around. I think 15:14weddings are hard to predict. So, there 15:17you go. Uh 41% of my of my predictions 15:20were broadly correct. 35% eh give 15:23yourself some partial credit and 25% 24% 15:27no good at all just didn't get it right 15:29what do you think about what's next what 15:31would I what would I wish I had added to 15:34these 17 predictions that's the last 15:36part of this if I if I look at what has 15:38happened so far if I could go back and 15:40do it again what are three things that I 15:42would I think first and foremost I 15:44should have predicted the kind of LLM 15:47affinity that we saw with the chat GBT5 15:49roll out where people just demand anded 15:5240 to come back. I had all the 15:54ingredients there. I talked about 15:55companionship, emotional bonding. I 15:57should have seen it. I didn't see it. 15:59Didn't predict it. Don't get to credit 16:00it. Another one that I should have 16:01predicted. I should have leaned heavier 16:04on this being the year of code. AI and 16:06code is something that has been so 16:08heavily invested in by every single 16:10major lab and frankly by a lot of the 16:12tool builders themselves. And I just did 16:14not put enough effort into sort of 16:16investing in smart predictions there. I 16:18think if I had been smarter, I would 16:19have seen the relationship between code 16:21and agents. I would have seen the 16:23importance of code in flagship releases 16:25do it again. All the ingredients are on 16:27the table. I could have seen it. And 16:29then three, like stepping back, I did 16:31not correctly predict the downfall of 16:35llama and the proliferation of open- 16:37source. There are now hundreds and 16:39hundreds of open source models that are 16:41excellent. And you may hear about open 16:42AI models, but there are lots and lots 16:44of Chinese models that are incredible as 16:46well that are open source, open weights 16:48that you can then run on your own local 16:49machines, and you don't have to run on 16:51Chinese servers at all. I should have 16:53done a better job noting how quickly LLM 16:56technology proliferates and that was a 16:58miss on my part, too. So, there you go. 17:00Three misses, things that I think I 17:01should have predicted, the 17 original 17:03predictions and how I broke them down. 17:05What do you think I missed? Where do you 17:07think I graded wrong? Let me know in the 17:10comments.