TRANSKRIP 🔗
WEBVTT
00:00:00.959 --> 00:00:01.520
17?
00:00:01.840 --> 00:00:02.319
Mm-hmm.
00:00:02.879 --> 00:00:03.200
Okay.
00:00:03.919 --> 00:00:04.879
Ron, are you ready?
00:00:05.040 --> 00:00:05.599
Oh yes.
00:00:05.759 --> 00:00:06.000
Okay.
00:00:06.559 --> 00:00:20.719
Because this is it is just a model, a series of ongoing conversations on the interstitium of innovation, design, philosophy, and tech with my friend Ron Kesik and myself, Ace Nakers.
00:00:21.519 --> 00:00:30.079
And apparently I don't need any topics because you will probably butt in anyway.
00:00:33.759 --> 00:00:34.799
So, all right.
00:00:35.119 --> 00:00:49.200
So we just came out with the news because the the American government they basically said, oh, you got uh the kids version of um uh what's the what's the thing called?
00:00:49.359 --> 00:00:51.119
No mythos, mythos.
00:00:52.079 --> 00:01:04.640
And uh it seemed to be able to hack like some obscure uh in some obscure way, and uh therefore Anthropic has been shut off, basically shut off their model.
00:01:04.879 --> 00:01:12.319
And they said you can't use it anymore because it's unsafe for our for us, but also an unfair advantage for our enemies.
00:01:12.640 --> 00:01:14.400
No, it's an export restriction, right?
00:01:14.480 --> 00:01:22.319
So not non-US citizen cannot use the model, yes, which is probably half of the staff of anthropic.
00:01:23.280 --> 00:01:33.280
And there's it's probably also um pretty much all their customers because most of the customers are outside of the US, I'm guessing.
00:01:33.840 --> 00:01:34.640
I'm not sure actually.
00:01:34.799 --> 00:01:44.640
No, no, but um to me it sounded like why would they do that to in two twofold?
00:01:45.200 --> 00:01:56.719
And my conclusion was they tried to force a uh they try to force a situation where legislation is uh is coming up for AI.
00:01:57.200 --> 00:02:10.159
So they basically want to make sure that everybody is uh included that plays the game, and everybody does that doesn't play the game is excluded uh from uh the US.
00:02:10.400 --> 00:02:16.479
Who is they uh the the the people that are the people that are excluded?
00:02:17.199 --> 00:02:19.599
No, no, who is they that want to exclude people?
00:02:19.680 --> 00:02:21.520
Is it anthropic or the US government?
00:02:21.680 --> 00:02:22.240
Yes.
00:02:22.800 --> 00:02:24.719
Okay, both one through the other.
00:02:24.960 --> 00:02:30.879
Yeah, I mean, because an anthropic want wants their competitors to be regulated, definitely, yeah, yeah.
00:02:31.199 --> 00:02:45.840
So but and also and it seems like I I've heard this in a uh some fragments of a podcast, it seems also be the uh people working on open source models, uh, in particular the Chinese open source models.
00:02:46.159 --> 00:02:51.759
Open weight, they they want to exclude them from the market, basically, or at least the American market.
00:02:52.719 --> 00:02:53.439
I'm not sure.
00:02:54.560 --> 00:03:07.919
But um yeah, that's the only thing I found reasonable why they would shut off such a model because I I didn't see the that big of a risk in the way in the things it does.
00:03:08.159 --> 00:03:12.479
Well, apparently, I mean you call it what what was it again, the kiddie version of of uh Mythos?
00:03:12.800 --> 00:03:14.639
Yeah, it was the where the hampered version.
00:03:14.879 --> 00:03:25.199
And apparently you could jailbreak Fable 5 so that it gets back to Mythos level capabilities in uh in uh in cybersecurity and uh biology.
00:03:25.360 --> 00:03:25.680
Yeah.
00:03:25.840 --> 00:03:34.960
So apparently because a competitor of uh Anthropic showed the United States government how that is possible, yeah.
00:03:35.280 --> 00:03:42.960
So it's the a certain unnamed competitor, open AI, yeah.
00:03:43.039 --> 00:03:46.719
Well I I have to or it might also be Google or whomever.
00:03:47.199 --> 00:03:54.000
And it's uh it's uh it's a government that thinks in zero sum um um words about the world, yeah.
00:03:54.159 --> 00:03:54.400
Right?
00:03:54.560 --> 00:03:57.680
If somebody else uh wins, then it must mean we lose.
00:03:57.919 --> 00:03:58.479
Yeah, definitely.
00:03:58.560 --> 00:03:58.879
Yeah, yeah.
00:03:59.120 --> 00:04:05.759
So it's the uh I found this both very expected and very boring.
00:04:06.000 --> 00:04:06.400
Yeah.
00:04:06.719 --> 00:04:24.160
Yeah, to me it was I just wanted to know why somebody would do it, but also let's say if it's intentional, then it's probably well played, entropic, yeah, forcing the situation into getting your ways anyway.
00:04:24.480 --> 00:04:25.519
IPO coming up, right?
00:04:25.680 --> 00:04:30.800
And we have technology that is so powerful that the United States government banned it.
00:04:30.959 --> 00:04:31.759
Come and watch us.
00:04:32.079 --> 00:04:32.720
Yeah, yeah, yeah.
00:04:32.800 --> 00:04:35.040
It's great, uh, great uh marketing material.
00:04:35.279 --> 00:04:40.000
Yeah, I mean it's uh it's like putting explicit lyrics on your on your album sleeve, right?
00:04:40.079 --> 00:04:41.600
And then hoping that people won't buy it.
00:04:42.000 --> 00:04:48.160
Yeah, I think uh the so research also showed this is also the same as the packaging of cigarettes.
00:04:48.480 --> 00:04:54.079
So these are in the Netherlands at the least, they're very off-putting or were off-putting, I'm not sure.
00:04:54.240 --> 00:05:06.399
But anyway, so the packaging was very uh graphic, if you will, and um that seemed to be that seemed to attract people because it's cool to live dangerously, I guess.
00:05:06.879 --> 00:05:08.800
Yeah, well, I can see that.
00:05:09.040 --> 00:05:15.120
Yeah, but uh so it didn't uh didn't really impress you, uh the the entire exercise, not at all.
00:05:15.360 --> 00:05:21.040
It's the uh it's the uh because this is the boring side of tech, to be honest.
00:05:21.199 --> 00:05:22.639
Yeah, at least boring to me.
00:05:22.879 --> 00:05:26.240
Well, I guess it's uh like an up
Load full transcript
ramp to really boring stuff, right?
00:05:26.319 --> 00:05:43.759
It's they're building these AIs, of course, to for uh for companies to also use, and to me it sounded a little bit like the um um the the play uh from uh the 92s or something where meshups were really a thing, yeah.
00:05:44.000 --> 00:05:58.959
So everybody was building and combining all kinds of big huge platforms, and at a certain moment, when everything was set in, uh everybody had their motors running and their commercial engines uh running.
00:05:59.600 --> 00:06:07.839
One by one, the big platforms basically shut off the free or or any access to their platforms and built their own, right?
00:06:08.000 --> 00:06:13.600
Replaced, for example, Twitter replaced their search API with their own interface.
00:06:13.759 --> 00:06:25.360
Uh, similar with Facebook, the web interface does exist, but let's be honest, they rather have us use the app, make it as appealing to the app, and and they you can't leech it, right?
00:06:25.439 --> 00:06:30.480
So you can't combine news with Facebook, with Twitter, etc.
00:06:30.879 --> 00:06:33.439
This was the mashup mindset.
00:06:33.759 --> 00:06:36.959
And I'm thinking this might also be the case for AI.
00:06:37.279 --> 00:06:50.000
Uh the combination the combination of all kinds of different agentic and also generative AI might be very appealing at the moment, but if you build your commercial model on it, i.e.
00:06:50.240 --> 00:06:56.000
getting to the boring stuff, yeah, that might be a little bit dangerous if it's not regulated.
00:06:56.240 --> 00:07:06.879
So I'm guessing that might also be an interesting uh or it might also be a direction there why they want to go towards regulation, not just exclusion.
00:07:07.360 --> 00:07:08.560
Again, who who is they?
00:07:08.879 --> 00:07:10.879
The the the producers of the AIs.
00:07:11.199 --> 00:07:15.279
Yeah, I mean what what they want uh they want to use regulation as a competitive mode.
00:07:15.360 --> 00:07:18.560
Yeah, I mean that's the that's the only motive they have, right?
00:07:18.720 --> 00:07:22.399
I mean they they have an advantage, they think, and they probably have.
00:07:22.560 --> 00:07:26.079
Yeah, so let let's not freeze any everything, right?
00:07:26.160 --> 00:07:28.720
And because this is dangerous, let's stop any development.
00:07:28.959 --> 00:07:30.079
Now that we are ahead.
00:07:30.240 --> 00:07:44.560
Yeah, that's the uh yeah, but also think the legislation as being predictable, reliable, and really get uh build a solid commercial foundation on top of the basically the research models that we have now.
00:07:44.800 --> 00:07:57.040
I mean, for another company, let's say it's not I think it's not for not that um um uh Microsoft got their own model, for example, for LinkedIn and their products.
00:07:57.199 --> 00:08:04.959
They built their own model, they're not dependent on, as far as I know, not dependent on OpenAI or any other model generator.
00:08:05.040
--> 00:08:07.600
Uh I have no idea what they're doing in LinkedIn.
00:08:07.920 --> 00:08:13.600
No, yeah, but let's say Office, you know, that it's it's all it's all their own model as far as I know.
00:08:13.920 --> 00:08:14.720
No, it's not.
00:08:14.959 --> 00:08:20.399
It's not no, it's that they're uh that they they have been and they still are relying on open AI stuff.
00:08:20.639 --> 00:08:26.560
I thought they they stopped that uh that their licensing because they had no, they want to stop it.
00:08:26.720 --> 00:08:29.600
Yeah, because they have conflicts and stuff like that, right?
00:08:30.639 --> 00:08:34.639
No, I mean not interpersonal conflicts, but I meant commercial conflict conflicts.
00:08:34.879 --> 00:08:40.240
No, they I think they don't want to be dependent about a company they they don't control the board of.
00:08:40.320 --> 00:08:40.559
Yeah.
00:08:40.720 --> 00:08:40.960
Right?
00:08:41.039 --> 00:08:47.440
I mean, nothing to do with legislation or safety or whatever, it's just you're dependent on a party you cannot control.
00:08:47.600 --> 00:08:47.919
Yeah.
00:08:48.080 --> 00:08:48.799
That's it.
00:08:49.120 --> 00:08:56.559
Yeah, well, I think legislation and stuff like that also leads into being more predictable as a as a commercial partner.
00:08:56.799 --> 00:09:01.759
So so you can do stuff like well, I mean having part integrated partnerships like that.
00:09:02.000 --> 00:09:08.399
If the legislation doesn't prevent it, doesn't help prevent the provider of the service to switch to switch it off.
00:09:08.480 --> 00:09:10.960
I mean, I don't see any reliability at all.
00:09:11.120 --> 00:09:12.159
No predictability.
00:09:12.399 --> 00:09:12.799
Yeah, yeah, yeah.
00:09:12.879 --> 00:09:13.360
No, no, no.
00:09:13.440 --> 00:09:18.320
It's the I mean the the biggest risk is just that your provider says no more soup for you.
00:09:18.639 --> 00:09:19.519
Yeah, exactly.
00:09:19.679 --> 00:09:20.080
Yeah.
00:09:20.320 --> 00:09:25.279
And uh whether whether they do it or they are linking the show globes, no globes.
00:09:25.519 --> 00:09:26.960
Yeah, that's a that's a great one.
00:09:28.080 --> 00:09:28.240
Right.
00:09:28.399 --> 00:09:29.440
I mean Seinfeld.
00:09:29.600 --> 00:09:30.000
Yeah, yeah.
00:09:30.240 --> 00:09:33.519
Or they are forced by the by the United States government, right?
00:09:33.600 --> 00:09:36.320
I mean, I think this is this this is the biggest thing.
00:09:36.639 --> 00:09:46.559
Yeah, similar to their I mean, I don't want to go to the weapons uh industry, but do you think it's similar to the defense industry?
00:09:46.799 --> 00:10:03.440
I mean, uh so the that they switched it off very very much sounded, or at least from everybody outside of the of the US, kind of like sounded like the uh the law, the export laws you had for encryption uh uh in the in the past.
00:10:03.759 --> 00:10:07.759
Yeah, too bad, too bad you cannot tattoo the weights of a model on your back, right?
00:10:07.840 --> 00:10:10.399
I mean, yeah, like the guy did.
00:10:10.480 --> 00:10:11.759
What was what was it again?
00:10:12.000 --> 00:10:12.720
PGP.
00:10:13.440 --> 00:10:15.039
I I don't know, I don't know the story.
00:10:15.120 --> 00:10:22.480
Uh okay, so that was uh uh I think it was pretty good privacy, so in encryption algorithm, and that was also export restricted.
00:10:22.720 --> 00:10:23.200
Okay, great.
00:10:23.440 --> 00:10:27.120
So you I mean because but how how can you not export an algorithm?
00:10:27.360 --> 00:10:31.039
So so a guy, I think I think it was the the Pearl code.
00:10:31.120 --> 00:10:35.440
I but the the code of the algorithm tattooed on his back.
00:10:35.679 --> 00:10:40.799
Oh so by him leaving the country, he was actually exporting the algorithm.
00:10:41.120 --> 00:10:41.679
That's awesome.
00:10:41.840 --> 00:10:43.360
Because well this was quite deliberate.
00:10:43.440 --> 00:10:53.279
I'm that was I'm guessing if you don't tattoo PGP on the on the on your back if you're just browsing to the to a book, right?
00:10:53.600 --> 00:10:56.240
A dragon, a dragon, an eagle.
00:10:56.399 --> 00:10:58.000
No, this tool is not based.
00:10:58.080 --> 00:10:58.320
What is it?
00:10:58.480 --> 00:10:59.200
An algorithm.
00:10:59.279 --> 00:11:00.159
Oh, yeah, let's do that.
00:11:01.120 --> 00:11:01.759
Interesting.
00:11:02.000 --> 00:11:04.720
You don't say my friends will think I'm so cool.
00:11:04.799 --> 00:11:09.519
It's the yeah, yeah, that's it probably the same, the the same play again.
00:11:09.600 --> 00:11:12.000
It's the uh as you can hear my voice.
00:11:12.080 --> 00:11:13.519
I find this wholly uninterested.
00:11:13.679 --> 00:11:14.080
Yeah, yeah.
00:11:14.159 --> 00:11:16.399
Well, please interest us in the stuff then.
00:11:16.720 --> 00:11:18.639
Well, episode 16.
00:11:19.360 --> 00:11:20.240
Episode 16.
00:11:20.399 --> 00:11:22.399
Yes, that was a really nice one.
00:11:22.639 --> 00:11:24.720
Yeah, I got a lot of memes uh from that.
00:11:24.960 --> 00:11:26.960
Yeah, it's the um that was a really interesting.
00:11:27.039 --> 00:11:28.399
I mean, so in typical fashion, right?
00:11:28.480 --> 00:11:33.600
We we we we zig and we zag and up and down and yeah, like puppies chasing a bus.
00:11:33.919 --> 00:11:39.919
Yeah, and somehow, typically in the final third, we end up on something meaningful.
00:11:40.559 --> 00:11:45.519
Yeah, and I guess on the way we also hit on uh some interesting thoughts at the least.
00:11:45.759 --> 00:11:52.960
But I think the uh the world building uh thing, the uh is uh uh AI engineers artists, right?
00:11:53.120 --> 00:11:57.440
Growing AI just like you well, make wine, for instance.
00:11:57.519 --> 00:11:59.200
Yeah, it's all nice metaphors.
00:11:59.279 --> 00:12:03.440
No, and the reason why I came to that one because I made a spelling error in my snow globes.
00:12:03.679 --> 00:12:08.480
Oh, so instead of world building, yeah, I said word
building.
00:12:08.720 --> 00:12:09.519
Oh, okay.
00:12:10.080 --> 00:12:15.440
That will probably appeal to your uh to what you do often.
00:12:15.679 --> 00:12:30.240
No, I mean and and I think why why that one actually so resonated because I actually realized it's the um um so I I th I for the all all peop many people probably have this, right?
00:12:30.320 --> 00:12:32.320
But I have errors in my career, right?
00:12:32.399 --> 00:12:35.679
Or epochs or yeah, periods, whatever you want to call it, right?
00:12:35.840 --> 00:12:42.000
Yeah, I think for the past 10, 12 years, I had a certain error, right?
00:12:42.159 --> 00:12:47.360
Right, and every time I enter a certain error, I'm looking for words and language.
00:12:47.519 --> 00:12:48.159
Yeah, yeah, yeah, yeah.
00:12:48.480 --> 00:12:49.600
I can see where it's going, right?
00:12:49.759 --> 00:12:58.320
I mean, that's the uh and my previous error, let's call it I don't know, the platform thinking error, but whatever, right?
00:12:58.399 --> 00:13:16.240
I mean, that started with the uh KuCon conference in San Francisco, seeing the Netflix people present, and that's where the first time I saw actual software engineers, actual damn good software engineers, yeah, able to communicate, which was a totally alien concept to me.
00:13:16.559 --> 00:13:23.120
Okay, so I thought these people are excellent software engineers and they can communicate, and these things are related.
00:13:23.279 --> 00:13:24.159
Yeah, yeah, yeah.
00:13:24.559 --> 00:13:28.320
Going from your own experience in the Netherlands, yeah, yeah, yeah, yeah.
00:13:28.399 --> 00:13:39.039
No, I mean, I mean, I'm I'm not to yeah, I mean it so it sounds a bit pretentious, but for many people I couldn't be a proper software engineer because I could somewhat communicate.
00:13:39.200 --> 00:13:42.080
Yeah, or actually, I wanted to somewhat communicate.
00:13:42.240 --> 00:13:42.720
Yeah, yeah, yeah.
00:13:42.960 --> 00:13:44.559
I wanted somewhat to communicate, right?
00:13:44.639 --> 00:13:48.799
I mean, it's uh it's the uh but in a certain amount, but also in a certain fashion.
00:13:49.039 --> 00:13:49.759
Yeah, yeah, yeah.
00:13:49.840 --> 00:13:50.799
And there they were, right?
00:13:50.879 --> 00:13:52.240
So that there was a big eye opener.
00:13:52.480 --> 00:13:58.720
We had the SX conference also in San Francisco, the service experience conference organized by Adaptive Path.
00:13:59.039 --> 00:14:00.720
Okay, late Adaptive Path.
00:14:00.879 --> 00:14:09.039
Yeah, just mind-blowing to me how these people were talking about well, service experiences, that experiences are something you can design.
00:14:09.360 --> 00:14:09.919
Yeah, definitely.
00:14:10.000 --> 00:14:10.879
Yeah, sure.
00:14:10.960 --> 00:14:13.360
I mean my previous uh occupation.
00:14:13.759 --> 00:14:17.519
But this was 2010-ish, 2012.
00:14:17.919 --> 00:14:24.720
The first time I went to the surf desk conference, so that's a academic conference about service design, right?
00:14:25.120 --> 00:14:26.399
Just mind-blowing.
00:14:26.559 --> 00:14:30.000
Yeah, actually, to actually see how people respond to each other.
00:14:30.159 --> 00:14:35.600
Somebody does a presentation and somebody stands up and says, Thank you for your wonderful presentation.
00:14:35.759 --> 00:14:40.559
Yeah, I wasn't used to that one because I was used that people would stand up and say this was crap.
00:14:40.879 --> 00:14:41.440
Yeah, exactly.
00:14:41.600 --> 00:14:42.879
Yeah, yeah.
00:14:42.960 --> 00:14:45.840
Well, that can also be uh happen in academic worlds, yeah.
00:14:45.919 --> 00:14:46.720
Oh yeah, totally.
00:14:46.799 --> 00:14:49.360
But that's the uh no, I mean, but but I I kept thinking, right?
00:14:49.440 --> 00:14:53.919
I mean, it's and I can I can remember that day and age, I was looking for words, right?
00:14:54.080 --> 00:14:59.919
Right, like for instance, but it sounds a bit silly, but experience, yeah, and experiences, right?
00:15:00.080 --> 00:15:01.120
I mean, that was that was acting.
00:15:01.200 --> 00:15:11.600
I mean, I my team back then at the bank I was working, I sneakily called the Office for Experience Design, OXD.
00:15:12.240 --> 00:15:17.519
I mean, that was really but so those were words I was donning as a code, so to speak, right?
00:15:17.600 --> 00:15:18.879
Yeah, and now it seems obvious.
00:15:19.440 --> 00:15:24.480
And now I I realize I'm in exactly that same period.
00:15:24.799 --> 00:15:28.000
I'm looking for words or language, yeah.
00:15:28.159 --> 00:15:32.639
Like, for instance, building an A model is that yes or no?
00:15:32.799 --> 00:15:35.679
Like, well, making wine, for instance, right?
00:15:35.840 --> 00:15:36.960
That discussion we had.
00:15:37.200 --> 00:15:44.399
Yeah, can we can we use the word building or should we use use growing or fermenting or yeah, yeah, yeah.
00:15:44.480 --> 00:15:53.360
I mean, and and it's and it sounds silly, but I think this is this is what I'm been trying to do for the past year or so, trying to find uh uh find find words.
00:15:53.519 --> 00:16:01.840
And recently uh I found another word which I think is probably the most descriptive one yet.
00:16:02.159 --> 00:16:02.480
Okay.
00:16:04.320 --> 00:16:05.200
Come on.
00:16:05.440 --> 00:16:06.639
No, keep us waiting.
00:16:06.879 --> 00:16:09.279
And the word is exoskeleton.
00:16:09.759 --> 00:16:10.399
Yes.
00:16:12.639 --> 00:16:13.679
Any more questions?
00:16:13.840 --> 00:16:14.960
No, no, no, no.
00:16:15.039 --> 00:16:22.000
Well, I mean, an exoskeleton implies like uh kind of like an addendum to what you already have, right?
00:16:23.039 --> 00:16:27.519
From and it being outside your your corpus, yeah, but outside of your body.
00:16:27.759 --> 00:16:28.720
Yes, instead of inside.
00:16:28.879 --> 00:16:29.440
Yeah, exactly.
00:16:29.600 --> 00:16:29.759
Right?
00:16:29.840 --> 00:16:38.480
And that's the and uh I mean, and and so two two weeks ago I
was at a conference in London uh about AI, and I noticed, and and we talked about this, right?
00:16:38.639 --> 00:16:49.360
Yeah that up till now in software, all the decisions, all the intent, all the taste, all the well requirements are manifest in code.
00:16:49.519 --> 00:16:52.720
Yeah, right, and code is the central artifact.
00:16:53.039 --> 00:16:59.919
Yes, yeah, yeah, yeah, right, and everything is in there, and and and and and our whole software development lifecycle is built around code, right?
00:17:00.080 --> 00:17:07.440
Everything we do in front is just to make sure that our requirements or intent or understanding of the world ends up in there somehow.
00:17:07.519 --> 00:17:12.960
Yeah, and everything after that is to make sure that we validate that indeed it ended up in there somehow.
00:17:13.039 --> 00:17:13.519
Yeah, right.
00:17:13.599 --> 00:17:16.079
So it's yeah, so the skeleton is inside.
00:17:16.640 --> 00:17:20.160
And what you now do is you pull it out, right?
00:17:20.400 --> 00:17:38.319
You you make your architecture manifest, understandable for an agent, your uh uh your design system, you make that explicit that an agent can actually understand it and act on it instead of infer it from your your code, right?
00:17:38.400 --> 00:17:39.920
So you and also being flawed.
00:17:40.160 --> 00:17:41.759
It's kind of being flawed, right?
00:17:41.839 --> 00:17:49.119
If you if you make code for better and for worse, it's an abstraction layer for what you actually want.
00:17:49.759 --> 00:17:52.480
You encode it with a limited language, if you will.
00:17:52.720 --> 00:17:53.759
Yeah, it's a limited language.
00:17:53.839 --> 00:17:58.720
I'm not sure if it is an abstraction layer, it's actually isn't what's the opposite of abstraction?
00:17:59.119 --> 00:18:00.000
Generalization?
00:18:00.240 --> 00:18:01.839
No, I don't know.
00:18:02.559 --> 00:18:04.400
Uh make it less abstract.
00:18:04.480 --> 00:18:06.480
Well, it's more so more concrete.
00:18:06.720 --> 00:18:10.960
Yeah, I mean, because it has to be concrete for this thing that needs to execute called the computer.
00:18:11.119 --> 00:18:11.839
Yes, exactly.
00:18:12.000 --> 00:18:12.160
Right?
00:18:12.240 --> 00:18:22.960
So it's very strict things, yeah, but nevertheless, for example, uh object object-oriented programming is a way to uh make an abstracted model of the thing that you assure.
00:18:23.359 --> 00:18:25.920
For sure was the intent, yeah, and failed horribly.
00:18:26.480 --> 00:18:31.920
Well, it was a step in our evolution and definitely an artifact that we should leave behind.
00:18:32.160 --> 00:18:36.559
Yeah, yeah, it's the for sure was a branch in the evolution of this, right?
00:18:36.720 --> 00:18:40.000
So but it's but and then but this is really what it is.
00:18:40.079 --> 00:18:45.359
You pull everything outside to the outside so that it is legendable for the agent, right?
00:18:45.440 --> 00:18:48.079
And what I found so it's the
exoskeleton, right?
00:18:48.160 --> 00:18:49.519
It's at the outside, okay.
00:18:49.680 --> 00:18:53.839
And what's cool about an exoskeleton is you rear the exoskeleton, right?
00:18:53.920 --> 00:18:59.920
Yeah, so you can so you can jump higher and lift more things, so you can build worlds better.
00:19:04.160 --> 00:19:12.079
So you're literally seeing um images of uh the the thing with the blue man group uh kind of people.
00:19:12.319 --> 00:19:13.519
Uh come on.
00:19:14.240 --> 00:19:19.839
The the the three films with uh hyper-realistic uh CGI.
00:19:20.079 --> 00:19:21.920
Oh, the uh avatar.
00:19:22.079 --> 00:19:23.359
Avatar, that's it.
00:19:23.599 --> 00:19:24.240
Yeah.
00:19:25.039 --> 00:19:30.799
So kind of like the exoskeletons that we have of or the suit that Tony Stark is wearing, right?
00:19:30.880 --> 00:19:34.960
Yeah, or the the really big mover that uh Corny Weaver uh wore.
00:19:35.680 --> 00:19:37.119
That's one in aliens, yeah.
00:19:37.279 --> 00:19:38.000
That's one, right?
00:19:38.079 --> 00:19:40.480
I mean Mech Warrior, those kind of things, right?
00:19:40.720 --> 00:19:44.000
But no, but it's and you see it everywhere.
00:19:44.240 --> 00:19:46.400
So what is the exoskeleton to you?
00:19:46.480 --> 00:19:52.240
Is it is it words, is it like no language, or is it more the the what happens after?
00:19:52.559 --> 00:20:03.920
It's everything we use to communicate to to ourselves, our understanding of the world, yeah, the problem we are trying to address in our world, and how we think it should be addressed.
00:20:04.079 --> 00:20:05.359
Yeah, so everything, right?
00:20:05.440 --> 00:20:06.559
And that is words, right?
00:20:06.640 --> 00:20:08.240
That's requirements, documents.
00:20:08.319 --> 00:20:11.279
Yeah, but and but this is the cool thing about this technology, right?
00:20:11.359 --> 00:20:13.440
It consumes anything and everything.
00:20:13.519 --> 00:20:16.000
Yeah, it could even be a doodle on a whiteboard.
00:20:16.160 --> 00:20:16.799
Definitely, yeah.
00:20:17.039 --> 00:20:19.839
Maybe it doesn't have the fidelity needed, but still, right?
00:20:19.920 --> 00:20:26.160
A doodle on a whiteboard is now something legendable in the whole creation process, right?
00:20:26.240 --> 00:20:29.839
Not this this this this spurious moment and now it's gone.
00:20:30.079 --> 00:20:32.559
No, yeah, because once you go to the code, it's gone.
00:20:32.720 --> 00:20:33.039
Exactly.
00:20:33.119 --> 00:20:45.839
Yeah, it's an it's a first-order construct of expression to us, oh, and by implication the agent, or if you want, for the agent and by by happy uh circumstance, also for us, right?
00:20:45.920 --> 00:20:48.319
Because it actually makes it better for us too.
00:20:48.480 --> 00:20:50.480
That's what I find so fascinating.
00:20:50.720 --> 00:20:55.920
But it makes us better, or it it's more useful for us, it's more useful for us as a communication tool.
00:20:56.160 --> 00:20:58.000
Because I think
it makes us better as well.
00:20:58.160 --> 00:21:06.400
Yeah, I mean, and not better prompters, as we now would say, but I think it would make us make us better communicators.
00:21:06.640 --> 00:21:07.519
Yeah, and neat, right?
00:21:07.599 --> 00:21:09.279
So so or we can better, yeah.
00:21:09.359 --> 00:21:20.799
I mean, so to because yesterday a better also link is no cload about a uh a designer that said that uh now she's working with AI for the first time.
00:21:21.039 --> 00:21:31.279
Design systems are actually needed, as in they need to be there instead of we have a design system, uh yeah, great design system, you know as like a reference, right?
00:21:31.519 --> 00:21:38.000
As a reference, as an intention, it's cute, but I'm coding and you know what, as instead of a I'm going to use a space anywhere.
00:21:38.079 --> 00:21:40.160
And I I just so who's for?
00:21:40.240 --> 00:21:42.240
Who's the design system for?
00:21:42.480 --> 00:21:43.200
For the agent.
00:21:43.759 --> 00:21:46.880
So the agent understands the design system, right?
00:21:46.960 --> 00:21:54.720
So the design as a system, and it makes sure that anything the agent creates, code or otherwise, adheres to the design system.
00:21:54.960 --> 00:21:56.880
Well, this is it this is very interesting.
00:21:57.039 --> 00:21:59.759
So you're saying a entity that creates.
00:22:00.640 --> 00:22:09.920
Experiences should read or abide by those design uh structures, those design style guides, etc.
00:22:11.200 --> 00:22:19.519
But now we're just saying, well, at the least the A because of we we now have AI, we're making it for them.
00:22:19.680 --> 00:22:23.440
No, and to me, it's like no, no, no, this is exactly the it was meant for humanity.
00:22:23.599 --> 00:22:25.200
Yeah, but if I mean is it a chicken neck thing?
00:22:25.279 --> 00:22:27.279
No, I mean, one of the reasons why you need to make it, right?
00:22:27.440 --> 00:22:36.480
Because if you give agents the code, right, but with all these assumptions hidden and people adding their own interpretation of the design system, yeah, and bevility.
00:22:36.799 --> 00:22:38.640
Well, I mean, yeah, yeah, ambivality.
00:22:38.880 --> 00:22:42.720
The agent reads that, yeah, right, take that as gospel, yeah, yeah, yeah.
00:22:42.880 --> 00:22:47.359
And so iteration after iteration, you get the interpretation of the system inside of the system.
00:22:47.440 --> 00:22:53.440
So it starts diverging widely, and then we said, Oh, bad AI, but it's actually it's us.
00:22:53.599 --> 00:22:54.400
Yeah, right?
00:22:54.799 --> 00:22:57.839
It's uh it's it's a bit like those time-lapse movies, right?
00:22:57.920 --> 00:23:00.240
Where you see, I don't know, a garden, right?
00:23:00.400 --> 00:23:05.039
And then they speed it up, and you actually can see how a garden is overgrowing, yeah, right.
00:23:05.200 --> 00:23:06.400
And you say, Oh, that's bad.
00:23:06.480 --> 00:23:08.960
But if you just do it in real time, you don't
see it, right?
00:23:09.039 --> 00:23:09.279
Right.
00:23:09.440 --> 00:23:11.920
And now we have this spat of version of reality.
00:23:12.079 --> 00:23:12.960
So now we can see it.
00:23:13.119 --> 00:23:13.920
Now we can actually see it.
00:23:14.079 --> 00:23:17.599
Yeah, we see what happens if you don't make yeah.
00:23:17.680 --> 00:23:24.720
So in in order to prevent that, right, the problem we caused, yeah, we now do this for the agent, which we do for us.
00:23:24.960 --> 00:23:25.359
Yeah.
00:23:25.599 --> 00:23:28.079
So this this this almost symbiosis, right?
00:23:28.160 --> 00:23:31.359
What's good for them is good for us, what good is for us is good for them.
00:23:31.519 --> 00:23:32.880
So that's fascinating.
00:23:33.119 --> 00:23:48.720
Yeah, but if you if you post it like yes, and if you post it like this, I don't want to be contrarian, but um uh if you say it like this, I would think the how do you call it the style guide?
00:23:48.799 --> 00:23:51.119
No, a design system.
00:23:51.839 --> 00:24:07.440
If you make a design system, it should be an uh uh an agent on itself, it should be an agent itself because maintaining a design system is an activity, it's not a document to to adhere by.
00:24:07.680 --> 00:24:22.319
So I'm thinking so let's say this sped up version of things overgrowing, then you need a like an automated uh grass cutter that basically cuts the grass frequently in the shape that we agreed upon.
00:24:22.480 --> 00:24:25.039
Yeah, based on the layout of the garden that you provided.
00:24:25.519 --> 00:24:25.839
Exactly.
00:24:25.920 --> 00:24:29.119
So the design system is also the grass cutter itself.
00:24:29.359 --> 00:24:30.160
I don't think so.
00:24:30.400 --> 00:24:35.440
It it it should be something that's well it's that that works independently, right?
00:24:35.759 --> 00:24:39.920
Well, it's it's a medium uh as well, yeah.
00:24:40.160 --> 00:24:40.960
It's a system.
00:24:41.200 --> 00:24:50.079
No, I mean the reason why you see me smiling because I always uh take that transcript for a podcast and I give it to Chat GPT or Claude.
00:24:50.240 --> 00:24:52.799
Oh just what what do you think about this one?
00:24:53.200 --> 00:24:55.359
I thought Ron was an artisan, guys.
00:24:55.599 --> 00:24:59.920
No well, I still make my own snow globes, but just to see what it knows.
00:25:00.000 --> 00:25:06.000
I mean, yeah, and the last time it said uh Pete is a productive contrarian, definitely, yeah, yeah.
00:25:06.240 --> 00:25:10.799
And Ron has sharp something sharpish.
00:25:10.960 --> 00:25:14.079
Okay, it's the uh so it just gives you compliments along the way.
00:25:14.319 --> 00:25:16.480
No, it's no, it's it's it's it's observations, all right.
00:25:16.720 --> 00:25:20.720
I mean it's uh uh uh depends on you on what what you think that's a compliment.
00:25:20.880 --> 00:25:23.839
Um definitely uh well a functional contrarian, yeah, definitely.
00:25:24.079 --> 00:25:25.680
But but that's what you can see to challenge.
00:25:25.920 --> 00:25:28.640
No, I mean but that's why I said medium set of agents, right?
00:25:28.720 --> 00:25:31.119
I mean it's the I don't think it has agency.
00:25:31.599 --> 00:25:32.240
Oh, okay.
00:25:33.039 --> 00:25:40.240
Well, uh, I mean uh the to me it's more like it uh fulfills a it does a function rather than it is a reference.
00:25:40.480 --> 00:25:41.039
Oh, yeah, totally.
00:25:44.160 --> 00:25:50.720
Something that's static, something that doesn't grow or it's uh it's something that it's something that gives rise to worlds.
00:25:50.880 --> 00:25:51.279
Yeah, exactly.
00:25:51.680 --> 00:25:58.160
I mean it's see it it's yeah yeah yeah, generative first time I clapped ever in this uh podcast.
00:25:58.240 --> 00:25:59.920
Yeah, no, no, no, definitely.
00:26:00.000 --> 00:26:00.720
Yeah, that resonates.
00:26:04.319 --> 00:26:05.519
25 minutes.
00:26:05.839 --> 00:26:14.640
No, that definitely, and specifically because you also make the so you looked back to words, you looked back to prose, and to me, that was the inspiration, right?
00:26:14.799 --> 00:26:16.160
I made this visual, yeah.
00:26:16.400 --> 00:26:20.559
And to me, it's like it was kind of you.
00:26:20.640 --> 00:26:42.400
You I think you you commented on some post somewhere, and you you said, I don't feel like a person that makes things with AI rather than I give I give an assignment to an agent that does things and I appraise it, the results.
00:26:42.480 --> 00:26:48.000
Yeah, and I feel like I'm more like a patron, I think that's the word that you used, a patron of the system.
00:26:48.160 --> 00:27:17.440
Yeah, and um to me that really connected with okay, so if that's the case and we're working with AI, and this is the relationship we're bound to have, yeah, then world building is not per se describing the world, but it's describing um let's say the set the the the the context of of where the world might grow, right?
00:27:19.119 --> 00:27:20.240
Yeah, exactly, yeah, definitely.
00:27:20.960 --> 00:27:21.839
No, yeah, yeah, totally.
00:27:22.000 --> 00:27:30.960
No, I mean I made that comment, um, or actually I copied that comment in a uh review by Ethan Mollik from Fable of Fable 5.
00:27:31.119 --> 00:27:36.640
Yeah, and he said, uh, I I was used to be a conductor with my AI, yeah, yeah, yeah.
00:27:36.799 --> 00:27:42.559
But with this one, I give it a task, it makes a thousand decisions.
00:27:42.720 --> 00:27:47.279
I have I'm not aware of you don't have a say, and I pay for it, yeah, yeah, yeah.
00:27:47.440 --> 00:27:47.599
Right?
00:27:47.759 --> 00:27:53.920
I give it the money, yeah, and you don't have a say, and it comes back with something as a I like, yeah, or not.
00:27:54.000 --> 00:27:55.440
Yeah, I'm a Patreon, right?
00:27:55.519 --> 00:27:56.160
Yeah, yeah, yeah.
00:27:56.480 --> 00:27:59.279
This is a really nice shift, I think.
00:27:59.920 --> 00:28:02.000
Well, it's definitely a different relationship.
00:28:02.079 --> 00:28:12.000
Also, yeah, I mean, if you get the design uh spectrum on that, we see it, we see a different shift on our role within the design system.
00:28:12.319 --> 00:28:14.799
And that I mean, and there's another metaphor.
00:28:14.880 --> 00:28:18.000
I mean, we're just no throwing metaphors left and right, right?
00:28:18.400 --> 00:28:19.279
No, it's just a model.
00:28:19.440 --> 00:28:22.160
No, yeah, it's that's totally true.
00:28:22.400 --> 00:28:25.359
No, and that's uh that's adventure games, yeah.
00:28:25.599 --> 00:28:28.160
I mean, if you know Zorg and that kind of stuff, right?
00:28:28.240 --> 00:28:31.680
And uh Dungeons and Dragons.
00:28:32.000 --> 00:28:34.720
Dragons and Dungeons, Dungeons and Dragons, Dungeons and Dragons, yeah.
00:28:34.880 --> 00:28:34.960
Right.
00:28:35.039 --> 00:28:37.920
I mean, and because that I think that's what we're doing.
00:28:38.079 --> 00:28:40.640
We're giving them a Zork world, right?
00:28:40.799 --> 00:28:46.079
We describe it to the agent and we disclose gradually to the agent, you are now in this world.
00:28:46.160 --> 00:28:52.640
Yeah, and then the agent goes goes left, opens the chest, right, looks in the uh uh in the mailbox.
00:28:52.720 --> 00:28:55.920
Uh if you know Zorik, you you know why I made that remark.
00:28:57.920 --> 00:29:00.960
But but you but you give it a world to act in.
00:29:01.359 --> 00:29:03.200
Exactly, and that's it, right?
00:29:03.279 --> 00:29:09.599
You you talk about the gravity, you say there like a total surprise storm is a is possible.
00:29:10.480 --> 00:29:15.359
And if something goes wrong, right, or uh it dies in the game or whatever, right?
00:29:15.440 --> 00:29:18.720
I mean, there are two two two types of actions.
00:29:18.880 --> 00:29:29.440
You can correct it, stupid AI, do this, yeah, or you correct the world, you change the physics of the world, or you add something, or you add a constraint, or you relax a constraint.
00:29:29.759 --> 00:29:31.599
There's a magic spell, you can't die.
00:29:32.319 --> 00:29:34.319
You're actually asleep, yeah.
00:29:34.799 --> 00:29:38.079
But so it's all I mean, so in that sense, we are the dungeon masters, right?
00:29:38.160 --> 00:29:48.160
And is it and the reason why I make this we should bookmark this particular word because you are you're a dungeon master, yeah.
00:29:48.319 --> 00:29:59.680
And yeah, but but often, certainly when ChatGPT happened, people were framing it that it is the dungeon master, yeah, right, and that we are playing to uh its game to its world.
00:29:59.759 --> 00:30:04.559
I believe I mean I don't believe that it's literally the other way around, but literally it's the other way around.
00:30:05.119 --> 00:30:11.039
I guess the depending on the amount of prompting or let's say dungeon mastering you do yourself.
00:30:11.200 --> 00:30:11.519
Yeah, okay.
00:30:12.400 --> 00:30:21.920
But if you go if you go into the wide world of matrix, right, where it
where we just need guns, more guns, it's of all or all the guns, I'm not sure.
00:30:22.160 --> 00:30:34.319
It's like what we uh if we decide there's nothing there, then well, that's also a world where we might relinquish relinquish some of the powers that we expect.
00:30:35.519 --> 00:30:42.079
Nevertheless, it is something we can also fill in and make yeah, more concrete.
00:30:42.319 --> 00:30:51.839
And the the reason that I say, aren't we also more designing the context in which we design the things rather than the thing itself?
00:30:51.920 --> 00:30:58.079
Yeah, like you also say, being a patron or quoting the other person that talked about that.
00:30:58.720 --> 00:31:13.119
Um to me, it's like, okay, so if that this is the case, and at the moment language or in prose specifically, so ambivalent language is our tool to make to create this context.
00:31:13.440 --> 00:31:24.960
Could we also make tools that also appeal to our other senses, like the the the other intuitive senses that humans have, like a visual interface that we can see?
00:31:25.119 --> 00:31:31.839
Oh, I forgot this little, I forgot the aspect in my dungeon mastering that people could not die, right?
00:31:31.920 --> 00:31:41.440
It's it's like, oh, I should change that first, because that's something I'm kind of missing if we talk about LLMs and talking about how we create our worlds.
00:31:41.599 --> 00:31:46.880
Yeah, it doesn't seem we we don't see everything, or I don't know.
00:31:47.200 --> 00:31:48.079
Oh, that's a nice one.
00:31:48.240 --> 00:31:51.680
No, I mean, but that's why I said what even a doodle on a whiteboard can be an input, right?
00:31:51.839 --> 00:31:52.319
Yeah, yeah, yeah.
00:31:52.480 --> 00:31:57.200
I think we are we by and large are still limited to bringing text to it.
00:31:57.440 --> 00:31:58.160
Yes, right?
00:31:58.319 --> 00:32:00.559
For sure, we could bring images, right?
00:32:00.880 --> 00:32:02.079
Could we bring sound?
00:32:02.240 --> 00:32:02.960
I have no idea.
00:32:03.039 --> 00:32:04.720
It's the uh it's an interesting one.
00:32:04.880 --> 00:32:06.400
It would be yeah, that would be awesome.
00:32:07.200 --> 00:32:21.839
And uh specifically, also um I'd love I'd love to also see things I forget and things that might be implied uh if you don't if you don't say something.
00:32:22.079 --> 00:32:27.759
And this is kind of has to do with the biases play that we also work with LLMs, right?
00:32:27.839 --> 00:32:34.559
It's based on data, blah blah blah, and it has definitely has biases because it's based on other things.
00:32:34.799 --> 00:32:36.319
And I'm very much interested.
00:32:36.480 --> 00:32:42.480
So if we're creating a world and and we forget a basic principle like gravity, yeah.
00:32:42.799 --> 00:32:46.240
Yeah, but yeah, could could we could we say something about it?
00:32:46.319 --> 00:32:51.519
Could we see something like that that we might want to say uh something important about that?
00:32:51.759 -->
00:33:04.559
Yeah, that's an I I mean so one thing I noticed recently that's the re reason why I'm now talking slowly, is how powerful these systems also are in introspection on their own actions.
00:33:04.720 --> 00:33:05.599
Yeah, yeah.
00:33:05.839 --> 00:33:09.680
So I mean I try to get a certain image and somehow it just goes wrong, right?
00:33:09.759 --> 00:33:10.000
Yeah, yeah.
00:33:10.160 --> 00:33:12.559
And then I ask, why is it going wrong?
00:33:12.799 --> 00:33:17.920
Yeah, and it comes back more often than not with a proper analysis.
00:33:18.079 --> 00:33:34.079
Well, you said XYZ, which means it uh the model is forced to focus too too much on this aspect if you would have used the words x accent, uh y accent, and I'm sitting there really, and I try it, and guess what?
00:33:34.720 --> 00:33:36.319
So succeed, success.
00:33:36.799 --> 00:33:39.599
It knows why it went wrong.
00:33:39.759 --> 00:33:40.079
Ah.
00:33:40.720 --> 00:33:41.839
After the fact, you have to ask.
00:33:47.599 --> 00:33:55.440
Uh the visual uh generative AI, and I can't talk, I I don't seem to be able to have it talk back to me.
00:33:55.680 --> 00:34:04.240
Yeah, but I mean, but but this is specifically for uh GPT, yeah, because it has a text and image uh embedded in one user, right?
00:34:04.480 --> 00:34:09.760
And and and the text model can reason about what the image model is doing.
00:34:09.920 --> 00:34:17.039
Yeah, so this what I just said is a discussion I have with the text model, which gives me then the prompt that I then hand to the image model, right?
00:34:17.119 --> 00:34:20.320
So but this this introspection is fantastic.
00:34:20.400 --> 00:34:22.000
You can it is, yeah, yeah.
00:34:22.320 --> 00:34:23.280
Why did this go wrong?
00:34:23.440 --> 00:34:24.719
Well, I think it's this.
00:34:24.960 --> 00:34:27.599
So that might be like a key feature, you know.
00:34:27.920 --> 00:34:43.119
People are talking about agents being like very autonomous and very helpful in uh speed and capacity, but it seems like one of the superpowers might be being able to reason smear with them.
00:34:43.280 --> 00:34:45.679
Yeah, I mean it's the I mean to get to know each other.
00:34:46.000 --> 00:34:50.719
At least in my bubble, I mean the the the word of the week was loops.
00:34:51.199 --> 00:34:52.159
Yeah, I read that.
00:34:52.400 --> 00:34:55.840
Yeah, so you don't write prompts anymore, you write loops, right?
00:34:55.920 --> 00:34:59.280
So like the what is this programming 101?
00:34:59.599 --> 00:35:03.039
I didn't understand it because this this word has been reserved in my mind.
00:35:03.199 --> 00:35:05.039
Yeah, but I mean what's the programmer, yeah.
00:35:05.199 --> 00:35:22.159
So yeah, sure, but but yeah, but in this context is instead of actually telling the model what to do, yeah, you come up with a wider arching, more global uh goal, and then the model starts, and there is also an introspection phase.
00:35:22.400 --> 00:35:27.360
How well did how well did I do uh accomplishing this this goal, right?
00:35:27.519 --> 00:35:28.880
And how can I adjust?
00:35:29.039 --> 00:35:30.320
And that's the looping thing.
00:35:30.960 --> 00:35:32.960
Sounds a little bit like a design thinking loop.
00:35:33.199 --> 00:35:35.360
Yeah, because it also has introspection.
00:35:36.000 --> 00:35:36.880
That's a nice one.
00:35:37.039 --> 00:35:38.559
Yeah, didn't think about that one.
00:35:38.639 --> 00:35:39.440
Oh, that's cool.
00:35:39.599 --> 00:35:40.159
Yeah, it is.
00:35:40.320 --> 00:35:46.960
Oh, I also gonna oh my I mean talking about dropping dropping in in a sentence.
00:35:47.199 --> 00:35:52.320
Yeah, because I I wanted to say I also gonna then try to take my old love, the double diamond.
00:35:52.480 --> 00:35:55.280
I have no idea how or what, but I'm gonna do something with that.
00:35:55.440 --> 00:35:58.880
Well, it that it's definitely the pattern is definitely there.
00:35:59.119 --> 00:36:03.920
The double diamond was my language from the previous error when I saw that thing for the first time.
00:36:04.079 --> 00:36:05.599
I mean, literally.
00:36:07.760 --> 00:36:09.440
Can I quickly yes, please?
00:36:09.599 --> 00:36:14.000
So it's the uh I was uh um still consulting, I think.
00:36:14.400 --> 00:36:19.199
I was consulting for a uh uh a very large uh Dutch originated bank.
00:36:19.360 --> 00:36:19.519
Right.
00:36:19.679 --> 00:36:20.960
Who loves the colour orange?
00:36:21.119 --> 00:36:24.960
Yes, uh there are two banks who love the colour orange, but one has a really nice colour orange.
00:36:25.119 --> 00:36:29.760
Yeah, that's that bank, not with the ugly blue, yeah, not them, right?
00:36:29.840 --> 00:36:35.519
And uh was working with my uh co-conspirator who is still in my life, really enjoyed that one.
00:36:35.679 --> 00:36:40.239
And he said, and and we basically what we were uh we were implementing agile, scrum.
00:36:40.639 --> 00:36:41.280
Right, right.
00:36:41.760 --> 00:36:42.960
That was also a thing back then.
00:36:43.199 --> 00:36:43.760
Yeah, yeah, yeah.
00:36:43.840 --> 00:36:45.360
I've I've I remember.
00:36:45.599 --> 00:36:47.840
Yeah, and he said, Well, I think we're done.
00:36:48.000 --> 00:36:51.519
I think this whole scrum thing is done, done as in done done.
00:36:51.599 --> 00:36:54.239
Not done dead, but it's here.
00:36:54.400 --> 00:36:54.880
Yeah, yeah, yeah.
00:36:54.960 --> 00:36:56.079
And we are done, right?
00:36:56.320 --> 00:36:56.559
Right.
00:36:56.719 --> 00:36:58.719
And they had this- But are you done with it?
00:36:58.880 --> 00:37:00.079
Like not interested anymore.
00:37:00.320 --> 00:37:03.360
No, no, just just it's it's it's the organization is doing it.
00:37:03.519 --> 00:37:05.039
We we accomplished our goal.
00:37:05.119 --> 00:37:05.519
Ah, okay.
00:37:05.760 --> 00:37:12.320
And they had this this thing in the back of my mind that that just kept shouting to me, no Ron, no, no, no.
00:37:12.480 -->
00:37:13.679
But I didn't know what.
00:37:13.840 --> 00:37:16.159
I didn't have the word yet, right?
00:37:16.960 --> 00:37:21.199
One week later, within a week, I saw the double diamond for the first time.
00:37:21.360 --> 00:37:21.679
Right.
00:37:22.320 --> 00:37:24.559
Discover, define, develop, deliver.
00:37:25.039 --> 00:37:25.760
Two diamonds.
00:37:26.159 --> 00:37:26.400
Yeah.
00:37:27.679 --> 00:37:29.679
Scrum is the second diamond.
00:37:30.800 --> 00:37:31.760
Develop deliverer.
00:37:32.079 --> 00:37:32.239
Yeah.
00:37:32.639 --> 00:37:35.679
There is a st there is a whole diamond before that one.
00:37:35.760 --> 00:37:35.920
Yeah.
00:37:36.320 --> 00:37:36.559
Right.
00:37:36.639 --> 00:37:39.119
And this is basically how do you get to your backlog.
00:37:39.519 --> 00:37:50.719
Because if you read the scrum guide, it's assumed there is this magical person called the product owner, yeah, who just by sheer brilliance knows how to fill the backlog.
00:37:51.199 --> 00:37:51.760
Yes.
00:37:52.320 --> 00:37:54.320
The magic ability of typing.
00:37:54.639 --> 00:38:01.199
Yeah, and I had it on my on on personal experience that's most most product owners cannot.
00:38:01.519 --> 00:38:02.159
Indeed.
00:38:02.639 --> 00:38:05.840
Apart from, yeah, let's do the next version of Java, right?
00:38:05.920 --> 00:38:07.199
And these kind of things.
00:38:07.519 --> 00:38:09.760
It's the uh so but this was the epiphany.
00:38:09.920 --> 00:38:13.360
There is this whole world before you actually get to the article.
00:38:13.519 --> 00:38:15.440
And now that sounds bloody obvious, right?
00:38:15.599 --> 00:38:18.480
But back then I had to discover this model.
00:38:19.039 --> 00:38:27.760
I know, just a model, but still, a model to articulate the world I was feeling, but couldn't describe to anyone yet.
00:38:28.000 --> 00:38:28.880
But that's it.
00:38:29.119 --> 00:38:30.639
So you just nailed it.
00:38:30.880 --> 00:38:35.039
The model that you use is a way to convey the thing that you meant.
00:38:36.079 --> 00:38:37.360
It's basically okay.
00:38:37.519 --> 00:38:42.000
If you want really want to know the ins and outs, here's a nice book uh for you to read.
00:38:42.079 --> 00:38:50.320
But this is a very uh understandable way of explaining us uh a certain abstract thing in this case, yeah, right, or a process or whatever.
00:38:50.400 --> 00:38:52.480
Yeah, and I really val value this.
00:38:52.559 --> 00:38:58.639
So I think we should also incorporate this in how we, for example, be a patron, right?
00:38:58.800 --> 00:39:12.559
Yeah, we can we should use certain models to also have like uh to to have a I want to don't but I want to stop using the word prompt, but have a let's say a dialogue with with artificial intelligence.
00:39:13.039 --> 00:39:22.960
And I mean and and maybe ex maybe exoskeleton is not a thing, but for me it opened a whole bunch of windows by just having that yeah that that word, right?
00:39:23.039 --> 00:39:30.880
Yeah, pretty much every article I read over the past two weeks, I could see ah, what you're talking about is exoskeleton it, right?
00:39:31.039 --> 00:39:36.079
Yeah, but also on the double diamond, I have a something is in there, I just know it.
00:39:36.239 --> 00:39:47.119
Yeah, the funny thing is, I you see it in a kind of or it seems like where when people are working in agile to me, it sounds like production, building stuff, yeah.
00:39:47.280 --> 00:39:51.039
And there's there's there is some discovery going on, right?
00:39:51.119 --> 00:39:56.000
The flexibility to be realize you've made the wrong turn or whatever kind of thing.
00:39:56.239 --> 00:40:02.239
But basically it's like going to the end goal, yeah, it being like a quite clear end goal as well.
00:40:02.320 --> 00:40:06.719
Yeah, and to me, the double diamond is a discovery tool, something totally different.
00:40:06.880 --> 00:40:24.320
So making the the production part, the diamond that you inhibited from the get-go, to me is a prototyping part, it is a ways to make things uh uh uh to uh give to other people so they know what I meant, right?
00:40:24.480 --> 00:40:26.159
So I think let's say so.
00:40:26.239 --> 00:40:29.440
The first diamond is discovery, it's like what do we want to do?
00:40:29.599 --> 00:40:34.800
Okay, let from all the things that we want to do, let's choose this, because we think this is most important.
00:40:34.880 --> 00:40:44.800
That's the second part of the diamond, and from that point is making a prototype so we can show other people and see whether or not this will work in uh in our world, right?
00:40:45.039 --> 00:40:55.280
And that's and the end result is like reflection upon that, see testing, but also reflection afterwards, like okay, so the process that we just did, can we improve it a little bit?
00:40:55.440 --> 00:41:04.800
Well, that's kind of like what you do in stand-ups or stuff like that in the agile world, but mostly also can we improve our understanding of where we actually want to go?
00:41:05.280 --> 00:41:09.760
Did we learn things that we either were right on the money or way off base?
00:41:10.000 --> 00:41:11.760
Yeah, but that's the first diamond, right?
00:41:11.920 --> 00:41:14.719
Well, it looks back, and it always looks back, yeah.
00:41:14.800 --> 00:41:26.719
Yeah, but you need to have the exercise whether or not you the thing that you just did, whether or not you were satisfied, and that you were curious, and your end goal was also supported, right?
00:41:26.960 --> 00:41:31.440
It's uh by testing or having people look at it or whatever kind of thing.
00:41:31.760 --> 00:41:41.840
And of course, I made a model with two double diamonds, but it's the yeah, and to me, it's uh the like that even the diamonds can be iterated upon on it on itself, yeah.
00:41:41.920 --> 00:41:42.400
Totally.
00:41:42.559 --> 00:41:45.199
Uh, because that's something most people forget.
00:41:45.360 --> 00:41:50.480
I even
made like an illustration, I think I used it in uh one of our public speaking gigs.
00:41:51.760 --> 00:41:58.880
Uh I come to realize it, I think it was illustrated the wrong way, but it doesn't really matter.
00:41:59.039 --> 00:42:12.559
But mostly most of the so how I use the double diamond most of the time, and I I do have this misunderstanding with other people, is it's not built to make things not to make end results.
00:42:12.800 --> 00:42:15.280
It's it's a wayfinding tool.
00:42:15.760 --> 00:42:17.840
Well, there is also develop and deliver.
00:42:17.920 --> 00:42:19.519
So, what are you developing and delivering?
00:42:19.840 --> 00:42:23.360
Well, I use different words in in those for those diamonds.
00:42:23.599 --> 00:42:24.239
Yeah, yeah.
00:42:24.320 --> 00:42:26.559
It's not deliver, definitely not deliver.
00:42:26.719 --> 00:42:29.280
Well, it's it's it's build and test.
00:42:30.639 --> 00:42:37.519
It is it it is absolutely develop and deliver in the original double diamond, in some books, yeah.
00:42:37.679 --> 00:42:40.960
No, no, no, no, no, in the original uh design console documentation.
00:42:41.440 --> 00:42:59.039
I understand what you're saying, but I'm saying I I I took that model and applied it to let's say a world where uh either innovation or discovery has a very strong um uh well was like the deliverable itself, let's say.
00:42:59.840 --> 00:43:12.559
So in that because in that case we could also go a little bit nuts, we could let's say uh the first diamond we could discover like weird things and really go into tangents.
00:43:13.119 --> 00:43:22.320
Whether if you would be in a building mode, you would do away with tangents because those might be distractions most of the time.
00:43:22.719 --> 00:43:27.039
I mean, develop is diverging, that's why it's drawn like like yeah, yeah, yeah, definitely.
00:43:27.199 --> 00:43:34.800
No, I think you're uh uh I think I think there might be value going back to the to to to the original definition of the diamond page.
00:43:34.960 --> 00:43:49.119
I I think I think we both have uh a decent understanding of the model, and uh but I definitely applied it differently than than I nowadays see with uh with people that basically just want to build a thing.
00:43:49.280 --> 00:43:50.000
Yeah yeah.
00:43:50.239 --> 00:43:53.039
And uh and nevertheless the model is very strong and very rich.
00:43:53.280 --> 00:43:59.599
No, I mean and and and what and and what is interesting now is that apparently we're now discussing that this model as a potential use.
00:44:00.159 --> 00:44:00.719
For agents.
00:44:00.800 --> 00:44:01.039
Yeah.
00:44:01.199 --> 00:44:04.559
Or in an agenc context or agentic development context.
00:44:04.880 --> 00:44:07.199
Wouldn't surprise me that's already in there.
00:44:08.800 --> 00:44:11.199
What where is there?
00:44:11.519 --> 00:44:14.880
With within uh the function of an LLM.
00:44:15.119 --> 00:44:20.079
I mean it's so the ways for it to do a sanity check for itself, for example.
00:44:20.320 --> 00:44:23.039
Yeah, certainly the diverging and converging bit is there.
00:44:23.679 --> 00:44:26.639
So it's the uh that is that is totally true, and you get to the define.
00:44:27.199 --> 00:44:36.239
I mean, you could probably say oh you could probably say that the reasoning phase in the in the agent is the first diamond, yeah, and the answering phase is the second diamond.
00:44:37.360 --> 00:44:37.760
Yeah.
00:44:39.840 --> 00:44:40.960
Okay, we're done.
00:44:41.599 --> 00:44:43.039
Yeah, it's gonna be a short episode.
00:44:43.119 --> 00:44:44.239
It's the uh no, we're not done.
00:44:44.400 --> 00:44:46.079
I was gonna I'm I'm gonna try that one.
00:44:46.320 --> 00:44:56.719
No, I I I I I allowed myself to have a uh three-month mini sabbatical to do this kind of thinking and doing pondering, yeah.
00:44:56.880 --> 00:45:06.480
Yeah, pondering and just try things uh with a somewhat unbounded uh amount of time, yes, instead of the two hours I have on Saturday morning.
00:45:06.960 --> 00:45:07.360
Okay, okay.
00:45:08.559 --> 00:45:11.920
It's an odd uh odd time to do stuff like that.
00:45:12.480 --> 00:45:14.719
To me, it's like oh I'm exhausted.
00:45:15.199 --> 00:45:17.039
No, no, I'm an I'm this week.
00:45:17.199 --> 00:45:18.480
I'm an I'm an early riser.
00:45:18.960 --> 00:45:20.719
And my girlfriend is not, yeah, yeah, yeah.
00:45:20.960 --> 00:45:24.559
So the delta between me getting up and you're getting to you.
00:45:24.800 --> 00:45:28.079
That's the time I can do my uh my uh introspective stuff.
00:45:28.159 --> 00:45:28.639
Oh, that's nice.
00:45:28.800 --> 00:45:29.920
I need to do it by myself.
00:45:30.159 --> 00:45:33.119
Yeah, yeah, I I get that the same only late at night.
00:45:33.280 --> 00:45:38.320
Yeah, that's the way you can't sleep, it's like okay, let's think about uh yeah, that's the horrible thing.
00:45:38.400 --> 00:45:43.679
I'm I'm I'm very much a morning and a night person, and I'm quite horrible from nine to five, actually.
00:45:43.760 --> 00:45:45.119
So oh yeah, me too.
00:45:47.440 --> 00:45:53.679
Well, I I guess I'm busy, I just like to do busy things, busy work from nine to five.
00:45:53.920 --> 00:46:04.559
Basic, so basic treading the water, making sure things work, making things, just doing things that I already thought of and do the busy work, yeah, drawing stuff and stuff like that.
00:46:04.719 --> 00:46:20.480
Uh, but most of the actual work and thinking either happens over here, hence, uh or when I do the editing of this podcast, yeah, because I pick up on things like okay, this is gold, and we need to make an yeah, make an illustration.
00:46:20.559 --> 00:46:29.760
I really like the uh making illustrations with with those thoughts, just to I don't know, have others also to take part in what we make, yeah.
00:46:29.920 --> 00:46:30.079
Right.
00:46:30.320 --> 00:46:35.199
Yeah, I want to do an illustration about Zorik and an
agent uh running around, but it's the I ran out of time.
00:46:36.639 --> 00:46:43.360
Yeah, yeah, I really uh I'm just want to check if we ignored all our topics.
00:46:43.679 --> 00:46:47.840
Yeah, we must make sure we must make sure we ignored all the topics, yeah, exactly.
00:46:48.079 --> 00:46:49.760
No, I actually mentioned it.
00:46:50.320 --> 00:46:54.159
It's it's like the the sandbox that I talked about.
00:46:54.320 --> 00:47:05.840
Um, so is PROS the only language that we can build, but also check up on the the things that we do with AI, or specifically with LLMs.
00:47:05.920 --> 00:47:16.480
Yeah, I want to include I I read an article from six months ago of this person that left a uh important um LLM building company.
00:47:16.639 --> 00:47:24.000
I'm not sure if it was OpenAI or uh Entropic, but basically he said, Well, this is limited.
00:47:24.159 --> 00:47:26.800
So LLMs as a concept are limited.
00:47:26.880 --> 00:47:31.039
Yeah, they will be like you just said, at a certain moment, they will be done.
00:47:31.360 --> 00:47:35.920
And his vision was okay, so what's the next iteration?
00:47:36.000 --> 00:47:37.280
And he started working on that.
00:47:37.360 --> 00:47:41.760
He got some pocket money, a billion dollars, and he started working on that.
00:47:41.840 --> 00:47:52.639
And his premise that he developed beforehand, it seemed, was uh movement, uh trying to predict movement.
00:47:53.119 --> 00:48:11.360
So basically, if we got the ball over here and somehow build a world so you can predict it if you give it a little tap that the ball will appear over there, yeah, or or have a model that understands that if you give it a tap, the thing moves.
00:48:11.519 --> 00:48:11.840
Yes.
00:48:11.920 --> 00:48:12.320
Yeah.
00:48:12.559 --> 00:48:16.400
And that's kind of like I'm very curious about.
00:48:16.480 --> 00:48:28.079
I'm not sure if this is this is a valid answer or very valid uh addition to the tool set, but I'm definitely interested in different AIs, in different applications of the code.
00:48:28.239 --> 00:48:28.880
Oh, yeah, sure.
00:48:28.960 --> 00:48:31.920
I mean, I think what you what you're describing are our world models, right?
00:48:32.000 --> 00:48:32.400
Yeah, yeah.
00:48:33.039 --> 00:48:36.079
And maybe the researcher you mentioned is uh Jan Le Koon.
00:48:36.559 --> 00:48:37.440
Yeah, exactly.
00:48:37.760 --> 00:48:38.639
He worked at Meta.
00:48:38.880 --> 00:48:39.519
Yeah.
00:48:39.760 --> 00:48:52.079
And and uh the interesting because they call it world building, and I think it might not be because world building to me is like where where does the house stand, where's the road?
00:48:52.159 --> 00:48:55.679
You know, it's basically city building, uh the uh the the game.
00:48:56.000 --> 00:48:57.599
Yeah, but also the physics, right?
00:48:57.679 --> 00:49:04.719
I mean that action is reaction, well, and uh and spatial relationships and time relationships, but things change over time.
00:49:05.039 --> 00:49:06.960
Okay, little side tangent.
00:49:07.039 --> 00:49:19.360
I'm not sure if you've seen the video, but I think I sent it to you, but this is this very ancient video of uh me, uh very uh young and uh quite quite a lot uh more uh poundage.
00:49:20.159 --> 00:49:26.400
Um, and I'm talking more about one of my fascinations, which is sound, sound on itself.
00:49:26.559 --> 00:49:33.760
And I talked about Ben Burt with double T at the end, and Ben Burt is the sound engineer of Star Wars, yeah.
00:49:34.000 --> 00:49:42.159
Awesome, and the thing so this blew my mind lit, not literally, because that's kind of like uh gruesome.
00:49:44.000 --> 00:49:49.519
So he needed to design the sound that the the pods made in the pod race, yeah.
00:49:49.760 --> 00:50:06.480
And he said, Okay, we've got all kinds of yeah, exactly, it's got all kinds of source material, and on Earth, these engines would sound like and he said, No, no, no, this is these are alien planets, the atmosphere works in a different way.
00:50:06.719 --> 00:50:14.000
So he designed the atmosphere so he could make so he could design the the sounds.
00:50:14.159 --> 00:50:15.760
So the sounds were were indeed.
00:50:17.679 --> 00:50:21.199
And it's so to me this was okay.
00:50:21.760 --> 00:50:26.320
That is what we need to think about if we want a consistent world building.
00:50:26.559 --> 00:50:29.920
Yeah, this is literally the we say literally a lot.
00:50:30.559 --> 00:50:32.239
Did you know that we literally said literally a lot?
00:50:32.719 --> 00:50:40.960
The next time when we say literally, we need to have the quote on the screen so we can say it is actually literally, yeah.
00:50:41.679 --> 00:50:44.159
It's the but this is world building, right?
00:50:44.239 --> 00:50:59.360
In order to make the sound, you actually create a world where that sound could exist, yeah, or the other way around, create a world and then are curious on what kind of sounds live there when you set the set the rules.
00:50:59.760 --> 00:51:04.239
Or you or you start with the sound and start thinking about the world and get to the I mean this is intuitive.
00:51:04.480 --> 00:51:04.880
Yeah, yeah, sure.
00:51:04.960 --> 00:51:05.840
I mean totally.
00:51:05.920 --> 00:51:11.039
I mean, this is to me, the surprise on what kind of sounds you might hear over there.
00:51:11.199 --> 00:51:13.679
Yeah, to me, it does so.
00:51:13.760 --> 00:51:15.280
Let's say a very practical way.
00:51:15.360 --> 00:51:24.960
If you uh if you are going uh swimming uh soon, just dive underwater and hear how audio is so damn different.
00:51:25.119 --> 00:51:29.280
Yeah, I know the clicks that you hear of little pieces of metal hitting the tiles.
00:51:29.679 --> 00:51:30.480
Amazing.
00:51:31.280 --> 00:51:34.960
I mean, it's weird, it's both muted and very loud.
00:51:35.119 --> 00:51:36.159
Yeah, yeah, yeah.
00:51:36.320 --> 00:51:36.639
Yeah.
00:51:37.119 --> 00:51:46.159
And that's and just
thinking about, I mean, it's thinking about physics on the one hand, but it's also having your imagination uh go free, right?
00:51:46.239 --> 00:51:50.320
And and then experiencing on the water so great that your movement makes sound.
00:51:50.480 --> 00:51:50.800
Yeah.
00:51:51.039 --> 00:51:53.199
That you can actually hear yourself moving, right?
00:51:53.599 --> 00:51:54.800
It's fantastic.
00:51:55.440 --> 00:51:56.079
Yeah.
00:51:56.320 --> 00:52:16.320
So to it, and so if we go back, look back to the world building and also working on the exoskeleton, it's building tools to actually not only describe what we actually want to see in the end, what we nowadays do, but specifically create a world where things can appear.
00:52:16.400 --> 00:52:17.280
Yeah, right?
00:52:17.679 --> 00:52:23.360
Create the logarithms, if you will, where certain things can appear across our spectrum.
00:52:23.519 --> 00:52:24.000
Yeah, exactly.
00:52:24.159 --> 00:52:24.639
Yeah, yeah, yeah.
00:52:25.440 --> 00:52:32.639
And and be surprised, just like you, with the with the sound, that a certain instance of the possibility turns up that you never saw coming.
00:52:32.800 --> 00:52:33.199
Exactly.
00:52:33.440 --> 00:52:34.800
Is that what could have been possible?
00:52:34.880 --> 00:52:36.000
And the answer is yes.
00:52:36.159 --> 00:52:36.480
Yeah.
00:52:36.719 --> 00:52:42.000
Just like in adventure games, where people would do really funky stuff, yeah, right?
00:52:42.159 --> 00:52:43.679
Stack this on top of that one.
00:52:43.840 --> 00:52:46.000
Uh, speed runs in Quake.
00:52:46.480 --> 00:52:47.039
Oh, okay.
00:52:47.119 --> 00:52:47.760
Yeah, yeah, yeah.
00:52:48.159 --> 00:52:51.119
The things people come up with to actually run faster.
00:52:51.280 --> 00:52:56.000
It's the I mean, it's trust me, the designers of Quake never thought about that one.
00:52:56.159 --> 00:52:56.880
No, no, no.
00:52:57.039 --> 00:53:01.760
And that's is so abusing the system as well, which is one of my pet uh projects as well.
00:53:01.920 --> 00:53:05.679
It's like, okay, show me the rules and I'll show you how to hack them.
00:53:06.000 --> 00:53:07.440
Yeah, or yeah, yeah.
00:53:07.840 --> 00:53:16.000
Or abuse them or use them, or apply them in in in unexpected ways, unexpected by the designers of the rules, right?
00:53:16.239 --> 00:53:21.199
And for them it looks like hacking, but for you it's just applying the rules the way you see fit.
00:53:21.360 --> 00:53:21.760
Yeah, exactly.
00:53:22.239 --> 00:53:22.719
Yeah, it's fantastic.
00:53:22.960 --> 00:53:24.320
Yeah, with different moras, I guess.
00:53:24.559 --> 00:53:26.159
Yeah, yeah, different intentions, etc.
00:53:26.400 --> 00:53:30.320
But it's the uh yeah, but all systems are hackable, certainly if they are digital.
00:53:30.960 --> 00:53:35.440
Yeah, and in the in that vein, I mean you you said it correctly.
00:53:35.760 --> 00:53:42.800
But whether or not it's hacked is based on your own perspective on the things, right?
00:53:43.039 --> 00:53:46.000
Yeah, you say hack, I say use, and the other way around.
00:53:46.079 --> 00:53:46.639
Yeah, yeah.
00:53:47.039 --> 00:53:48.159
That's quite interesting.
00:53:48.320 --> 00:53:51.840
So we basically now got back to the first topic.
00:53:52.159 --> 00:54:05.199
Whether or not was is the American government hacking the system of Entropic, or is Entropic hacking the uh the American government with their uh with their marketing?
00:54:05.519 --> 00:54:06.639
It is a big game.
00:54:06.800 --> 00:54:13.119
I'm not sure it might be boring, but it uh but it is nevertheless, there is some intent over there, it seems to be.
00:54:13.360 --> 00:54:19.679
I think both parties are under overestimating their level of control, yeah, and also their importance.
00:54:20.320 --> 00:54:22.320
I don't think it really matters that much.
00:54:22.559 --> 00:54:23.360
Yeah, that's what I mean.
00:54:23.440 --> 00:54:24.800
I mean it's the it was also fine.
00:54:24.880 --> 00:54:26.159
I mean, we have a couple of minutes, right?
00:54:26.320 --> 00:54:31.760
But if if you if so when I read all the reviews about Fable 5, right?
00:54:31.840 --> 00:54:34.320
So the uh uh the kiddie version of Mythos.
00:54:34.960 --> 00:54:38.079
To me, it read like a campaign.
00:54:38.400 --> 00:54:39.679
Oh, for sure there's a campaign.
00:54:39.840 --> 00:54:41.920
No, no, no, like a massive campaign.
00:54:42.719 --> 00:54:47.440
I don't know if whether it there is a publication that didn't write a bio it.
00:54:47.519 --> 00:54:48.000
Oh, yeah, sure.
00:54:48.239 --> 00:54:48.559
Day off.
00:54:48.800 --> 00:54:49.519
Oh yeah, totally.
00:54:49.599 --> 00:54:55.360
It's the uh but it I mean, if the new iPhone comes out, yeah, and you you read a gazillion iPhone reviews.
00:54:55.920 --> 00:54:57.039
To me, this is a red flag.
00:54:57.119 --> 00:55:03.039
I mean, this this wasn't about the new product, it was about what it the the implications.
00:55:03.280 --> 00:55:04.159
Yeah, I don't know.
00:55:04.239 --> 00:55:08.719
I mean it's the uh uh the the ones I read read like product reviews.
00:55:08.880 --> 00:55:09.360
Oh really?
00:55:09.440 --> 00:55:15.920
And yeah, and you actually see that the uh that the product oversaturated the use cases.
00:55:16.239 --> 00:55:16.559
Yeah.
00:55:16.800 --> 00:55:23.760
Basically, it it does something really cool for very specific uh use cases, yeah, and then it's really awesome.
00:55:24.000 --> 00:55:24.320
Yeah.
00:55:24.719 --> 00:55:29.119
And for a very large range of existing use cases, right?
00:55:29.440 --> 00:55:31.119
It's totally overkill.
00:55:31.599 --> 00:55:32.079
Yeah.
00:55:32.320 --> 00:55:32.559
Yeah.
00:55:34.079 --> 00:55:34.400
Yeah, yeah.
00:55:34.480 --> 00:55:40.239
I mean, so uh uh sounds like Word or any office products, yeah.
00:55:40.320 --> 00:55:43.039
That's uh like overkill for most cases.
00:55:43.280 --> 00:55:46.079
Yeah, yeah, that's uh that's
uh that's that's a nice one, right?
00:55:46.159 --> 00:55:49.760
So it's the I mean we we all always had the capability overhanging these models, right?
00:55:49.840 --> 00:55:52.239
That they can do much more than we can absorb, yeah.
00:55:52.639 --> 00:55:55.280
But maybe now it's getting way out of rack.
00:55:55.599 --> 00:56:01.280
Do you think we need to make like an uh IA uh writer version of the of an AI?
00:56:01.440 --> 00:56:05.519
Like that is very specific in its intention and you and usage?
00:56:05.920 --> 00:56:06.800
I have no idea.
00:56:06.960 --> 00:56:15.199
It's the um, I mean, but what I do know is that uh the open weight models, right, that tend to lag by half a year to a to a year.
00:56:15.519 --> 00:56:21.920
Now with all this capability overhang and you lagging by half a year, yeah, actually gives you a really capable model.
00:56:22.079 --> 00:56:22.400
Yeah.
00:56:22.559 --> 00:56:22.880
Right?
00:56:23.039 --> 00:56:27.280
Not Mythos level model, but you can't apply that anyway.
00:56:27.360 --> 00:56:29.679
It's it's like a thousand horsepower sports car, right?
00:56:29.760 --> 00:56:33.119
It's it's great to show off when you go get ice cream, right?
00:56:33.199 --> 00:56:38.960
But there's no chance in hell that you can deploy even 300 of those thousand horsepowers and and survive.
00:56:39.119 --> 00:56:42.639
And survive because that seems to be like the hardest part.
00:56:42.800 --> 00:56:46.800
Yeah, yeah, but but but just find a spot where you can unleash a thousand.
00:56:46.960 --> 00:56:47.440
Yeah, yeah, yeah.
00:56:47.679 --> 00:56:47.840
Right?
00:56:48.079 --> 00:56:51.679
Even when but while while harming yourself, I mean it's just almost impossible, right?
00:56:51.920 --> 00:56:57.519
The Autobahn, it's I mean, you can do it technically, but it's uh it's unmanageable most of the cases.
00:56:57.760 --> 00:56:58.800
It's way too busy, yeah.
00:56:58.960 --> 00:56:59.760
It's way too busy.
00:56:59.840 --> 00:57:07.440
It's the uh this this is and the reason that I brought it brought it up was this uh direction that Apple also goes, right?
00:57:07.519 --> 00:57:10.719
It's uh it seems like they're hinting to okay.
00:57:10.880 --> 00:57:28.960
Now we still have the really big uh uh places uh like the big data centers to create our AI experience, but we expect within uh uh now and five years that all the AI capabilities will be in your pocket, yeah, will be in your device.
00:57:29.280 --> 00:57:30.079
Could very well be.
00:57:30.320 --> 00:57:32.800
So this is an this is very interesting, right?
00:57:32.880 --> 00:57:43.119
It probably means that a lot of investors are now break pumping the brakes on creating all those data centers because they might be obsolete for most applications, yeah.
00:57:43.760 --> 00:57:50.880
Unless, of course, we find a way to absorb the the the all all these wonderful leading edge frontier capabilities.
00:57:51.199 --> 00:57:53.119
I mean, if you build it, people will come, right?
00:57:53.199 --> 00:57:54.400
That's the th that's what I say.
00:57:54.639 --> 00:57:55.199
That's a theory.
00:57:55.360 --> 00:57:56.480
Yeah, yeah, yeah, yeah.
00:57:56.559 --> 00:58:02.320
I mean it's the uh but if you see, I mean, yeah, so it's no I wanted to say something really smart, but but I forgot about it.
00:58:03.840 --> 00:58:06.960
Yeah, but I but to me it's like oh that's interesting.
00:58:07.119 --> 00:58:16.159
I didn't I expected data centers to become obsolete because it cut this this is an unsustainable or it seems to be an unsustainable growth, right?
00:58:16.239 --> 00:58:46.639
It's like we have seen in the past, the switch from uh going uh to be centralized and decentralized compute power, working with terminals and with without terminals, uh having a all kinds of like a plethora of websites on the internet, and all of a sudden concentration to Twitter, Facebook, and some other limited platforms that we now see as like flakes of the internet or flakes of the function of the internet.
00:58:47.039 --> 00:58:50.960
I think a similar thing might also be going on with the with the AI.
00:58:51.119 --> 00:59:02.559
It's like okay, now we harnessed the economic power to put all the intelligence into a small chip that's manageable in your pocket or manageable in your own computer.
00:59:03.199 --> 00:59:06.960
Yeah, I was about to say that on your own laptop or your own Mac Mini, right?
00:59:07.119 --> 00:59:09.599
I mean it's the uh yeah, totally.
00:59:10.000 --> 00:59:36.159
Right, and that pleases me a lot because that that's also and it might also be a uh well, I wouldn't say an explicit knot, but definitely a solution to people that are very worried about the environmental impact that AI might have, not only like uh the energy consumption, but also let's say the noise and the amount of space it takes, etc.
00:59:37.199 --> 00:59:40.639
And that so that might be might be an interesting relief for people.
00:59:40.719 --> 00:59:42.480
It's like okay, we don't need them anymore.
00:59:42.559 --> 00:59:43.280
Oh, that's nice.
00:59:43.360 --> 00:59:45.840
We can spend our monies also on different things.
00:59:46.159 --> 00:59:48.000
Yeah, but I have no idea.
00:59:48.239 --> 00:59:48.800
Me neither.
00:59:49.599 --> 00:59:52.639
It sounded very promising because they kind of like hinted there, yeah.
00:59:52.719 --> 00:59:55.519
I mean, yeah, they hinted quite explicitly, yeah.
00:59:55.599 --> 00:59:58.000
But that but that but that is all for inference, right?
00:59:58.079 --> 01:00:02.000
I mean, yeah, training you won't do on your phone just using the year.
01:00:02.159 --> 01:00:03.599
So it's the year, yeah, definitely.
01:00:03.760 --> 01:00:07.119
So we have vineyards where we grow our models, yep.
01:00:07.360 --> 01:00:12.639
And then we have these little pocket book uh uh bottles where we uh zip from our wine.
01:00:13.440 --> 01:00:18.400
Secret flasks, yes, yes, yeah, yeah, okay, yeah, yeah, yeah.
01:00:18.559 --> 01:00:20.800
That's I think that that'll that will be.
01:00:20.960 --> 01:00:31.360
I'm not sure if you would put a vintage wine in in a uh in a pocket bottle, but nevertheless, it I think that's uh that's great.
01:00:31.599 --> 01:00:52.960
I think I I've got some work to do because I totally forgot about working out that uh vineyard um uh yeah, storyline to to uh like write write an article about that because I think I'm not sure, but I try to research the the visuals I wanted to make, and I haven't found them yet.
01:00:53.199 --> 01:00:55.760
I haven't found something equivalent.
01:00:56.000 --> 01:01:00.000
I I tend to make like works that are kind of like expressive, right?
01:01:00.400 --> 01:01:10.559
They're very um kind of like imaginary rather than a system model, like with okay, you got a note over here, a note over there, lines, etc.
01:01:12.719 --> 01:01:14.000
Blueprints are awesome.
01:01:14.159 --> 01:01:15.840
Yeah, you're a UML fan, right?
01:01:15.920 --> 01:01:18.239
You have uh I used to be a UML fan, yeah.
01:01:18.639 --> 01:01:22.719
Spent certainly five years of my life uh on the ins and outs of UML.
01:01:23.199 --> 01:01:23.760
Yeah, same.
01:01:23.920 --> 01:01:34.800
I I've uh this is one of my uh areas of uh of my study is uh using UML in a certain way, also with rapid application development and stuff like that.
01:01:34.960 --> 01:01:38.800
Yeah, and um I'm very glad that I learned it.
01:01:38.880 --> 01:01:41.760
I'm very happy that I don't have to use it anymore.
01:01:42.079 --> 01:01:47.039
It's the uh we now have a new way of doing exactly what we did back then.
01:01:47.920 --> 01:01:54.639
I think the bait of the past will be very glad in how we now build our worlds and our products.
01:01:54.719 --> 01:01:55.280
Yeah, yeah.
01:01:55.440 --> 01:01:58.079
I think that will be uh that's uh that's wonderful.
01:01:58.800 --> 01:02:04.400
And with that, I think this is also a wonderful uh conversation once again.
01:02:04.719 --> 01:02:13.039
Let's not let's not come with because I I'd really love to um continue this series next time.
01:02:13.199 --> 01:02:14.000
Yes, definitely.
01:02:14.320 --> 01:02:14.800
When Dr.
01:02:14.880 --> 01:02:17.119
Bob said oh Dr.
01:02:17.199 --> 01:02:17.519
Bob?
01:02:18.480 --> 01:02:18.639
Dr.
01:02:18.800 --> 01:02:21.440
Bob Link in the show in Snow Globes.
01:02:23.199 --> 01:02:24.000
Okay, everybody.
01:02:24.079 --> 01:02:25.679
This was It's Just the Model.
01:02:25.840 --> 01:02:32.639
If you want to know more or you want to see the wonderful snow globes of Rome, go to it'sjustamodel.com.
01:02:32.880 --> 01:02:38.079
There's also links to uh subscribe and stuff like that, and share it with your friends.
01:02:38.320 --> 01:02:40.239
And um I'll see you next time.
01:02:40.480 --> 01:02:40.960
Next time.