Does AGI Actually Matter to You?

[00:00:00]

Matt: All righty everybody, this is yet another edition of the Web News. And like many other content creators are covering, uh, ChatGPT, uh, 6 Astra is here. This is yet another, you know, crazy moment. It's kinda like the Mythos level whole Fable, I think it was Fable V, everyone running around going crazy, uh, kind of moment.

But, uh, we kinda wanted to ground this episode. So if you're watching the video version, there is gonna be the ChatGPT or the OpenAI website open to the GPT-6 Astra page. But we specifically, uh, want to sort of ground this with a, a bit of a thesis, if you will. And it's, you know, what does AGI mean to you?

'Cause there's a lot of people out there saying, you know, "AGI is here. Is AGI here?" You know, the questions are being raised, and there's an AGI test or a benchmark that we're gonna be talking about, uh, in these, uh, i- on this page, on this GPT-6 Astra page. We're gonna be talking about that and how, how well it scored in an AGI [00:01:00] test.

However, it's the third version of the test so, like, you know, is there a moving finish line for AGI? Y- you know, what are we talking about here? So let's, uh... I, I think we should, you know, just kinda dive in here. Now Mike, uh, you were ... You broke this news to me. I'm not, like, super up to date with a lot of AI stuff.

I'm just kinda, like, using it. And so you were like, "A new big," you know, "a new big thing kinda happened." And you sent me this page. And so I was like, "All right." And so of course, you know, read this in prep for this episode. But why don't you take us through, you know, s- maybe a, a key point or two of this, uh, GPT-6 Astra release

Mikhail: Yeah, absolutely. And I wa- I wanna put a couple dis-disclaimers in before. Um, the main disclaimer is I haven't used GPT Astra. I wasn't part of like the early access program for it, and this is all coming from like third party benchmarks and some other... Some people have used it, so I've watched their videos reviewing it and doing demos and stuff like that.

So it is out to a certain limited amount of people and [00:02:00] organizations, so it's not completely vaporware, but I have not personally ran my experiments on it and use it on a day-to-day basis. So take everything we're gonna be saying with that grain of salt, right? Like we know that this, this model is good, like it's very good and people are saying it's good. So there's no hiding from that. But until I can get my hands on it, I can't give you my honest like 100% feedback. Having said that, um, this seems to be, like you said Matt, a pretty big leap. This isn't, this isn't a iterative bump. This is why it's not a five class model like GPT 5 was the previous one. Uh, this is a six class model, and this is their top six class model which they're, they're name, naming Astra, right?

They-- uh, GPT has like OpenAI or ChatGPT have this like naming sch- uh, naming scheme of, um, planetary bodies usually, like Sol being the bigge- their biggest previous model which is like the sun, Terra the earth, and then Luna being their smallest model which is the [00:03:00] moon. Now there's like Astra which is, you know, a whole solar system. Um, and this is a big model, that's one thing, and the next thing is that this seems to be I, I don't know how to put it. Like I-- This is why this episode came out is like we don't really know what AGI is, and I h- I really hesitate to say that this is AGI because it I can't think of a way to just like say that it is.

It's not sentient, but I don't think AGI needs to be sentient or i-in that sha- any way, shape, or form. believe it, what AGI means is just, it's just supposed to be better or equal to human intelligence like a very layman perspective, right? Like from a o- a one liner perspective

Matt: Yeah, like I'll-- Like quite literally, you know, artificial... Like I'll, I'll read this right from, uh, IBM's website. Artificial general intelligence or AGI is a hypothetical stage in the development [00:04:00] of machine learning ML, uh, which is an artificial intelligence AI, uh, in which an artificial intelligence system can match or exceed the cognitive abilities of human beings across any task.

So this is like a big thing, any task. It represents the fundamental abstract goal of AI development, the artificial replication of human-level intelligence in a machine or software

Mikhail: It's tough. Like, this one's tough because I've been putting AI through its paces over the last three to four months. Like, I've been it more and more. How I started earlier was most of my AI tasks from s- six months ago were only running for three to five minutes at most. Everything was pretty quick because I was giving it very scoped down work. I was telling it, "Hey, edit this code in this section on this page to do this." Right? Like, I was giving it very direct work, and I was like, "Oh, this is amazing." It was cool. As I've gotten to trust it more and as I've worked out my end-to-end feedback workflows, I've [00:05:00] given it more and more work, and at this point, like in the last even few weeks, it's running loops of eight to 16 hours for me doing very big changes, like grandiose changes to different applications on multiple different levels. I've, I've gotten it also to do a lot of like computer work for me, like filling in spreadsheets and doing a bunch of like m- menial tasks. And I'm treating it very much like I would a person, and it is performing at the level that I would expect a person to perform to, even with the deficiencies that it has. So I have a hard time, like, saying that it's not better at cert-- at a lotta things. I don't know... Like, any is a big word in that, in that statement, Matt. But if that's really the, the metric that we're at then it's doing a lot. Like it is, it is doing a lot. It, it, it might still need me to, you [00:06:00] know, log in and do some

You know, give it some basic information, give it access to certain things but overall it's able to do anything I've thrown at it. Maybe I'm not throwing enough complexity at it, I'm not sure, but it's, it's been pretty intense over the last little while

Matt: Well, what I was gonna say, and we were talking about this briefly in the show to sort of like anchor it in the thesis, like what does AGI mean to you? Like clearly there's a definition there. But the problem when you say any task is like, I'm not gonna be very good at climbing Mount Everest whereas like someone else is going to be.

The argument even there is that the AI or a human if we take AI out of it for a moment... Uh, li- like I could learn to climb Mount Everest. I could learn to free climb, I could learn to climb, I could learn to be good at higher altitudes, I could learn survival stuff, I could become more fit and be able to actually handle harsher conditions.

I would know what equipment to bring, I would know my, you know, [00:07:00] my calories and my macros and things to allow myself to continue to climb up and down and, and perform at an athletic level, uh, in, in extreme conditions as well. And like so I could, I could do that. Now this is the thing though is like look at all those things that I mentioned just talking about climbing Everest.

I've watched some like movies on climbing Everest, and I've like seen some YouTube videos but I'm not an expert and those are-- Look at all the little things that I specifically mention there. And then there's e- an expert, if we had an expert on the show on climbing Everest he would in- he would say a bunch of other things that I would need to do in order to get to that stage.

My point is, is that when you say any task, there is a lot of nuance with th- with basically every task obviously some exclusions apply. But there's a lot of nuance with every task let alone any task. You know is this thing capable of learning anything? I mean i- what it makes it, what it makes me feel like is, is there is a, there's a [00:08:00] moving finish line to an extent here.

I mean, u- like off the top of my head wouldn't the very first version of the AGI test be enough because it's any task? But then we're on the, the third version of this AGI test and Mike you mentioned that there's a fourth version coming out or something. Um, and this is the Arc AGI 3 is what, uh, we're looking at on the screen here if you're watching the video version.

And so to me it's sort of like there's a, a, a moving finish line in a way because humans approach things you know so differently and, and eh, that it's, it's

It's, it's very difficult to take something analog and something that we take for granted and move it into the th- th-- into the digital world. And then also somehow test it like w- th- even with humans... I mean humans have --humans and traditional education has been around for a long time, and we have these sort of key metrics and key tests that will do for people.

So in Canada we [00:09:00] have a grade 10 literacy test which I think has evolved to something else in more recent years. Don't quote me on that but that's what we did when we were in high school not all that long ago. There was grade six testing which tested across a bunch of different things math and science and things there was also grade three testing did exactly the same thing

And um, we did those, like we did those testing and that's how they gauged like how good the students were doing and how the education system was faring and, and all these things . But there's criticisms of even those tests where its like hey this tests are very rigid you know? This person may be a complete fool when it comes to Math but their amazing at something else and is sorta like ,you know why are really

You know how are punishing them ? Like are we going drag down the whole school board because we just got batch of more creative students rather than mathematic centric studen--like theirs tons and tons ways criticize this and those test have different versions in and of themselves..And so my point being is i almost think [00:10:00] that definition AGI IS going to move around unless something absolutely blows us out of water In terms of Amazing Us.

We’re like oh My God!This new model whatever It Is IN The Future Is SO AMAZING Because Humans Do Alot Of DIFFERENT Things LIKE I Mentioned…I’ve Just mentioned Grade School I’VE MENTIONED CLIMBING MOUNT EVEREST LIKE HOW MANY Playing Retro Games Doing Electronics DOING MATHEMATICS BEING A Being A Traveler BEiNG YOU KNOW LIKE How Many THINGS YOU Know Are WE REALLY Going To Be ABLE TO TEST AND IM SURE That ANY Test THAT We Come UP WITH IS GOING TO DRAW CRITICISM AND HAVE PROBLEMS

Mikhail: Yeah, it-it's the like figuring out how to actually evaluate these models is I think becoming almost as big of a challenge as actually creating them. honest and I think that's going to only further become more difficult because of what you're saying. an infinite-- literally pretty much an infinite amount of things that humans are [00:11:00] capable of doing how do we actually...

Like we can never really put the check mark on this. Uh, yeah, we could --I don't think we could like i-i--we could ever really, you know say that it can do everything but at some point do we just You know, it-it's doing it. Like it's-- it's doing it, it's doing it. Like if it, if it's able to act as an employee, it's able to act as an employee, maybe that's the metric. Like if you're able to just hire a bunch of these GPT-6 Astra agents inside of a good coding harness and do... And it does everything that an em- uh, at an employee would do at the standard, uh, at or above the standard of an employee, maybe that's where your AGI matters for your company. So maybe it becomes not a moving target, but a specific target.

So maybe it's reached AGI, not again, for all of humanities, but maybe it's reached AGI for, uh, you know, [00:12:00] finance or lawyers or developers. Like if you're able to give it the tasks that you would give a developer at the same level of, you know, ownership and detail, and it gives you the output, why wouldn't it be a developer?

Like, why isn't it AGI for a developer, right? Like it's, going... I think it's gonna go that way, and I think, uh, like, uh, a lot of people have predicted this before, where we won't know its AGI until after its AGI. Like we'll evaluate it to its, you know, uh... We'll have a, a back date, be like, "Oh, we reached AGI six months ago because of X, Y, and Z."

So we won't know until it's there. So I'm not saying that this is AGI, by the way. Like I'm not, I'm really not hyping this to that degree. I know people might think that. I just, it-- I think it's a really difficult conversa- I re- It's a really difficult evaluation to have, and the fact that this model has [00:13:00] achieved a ridiculous score of ninety-nine point nine percent on the ARC-AGI3 test, where the previous best was thirty percent, it's... I don't know, like i-i-it, this, this is moving the target. Uh, uh, from what I've understood, ARC-AGI4 has ha- is coming out, and it's only scoring four percent tied with, I think, uh, Fable 5.1, which is the other like top end model from Anthropic. Um, so I don't know if it's AGI or not, but I can tell you right now that this thing is doing real work for me a day-to-day basis, and i-I am trying to treat it like I would an employee, uh, fro-from that perspective.

It's not maybe directly has all the capabilities of an employee quite yet, so maybe I'm not willing to say it's AGI. But that might not be a, a model problem; that might be a harness problem Like maybe something where I give it more access to a computer and more credentials if it's able to spin up its own GitHub apps or if it's able to spin up its own, you know, [00:14:00] databases and stuff.

I just don't give it that access 'cause I, I still don't have inherent trust. But maybe if I did, I would be m- feeling more AGI, like it's AGI ready. Um, it's just, that's more maybe of a personal thing on my end. Like I, again, as I've said before, I'm kinda slowly different things that it can do and not do

Matt: I think maybe we're, we're also... Y- there's also comes a point here like if, if that's true with the Arc AGI 4 test and it's only scoring 4%, I mean, that's a drastic drop. I mean, that's a, you know, a, a, a perfect grade versus an absolutely abysmal failing grade at the end of the day. Um, and so the, you know, the goalposts are definitely being moved in, in, in some regard.

Uh, I think maybe we're ... A- AGI is like a, is like a fixed point and it's something easy to report on, it's something easy to talk about, but when it comes to real world usage, I don't know how much AGI matters to the layman. It, at the [00:15:00] end of the day does this model do the things that I want it to do better?

Like you mentioned specific professions like lawyers and of course then there's gonna be things like developers and, and game devs and other various like office admin. The list goes on and on different professions. If those professions or th- those professionals at the end of the day are using ChatGPT and they're using this model, they wanna know that their tasks are being done properly and hopefully better than the previous model.

They're not sitting there worrying about AGI. I think AGI at the end of the day really is something to report on. It's something that's sort of considered an achievement point and it is certainly an achievement point for those involved in the actual creation of AI itself. It's something that they're trying to achieve.

It's try-- You know, they're, they're trying to get there. It's kind of the equivalent of, you know, tr- they're trying to get onto the moon. They're trying to hit that, that target right? They're trying to get to.... You know NASA was like you know, "Try to get to the moon." That was the, [00:16:00] that was the goal lik- At the end of the day it was a rudimentary goal that they set said, "We're going to get to the moon," and they go out and they do it right?

And then like this is... There has to be some sort of goal, some sort of goalpost internally and then in this case, it's, you know, we're all, we're all aware of it is these companies GPT models, Fable models, you know Claude, Anthrope. All these companies. These companies Gemini with, you know of course with Google and things, they're all trying to reach this AGI that, that's, that's ultimately what they're all fighting for But for us on the consumer side I do think that it matters less.

I think it's more of does it do my task that I wanted to do amazing amazingly because here's the thing hooray let's say ch- uh GPT seven comes out Hooray! GPT Seven is amazing at climbing Mount Everest

Mikhail: Mm-hmm.

Matt: I don't plan to do that So what else does the thing do Right from a consumer perspective

Mikhail: [00:17:00] We're, we're starting to get to a point of diminishing returns in my opinion on a lot of these models. Like where-- I, first of all we're at, we're at a pace like model release pace of it's, uh, I mean obviously unprecedented because this is a new thing but like it is unprecedented. We're having massive releases again every week.

Like right Fable 5.1 just came out two days ago or something like that then Astra Six launched before that there was like three or four different really good open source models that are low cost that launch that are on like you know three months or a month, a month before you know flagship performance levels.

Like there's a local model that came out that can run on my MacBook that's like at Opus 4.6 levels like there's just an absolute outrageous amount of competition in the space and an absolutely outrageous amount of shipping and like everyone is fighting to be the top the middle the cheapest the fastest now so really at what seems to be an acceleration point with, with the amount of stuff happening And [00:18:00] absolutely i think near the top we're at diminishing returns for the regular person well for the regular person i think we were at diminishing returns from a long time ago honestly For someone that's just trying to chat with it and doing basic things But that's from someone that doesn't understand how...

What these things can do Because when I talked to my dad I think I brought this up before Where my dad was like "Yeah I just asked ChatGPT questions and it gives me answers" And then I sat down with him and I showed him Hey You know you can actually just pull up Codex and have it go in and look at your file and edit your file for you and look at the logs and do this." And he was like oh and then two weeks later He's like yeah That's codex been Doing everything for Me It's going In and doing X Y and Z All this When I saw You use it Learned How to Use it Then Changed The way Work There Is Gap These Model Specifically Apparently Really Good At Why People Are Saying Closer To AGI Computer Use [00:19:00] Part Natural Consumer Uh Use Of Your Computer And natural consumer workflow of opening up Excel spreadsheet Checking the variables in the excel Spreadsheet mathematical Calculations inside Opening Up Pdf computer Logging Into Email Report Boss Stuff Was Doable Let Me Be Like Pretty Shape Version Kodak Out Ten Or Fifty Speed Efficiency Major Focuses Spatial Awareness And so understanding what's on the screen and understanding what's going on, that's why it's so much better at 3D as well. So like if you want it to create a Blender model for you or if you want it to create some game assets or actually like a full game, it's able to do a much more accurate job now because it has better spatial awareness, which is something that these models were missing. And so that also plays into, you know, being able to, you know, process what's going on in the screen [00:20:00] better and act on it and figure stuff out itself. Like it doesn't have to be-- I think one of the examples was like a government -- like some random government website, like being able to pull in logs or like, uh, or documents from a random government website, do some things with it and, and apply for some random permit. Um, that's not stuff that it's gonna be trained at, right? Like no one's giving it that training data. That's stuff that it has to kind of figure out from visual-spatial understanding of what's going on on that website and then accurately, you know, push the right buttons and go through the right workflows to get that done.

So that's apparently the biggest jump in this model. That plus what we're seeing with like terminal and decoding stuff. But I j- That's all to say that like consumers, the general consumer, a lot of them have actually given up on AI early on because it was very much like answer question or question answer, question answer. I get that, but [00:21:00] I do think that there's gonna be kind of like a renaissance from them coming back into it once these things become a little bit more solidified, easier to use, easier to access. Like it's not... It's super expensive, this model. Apparently, it's like sup- efficient as well, so it's not like crazy more expensive than the top model. But still not, you know, general consumer cheap. It's definitely not free, which a lot of consumers usually tend to go towards. So we're not-- I don't think we're gonna see the adoption of it in the same level as we would have, uh, as we would if this was like, you know, available for free in ChatGPT

Matt: We have to also remember something though, is that a lot of people are, are, are going to be using a free model

Mikhail: Yes.

Matt: Gemini. Like Gemini just comes on Android and they're just gonna use something like Gemini Free without caring about the, the pro model and they'll never hit the limits and they'll never bump into the limitations because they'll never push it.

Now I know-- I understand that you're, you're making the point that like, you know, they should be pushing it and things but I think that that might [00:22:00] happen more so at least in the interim in the workplace where they might go, "Oh yeah, when I go to work I have my more capable model that's like fully, you know, paid for" and stuff like this from the corporate account and then I go in and I do all my whatever it is design work or coding work or, or what have you.

Also a lot of the general public has dropped off due to political reasons, due to environmental re-due to environmental concerns, uh, due to, uh, like anger over potentially replacing humans. Big questions, you know big concerns many are, many are completely legitimate and so there's, you know, there's people are voting with their wallet and that's what I tell people.

Like if people bring it up I say v-v-"Do it." If you are anti-AI don't use it, don't pay for it, stay away from it and that's it. Like you're voting with your wallet and we'll see where it goes because what other momentum does the general consumer have? I'd understand you could do, you know protests or what have you within the con-confines of the law within your jurisdiction, I understand that um, but I mean these things are being [00:23:00] built for profit so i-the...

In my opinion anyway, it's a personal opinion in my opinion vote with that wallet. You're anti something? Don't pay for it. You know just as simple even something as simple as,"I don't like Game Pass", don't pay for it.

Mikhail: Yep

Matt: Vote with the wallet right make that because here's the thing is like AI is great I'm enjoying using it but if it was like AIs banned next week it'd be like oh alright and then I would just continue along my merry way doing things the way I did before

Like it wouldn't ,you know it w- I'm not like so enamored by AI that I'm like ,"Oh my God!" You know?"It's the best thing ever". Like uh I..."Oh we gotta go back to the old way"? Alright

Mikhail: I mean, I'd-- It would be a little bit of an adjustment period for me because I've been diving so deep into it, but I would be... Like if it was banned across the board, I would be relieved. That's for sure I would be more than glad to ramp back up in my original, you know, the, the normal way of coding.

Um, but like I, again, I, I'm [00:24:00] deep into the AI workflow stuff because I, I just wanna learn it front and back and make sure that I'm aware of what's going on. Um, but yeah, no, I agree with you. I, I think that it's gonna take some time for this kind of stuff to penetrate the market and, uh, get into the hands of the general consumer.

Businesses will obviously take advantage of it to a certain degree because it is expensive. not everyone's gonna be gain-gaining access to this. but if this is what it sho-... Like again, we have open source models that are now being able to be kinda like three months, four months ago flagship level performance. That's-- If this is the-- If that's the path, then those free people are might, might be getting that kind of, uh, performance soon. Like th-th-that might propagate to the, to the masses and faster than we think, and then if it's able to do all the computer stuff that it's reportedly able to do, it's another tick into the favor of AGI, right?

Like if it, again, the, from me pers-, from my perspective, if it's able to handle [00:25:00] the same workflows that same workload that a human employee is able to handle, then it's reached AGI for that specific task or for tha..., not even for that specific task, for that specific job. And I think I'm better... I'm, I'm more comfortable at calling it like an AGI developer or something than I am at like this is AGI. Um, that's the way I'm gonna

Matt: Which, which i-in and of itself is not AGI because it's the narrow... Like AGI is the non-narrow version. There's like the narrow, narrowed, narrow general intelligence I believe is what it's called, and then there's ar- there's uh, general artificial general intelligence. Um, I've messed up my acronyms there.

But the point is, is one that's narrower. It's like this one is very, very, very good at coding or this one's very, very good at like fixing cars, and then there's the art- like the, the general one where it's like I'm good at cars, and boats, and coding up websites, and driving, and doing this, and all these crazy things

Mikhail: Yeah. Yeah, may-- a-again, again that's probably a wrong thing for me to state that then, [00:26:00] uh, because there is a specific term for it. Maybe it's more of like hey this is able to do most of the work that I would need it to do cross domain just coding but administrative work and personal assistant work and my own stuff that I do personally Like I don't know maybe th-there will be something like that comes up f...like not even comes up But like Again at some point someone's gonna declare AGI And then people rebuke it I need to put a line in the sand In my own life At so-...and Just move on It's not a huge deal I agree I think ASI Is a much bigger deal again we might Not know

Matt: Well, the thing, the thing that you're talking about though is I think it's important that we, w- we even mention this, you know, very, very briefly, is that it... The thing is, is I feel like the general popula- populace, including myself, kind of put AGI on the same scale just sort of mentally as like sentience.

And like there's no way to confirm other people are sentient. I know there's that whole thing where it's like, "I know, I know I'm sentient. Is Mike sentient? Who knows?" Like, is the person listening to this sentient? You can't-- and the person listening, you can't [00:27:00] tell if I'm sentient. It's like this whole, this whole thing.

I feel like the general public, myself included, like without, you know, short of like stopping and thinking about it objectively, like AGI does not necessarily mean sentience. And you mentioned this earlier in the episode. But the thing you're mentioning where you're like, as long as it's AGI for coding, it's like you're kinda thinking of it, um, I'd imagine, correct me if I'm wrong, as like a sentient human that's so that is a developer.

'Cause that developer is n- not necessarily a mechanic and a, and a mountain climber and a great swimmer and a whatever, right? But yet that's still ... Like if we stop and we go wait, and then we get back to the definition of things, it does fall into different camps. It is different categories

Mikhail: Y- y- you're right, you're right. Yeah. I, I, I think m-... Again, at the end of the day it really doesn't matter at th- on the AGI definition side of things, at least for us, the general public. For companies like you said, yes. I mean, I, I think... I'm not sure if this is still the case 'cause OpenAI has redone their deal with Microsoft, but there was a, a contingency point in Microsoft's deal with OpenAI [00:28:00] early on that once they reach AGI, Microsoft's deal kind of ends, and everything moving forward it bel- like, is purely closed source or something, or purely closed to OpenAI so Microsoft can no longer has access to it or something like that. I, I believe that, that, that part of the contract is already gone. I, I could be wrong. But anyway, like t- just to say that there are some very significant stakes probably in a lot of internal, uh, internal labs that give AGI, the definition of AGI a very strong meaning. To us as general consumers or even general laboror-- like workers, doesn't really matter.

Again, for me, I, I... my own definition, not AGI whatever, is going to be something that I could trust as, as well as a human. Um, I don't want it to be I almost don't want it to be like an everything, right? Like I want it to be more scoped down and more narrow, with the ability to have like some of the human characteristics of like [00:29:00] solving a task and reporting it to me and, you know, informing, informing of issues, like being proactive.

Like whatever. Like i- I'm not gonna get into exactly the defini- like exactly the benchmark that I'm gonna have 'cause I don't have one. But that's, the perspective that I would have on like what I want from an "AGI", quote unquote

Matt: Yeah, and I think, I think y- you know, you mentioning the specific... Like, you're, you're saying like, you know, "This is my AGI. This is," you know, "the other person's AGI." It's al-... Like it is really, like almost, like just like a milestone. Like we're treating it as, like this is like, you know, the perfect ... This is the utopian AI for me 'cause if you, if you stopped coding your AGI would change.

If you b- suddenly became a painter you'd be like, "Oh, I would..." You know, I really want it to f-... You really want it to be able to paint. I would imagine you wouldn't because then it would replace you there and it's a whole mess there, but realistically speaking it's like, oh, I w- really want it to suggest the best color palette or the best paintbrushes to do, or whatever, like your moving goalpost is to an [00:30:00] ex-, you know, to an extent, right?

And e- and someone listening to this that knows all about the math and knows all about the physics or what have you is probably like, "You guys are fools", and we're j- we're consumers. I think, you know at the end of the day, like we're using... We're, we're talking about, you know, pretty complex stuff that matters to who it matters to, and then the consumers are taking it and, and they're applying like a layman definition or explanation to it, and I think that that is kind of ultimately what is happening here.

So like to answer that sort of thesis of like what does AGI mean to you? I mean Mike's answered it and I agree with Mike. It's, it really is, it really is like what it, what it means to you. I want this thing to be able to do the tasks that I give it like as if I delegated it to a real assistant that has experience

Mikhail: Yep. And is able to remember stuff and act on, et cetera, et cetera. Like, there's a lot, there's a lot that comes with that general statement, but for, like, that's, that's the overview of what I think I will label as AGI for myself

Matt: 100%. Well, like even just [00:31:00] something as simple as, like you know that a real person is gonna make assumptions. Like they, they know that like, you know m- like me, I like to have all the numbers and stuff in front of me so when they send me a report it better not just be one sentence. It better be okay there better be a table or something in there so I wanna be able to read and like

'Cause then I use that, I extrapolate it, and then I come up with new orders or new tasks or new ideas. Um, but like I don't know Mike could give you... Your report might be more, I'm just making it up but it's like your m- your report could be more artsy or something. That's something that's not really trained so much as the person realizes, "Okay Matt wants the chart and Mike wants the art or whatever."

Right? And then that's something that you would expect from an AGI. In fact, like months and months and months ago I was testing out Copilot, not GitHub Copilot just Copilot and it told me it was like, "Because you're precise I'm going to present it like this." And I was like, what? I'm precise

Mikhail: Apparently

Matt: So yeah.

That's it. Uh, we'd love for you to answer that question, answer that sort of thesis question whatever. What does AGI mean [00:32:00] to you? You know, what is AGI to you? What does it, what does it mean? Does the... You know, does this, you know, GPT-6, uh, Astra almost, I almost said Opus, that's the, that's the last one. Uh, does this GPT-6 Astra matter to you?

Do you care about these AGI tests or do you only want the GPT, the AI, the model, the whatever it is you're using? Do you just want your AI to do better, uh, iteratively like any other product you would, you would use? But that's it. Uh, that is the web news and we're signing off.

Mikhail: Goodbye

Matt: Goodbye