Why Marc Andreessen Says We Already Achieved AGI 3 Months Ago

Ajouté : 2026-05-25

[00:00:00]Fun about it was I could compose, you know, rap lyrics based on Shakespearean poetry or I could write a great wedding speech or like what, you know, I could do all kinds of fun stuff. But it had all these problems. It hallucinated and it made stuff up and it wasn't good at like it wasn't good at logic and it couldn't do basic math and it had all these issues. And so people >> a baby. It was a baby. It was a little it was a little Yes, a little tiny baby.

[00:00:17]Learning how the world works. The the the technology advances in the last 3 years have been like mind-boggling. Like crazy. Amazing. Impressive. Um and so I I actually people talk about this concept called AGI, which means artificial general intelligence, which basically means an AI that's as smart as a person. And I actually think we crossed that about 3 months ago. Um and I think it was it was with the very latest versions of the of the leading models. And and one of the reasons people are having a hard I'm going to come back to that. One of the reasons people are having a hard time understanding what's happening in AI is cuz it's moving so fast that if you don't use the latest thing, you don't understand what's happening. Cuz you're not seeing it. So a lot of people used ChatGPT last year or the year before and >> Right. they're not actually seeing the new thing. Right. The new thing specifically is um it's uh uh it's called uh uh GPT, I think it's 5.5. Uh and then it's this uh it's the Claude.

[00:01:06]Anthropic has this thing Claude. Um and and that's called uh 4.6 um was was the key release. And then Google has this thing Gemini. Uh it's just like 3.0 and then uh Grok. Um it's 4.3. So these models all have they in in each case I think in in in with those releases, they kind of hit this threshold. Uh where all of a sudden, I guess I'd say this, like in in in in in my line of work, 99% of the time the answer that I'm getting from the AI from those from the most advanced models is better than I would get from talking to it uh basically almost any expert I have access to.

[00:01:39]Um and I have access to, you know, in my job a lot of experts. Um and I'd say 99% of the time I'm getting a better answer from the AI. It meaning a better answer meaning smarter, better analysis. And and and I part of it is what they call fluid intelligence, which is the ability to conceptualize and process information. And And of it is what psychologists call crystallized intelligence, which is just memorization of everything.

[00:02:00]And so the what the AI brings you is it brings you both because it it's smart, but it also knows it's it's trained on all the data. It's trained on It's trained on like the complete corpus of human knowledge, right? And so it's a world-class doctor and a world-class lawyer and a world-class accountant right? And a world-class politi- you know, I don't know political operative.

[00:02:20]If you want to run for city council, and it's a world-class marketing expert if you want to market your podcast. Or and it's a world-class software coder if you want to write write write write some software code. And so so it knows everything about all of these fields all at the same time.

[00:02:33]And then of course it has the huge advantage and and I love people and I love talking to people. It has the huge advantage of is endlessly happy to talk to you about anything. Right.

[00:02:41]>> Right. It doesn't get impatient.

[00:02:42][laughter] >> Right. It doesn't get frustrated. One of the really fun things I do with AI is you know, I'll ask it a question I'll get back this complicated answer and I'll just be like I don't this is too complicated for me, you know, I don't know something in quantum physics or something and I'll say so you say explain it to me like I'm 10.

[00:02:56]Yeah.

[00:02:56]>> And it gives you that So like all of a sudden it's like talking to you in terms you understand and then you're like all right, this is still confusing. All right, explain it to me like I'm five.

[00:03:02]>> [laughter] >> Right. And then and then at night what I'll do is I'll I'll I'll I'll do that all the way back. And so I I do it all the way back and I'll do it explain it to me like I'm two.

[00:03:09]And it's like well, you know, it uses the metaphors get you know, it's like you know how your mommy and daddy love you, right? And it's >> [laughter] >> And you know you have a pillow you love to sleep on at night.

[00:03:17]What if that pillow could be in two places at once?

[00:03:19]Um and so like it is absolutely happy to like do this endlessly. I'll I'll give you the the the medical implications alone. I'll give you my personal experience. So over the holiday break, I you know, I go on vacation I immediately get sick. I'm one of those people.

[00:03:33]Um so I immediately get food poisoning.

[00:03:35]Um and so I know I'm going to have nothing to do for like five days, right?

[00:03:37]I'm going to be on my on on my back.

[00:03:39]Five days for food poisoning? I mean, I don't know. It depen- I This was rough.

[00:03:43]This was yeah. Damn. This was intense.

[00:03:44]>> I the yeah I will not I'll protect the guilty. Okay. Um I I know, but I I won't say. So um Tell me later.

[00:03:52]So I just decided I just basically said um I'm going to do is I'm just going to let Dr. GPT take care of me.

[00:03:57]Mhm.

[00:03:58]Right? So, and I went I went totally overboard on purpose. And I just basically said like so like every 20 minutes I gave it like an update of like, you know, and it literally I'm giving, you know, it's personal information. You know, I'm like, you know, okay.

[00:04:07]Diarrhea. I just had a visit to the, you know, here's what happened. Uh I I didn't do the thing you can do. You can actually send it photos now. I didn't Of your poop? Yeah, I didn't I I didn't do that although you can and it and it will it will do that. But I I was already nauseous enough.

[00:04:20]Um but I gave it like moment-to-moment updates. And then this is like I wake up at 4:00 in the morning, I feel terrible and it's like I, you know, and I literally type in it's 4:00 in the morning, I feel terrible.

[00:04:27]And it gave it's it was like amazing.

[00:04:29]It's just like it's have it's they have like the best doctor in the history of the world who is just like happy to be there at 4:00 in the morning with you holding your hand working through this.

[00:04:36]It's [laughter] just a completely different kind of experience than anybody has ever had in medicine.

[00:04:40]And then to have the the exact same opportunity for anything that legal that comes up and for anything in your business and for anything >> Right. By the way, how to parent how to parent you I do this all the time. I've got I've got an 11-year-old like how do I All right, what movie should we watch?

[00:04:51]All right, like which ones are safe?

[00:04:53]What kinds of content do I want not want? Mhm. You know, um it like it's and it's infinitely it's just like, oh, tell me what your guidelines are and then it's like infinitely sensitive. It gives me um uh so, I want to watch movies with them and I know there's like three scenes in the movie that I don't want them to see.

[00:05:05]There's like, well, what when are those scenes? And it gives me like the exact timestamps of the scenes and you know, it says, you know, pause it here. Could you run a movie through it and tell it eliminate those scenes?

[00:05:13]>> Yeah, you can uh so, you can for sure. I haven't done I haven't done that. Uh people have done that. Uh that that that has been done. But yeah, you You could do that. That would work now. Blur out the nudity? You you could do you could do you could do the blur you could do the blurring for sure. Yeah, it could definitely do that.

[00:05:26]>> Wow. But it's just like it it's this thing it requires this kind of mindset change.

[00:05:32]I'm maybe two parts to the mindset change. One is just realizing what this thing can do. And and it's a it's a bit of a black box in the sense of like you can tell it to do anything. And so, you you you but you have to like figure out what to tell it to do. And so, there's a there's a there's a learning process that kind of kind of of goes goes goes with that for sure.

[00:05:47]Uh but the other part of it is just like in in your day-to-day thought is just like, "Okay, when do I hit when do I hit the barriers of my own knowledge?" Like when do and and and in the past like I would have been frustrated, but I wouldn't have even been aware that I was frustrated just because I took it for granted that of course I have no way of answering this question.

[00:06:03]Um and now all of a sudden I I mean I just you know, you take your car to the mechanic and it's like, "Oh, you needs a new radiator." I I I I don't know. Like what should I look at?

[00:06:10]You know, and it gives you like the complete undressing of the whole thing.

[00:06:12]And it's just like it's a capability that you you know, unless you have a friend who's like a car expert that you bring with you, you never would have had a way to do that. You would have just given up from the very beginning. And now you've got something that's happy to hold your hand through it um and and happy to make sure You don't have to sell me on it. I'm I'm a giant [clears throat] fan. I I think it's pretty fantastic in terms of just use like in daily life. You can get a lot of information from it. I use it for what if I'm ever writing, I keep like my phone open. And so I have my computer on and my phone on like and I started asking questions to the phone. I just ask Perplexity like, "What is this? Why is that? Well, when did this start? Why why did people start doing that? And what's the argument against it? And what's this and what's that?" And you know, and when did Spain invade Mexico? When did people start speaking Spanish over there? You know, like that kind of Yes. And you said something interesting. You said you think 3 months ago it artificial general intelligence I think we hit the we hit the change. Yeah, I think we hit the change. So, I I forgot the name I can't believe I'm blanking on the name, but the the test Oh. The Turing test.

[00:07:14]Turing test. Alan Turing. I couldn't remember his name. You think it's there?

[00:07:18]Yeah, for sure. So, for sure. So, so >> But that would that should be like massive news. This is what's confusing.

[00:07:25]Correct. And I totally agree with you and we in the industry talk about this all the time that this is not massive news and it should be. And and and so here's Okay, so for people for people who haven't heard of the Turing test, the the the Turing test was for 60 years it was the gold standard in figuring out whether AI would work or not. And the basic goal of the Turing test was can can you if you're a human being can you tell whether you're talking to another human being basically in a chat room or whether you're talking to a bot. Um and for 60 years it was impossible. Nobody Many people tried to write software to pass the Turing test.

[00:07:52]Nobody ever succeeded. Um we blew right through the Turing test over the Christmas holiday of 20 2022 when ChatGPT came out. We just like blew right past it. We blew past it so fast and so hard nobody has even bothered to do the test.

[00:08:06]I I I mean there's probably a handful of papers where somebody's actually formally done it. But like it it it it it it it it we blew through it like tissue paper.

[00:08:13]To the point where it was not even it it it is and again people older people in the industry like me know we're just like wow.

[00:08:19]Exactly your reaction like that seems like it should have been a big deal and it's like oh no that was like yesterday's news. Like that turned it it turned out it turned out well what we now this is part of the what we now know is it actually turned out to be easy.

[00:08:31]Part of the miracle of what we have now there there's now a large language model that this this guy Andrej Karpathy who's one of the leading experts in the space has developed. He's developed a large language model in 300 lines of software code.

[00:08:41]Um there are people who are back porting large language models to run on PCs from 40 years ago.

[00:08:47]Um you can run somebody's got people have them running on I saw somebody has a large language model running on a on a on a on a Texas Instrument calculator.

[00:08:55]Woah.

[00:08:56]>> Um and and so it just it it it it it turns out