Unmasking AI Hype: Truth Mode, Content Slop, & more Dr. Arshavir Blackwell | Bankedoutt Radio Show

WEBVTT

00:00:00.000 --> 00:00:01.800
Hey y 'all, welcome back to Bankedoutt Radio Show

00:00:01.800 --> 00:00:05.519
Music and Business Talk. Every single business

00:00:05.519 --> 00:00:08.460
owner is being bombarded with a tech marketing

00:00:08.460 --> 00:00:10.839
hype telling them to let AI run their company.

00:00:11.619 --> 00:00:14.880
But what happens when your brand loses its heartbeat?

00:00:15.759 --> 00:00:18.739
Our guest today has been building artificial

00:00:18.739 --> 00:00:22.300
intelligence since long before it became a trending

00:00:22.300 --> 00:00:26.559
buzzword. Dr. Arshavir Blackwell holds a PhD

00:00:26.559 --> 00:00:29.199
in cognitive science and brings over 30 years

00:00:29.199 --> 00:00:32.899
of raw frontline tech experience to the table.

00:00:33.539 --> 00:00:35.799
He is an elite strategist who helps creators

00:00:35.799 --> 00:00:38.640
and founders leverage advanced machine learning

00:00:38.640 --> 00:00:42.079
without losing their unique sovereign human voice.

00:00:43.060 --> 00:00:46.320
Dr. Blackwell, welcome to the platform. Thank

00:00:46.320 --> 00:00:50.759
you very much. It's great to be here. So there

00:00:50.759 --> 00:00:54.469
is a a couple of videos I was watching as in

00:00:54.469 --> 00:00:56.530
preparation for this interview that you did.

00:00:58.350 --> 00:01:01.490
And one of the things I wanted to get into is

00:01:01.490 --> 00:01:06.409
understanding like AI, how consumers use it.

00:01:06.689 --> 00:01:12.430
So like, you know, chat, GPT, little Gemini are

00:01:12.430 --> 00:01:16.269
from my understanding, like the frontier of AI.

00:01:17.290 --> 00:01:19.430
And then those are the frontier models. That's

00:01:19.430 --> 00:01:23.239
right. You also have what you call the local

00:01:23.239 --> 00:01:26.939
model. Yes. Which is running the application

00:01:26.939 --> 00:01:30.159
basically on your own particular device. Can

00:01:30.159 --> 00:01:33.280
you give us a little more brief explanation of

00:01:33.280 --> 00:01:38.180
what that is? Sure. So Frontier models are the

00:01:38.180 --> 00:01:40.980
ones that everybody probably, for the most part,

00:01:41.140 --> 00:01:46.640
is using. And those are things like Claude, JetGPT,

00:01:46.900 --> 00:01:51.439
Gemini, and so forth. those are all running in

00:01:51.439 --> 00:01:55.079
the cloud. And the advantages of them are they're

00:01:55.079 --> 00:01:58.459
very powerful, of course, as we all know. The

00:01:58.459 --> 00:02:01.540
disadvantages of them are amongst, well, first

00:02:01.540 --> 00:02:03.519
of all, they get to be expensive. If you use

00:02:03.519 --> 00:02:06.280
them a lot, you notice that you're burning through

00:02:06.280 --> 00:02:08.680
a lot of tokens, and that can add up very quickly.

00:02:09.719 --> 00:02:11.860
The other disadvantage, well, there's actually

00:02:11.860 --> 00:02:14.419
two other disadvantages. One is it's not secure.

00:02:14.650 --> 00:02:18.030
So you can't really expect that whatever you're

00:02:18.030 --> 00:02:23.030
going to send out over the wire is going to be

00:02:23.030 --> 00:02:26.150
preserved and secure and safe. They could very

00:02:26.150 --> 00:02:30.349
well use what you're sending them for training

00:02:30.349 --> 00:02:32.990
for their own material. And of course people

00:02:32.990 --> 00:02:35.370
can intercept along the way and there's other

00:02:35.370 --> 00:02:39.110
security concerns as well. The other thing that

00:02:39.110 --> 00:02:42.110
a frontier model cannot do is it can't really

00:02:42.110 --> 00:02:45.930
speak in your own voice. And the reason for that

00:02:45.930 --> 00:02:48.469
is because you can't really, they're trained

00:02:48.469 --> 00:02:51.689
already. They come pre -trained. When I work

00:02:51.689 --> 00:02:53.909
with a local business in Macon, Georgia, we took

00:02:53.909 --> 00:02:57.069
their operational revenue from 45K to 88K in

00:02:57.069 --> 00:02:59.849
a single year. How? We didn't chase social media

00:02:59.849 --> 00:03:02.870
clout. We found hidden billing programs, cut

00:03:02.870 --> 00:03:05.669
cash based operational leaks, and fixed their

00:03:05.669 --> 00:03:07.569
pricing structures. Most independent creators,

00:03:07.810 --> 00:03:09.849
bookkeepers, and estheticians are amazing at

00:03:09.849 --> 00:03:12.409
their craft, but they run their back ends like

00:03:12.409 --> 00:03:14.789
an expensive charity. They carry uncollected

00:03:14.789 --> 00:03:17.210
fees and think customers would just appear out

00:03:17.210 --> 00:03:19.370
of nowhere. Don't let operational blindness kill

00:03:19.370 --> 00:03:21.789
your vision. The worksheets inside my guy will

00:03:21.789 --> 00:03:24.449
help you audit your company structure in 48 hours.

00:03:24.569 --> 00:03:27.129
Grab your copy right now for $19 .99. Type here

00:03:27.129 --> 00:03:29.110
to buy the e -book. I mean, that's in the name,

00:03:29.169 --> 00:03:33.340
right? So because of that. There's really no

00:03:33.340 --> 00:03:35.680
way, there are no levers that you can use to

00:03:35.680 --> 00:03:38.719
really change what the nature of the output is,

00:03:39.340 --> 00:03:41.400
other than what you put in the prompt. And what

00:03:41.400 --> 00:03:43.860
we find is that the prompt is really not a very

00:03:43.860 --> 00:03:46.740
good way to do that. So you can give it instructions

00:03:46.740 --> 00:03:50.599
about what you want it to sound like in the prompt.

00:03:51.199 --> 00:03:54.280
But what happens very often is you get a costume.

00:03:54.780 --> 00:03:57.979
So you don't get genuine sounding, authentic

00:03:57.979 --> 00:04:02.110
changes. You get a sort of performative version

00:04:02.110 --> 00:04:04.590
where it just sounds very fake and it sounds

00:04:04.590 --> 00:04:08.289
like it's kind of trying to put on whatever tone

00:04:08.289 --> 00:04:10.770
you've asked it to do, but it really can't do

00:04:10.770 --> 00:04:13.689
that. Now what I work with are what are called

00:04:13.689 --> 00:04:17.290
local large language models, and they're local

00:04:17.290 --> 00:04:20.430
because you can run them either on your own computer

00:04:20.430 --> 00:04:24.110
or we run them also on a dedicated server, but

00:04:24.110 --> 00:04:26.129
one that's not connected to Frontier models,

00:04:26.209 --> 00:04:30.259
one that's not connected to chat GPT. And you

00:04:30.259 --> 00:04:34.860
can actually train those on your own material.

00:04:35.899 --> 00:04:40.180
And those are not in their responses performative.

00:04:40.519 --> 00:04:42.720
Because when you train it on your own material,

00:04:43.060 --> 00:04:45.459
you're not training it through the prompt. You're

00:04:45.459 --> 00:04:48.860
actually changing the nature of the learning

00:04:48.860 --> 00:04:52.660
in the model. And so what we do is we can actually

00:04:52.660 --> 00:04:56.040
have it read. Let's say you have a blog or a

00:04:56.040 --> 00:04:59.800
substack or a bunch of marketing material. it

00:04:59.800 --> 00:05:02.620
actually reads through all of that and retrains

00:05:02.620 --> 00:05:08.579
itself to speak in your voice. And the advantage

00:05:08.579 --> 00:05:10.680
of that is that you don't have to spend a lot

00:05:10.680 --> 00:05:14.819
of time on chat GPT or on cloud having it write

00:05:14.819 --> 00:05:16.540
something and then having you have to rewrite

00:05:16.540 --> 00:05:19.120
it. I mean, it kind of misses the whole point

00:05:19.120 --> 00:05:22.019
when you have to do that. So that's basically

00:05:22.019 --> 00:05:24.720
what we're working with here is a local model

00:05:24.720 --> 00:05:27.560
that's safe. It's much cheaper because you're

00:05:27.560 --> 00:05:30.560
not burning up tokens on some huge frontier model.

00:05:31.100 --> 00:05:34.860
And it's trained on your own voice. So it sounds

00:05:34.860 --> 00:05:38.439
like you. How long would it take to be able to

00:05:38.439 --> 00:05:42.199
do that, to train your own voice? How much material

00:05:42.199 --> 00:05:47.060
would you have to input into it? Yeah. So the

00:05:47.060 --> 00:05:49.720
training time is maybe 10 minutes, 15 minutes

00:05:49.720 --> 00:05:53.560
max. It's not very long. And you only have to

00:05:53.560 --> 00:05:55.560
do that the first time you use it. And then after

00:05:55.560 --> 00:05:58.920
that, it's already trained. The more information

00:05:58.920 --> 00:06:01.720
it has, so let's say you have a large size blog,

00:06:01.920 --> 00:06:05.519
it would certainly be enough. If you have a lot

00:06:05.519 --> 00:06:08.540
of emails that you've written, so they're all

00:06:08.540 --> 00:06:11.420
in your own voice, that would be enough. With

00:06:11.420 --> 00:06:14.920
mine, I train it on about... oh, 50 ,000 words

00:06:14.920 --> 00:06:17.259
and it does pretty well. And that would be kind

00:06:17.259 --> 00:06:20.379
of, that's not huge, but that's kind of the size

00:06:20.379 --> 00:06:24.519
of maybe a medium sized blog or mixed in there

00:06:24.519 --> 00:06:27.300
with perhaps marketing documents that you've

00:06:27.300 --> 00:06:29.839
got or some other material. So it doesn't have

00:06:29.839 --> 00:06:31.620
to be just one thing. Now, if you want it to

00:06:31.620 --> 00:06:34.000
be in a particular style, everything that's written

00:06:34.000 --> 00:06:37.040
in it. So if you're doing marketing for a certain

00:06:37.040 --> 00:06:40.360
product, you shouldn't mix it in with your blog

00:06:40.360 --> 00:06:42.800
about cooking. or something like that. Those

00:06:42.800 --> 00:06:46.980
should be two separate ones. But you can actually

00:06:46.980 --> 00:06:51.360
get a pretty good result with a relatively small

00:06:51.360 --> 00:06:53.860
amount of information trained relatively quickly.

00:06:57.420 --> 00:07:00.939
So, Dears, you've been in this space for 30 years.

00:07:01.139 --> 00:07:04.240
Yeah. How are people using it wrong nowadays,

00:07:04.339 --> 00:07:06.699
especially when it comes to small business owners?

00:07:08.040 --> 00:07:11.139
Well, I've been doing this since before it was

00:07:11.139 --> 00:07:13.699
called AI back when I first started doing it.

00:07:13.839 --> 00:07:16.120
We called it neural networks. And that's essentially

00:07:16.120 --> 00:07:18.800
what we're calling AI now is basically just a

00:07:18.800 --> 00:07:23.600
very, very sophisticated neural network. I started

00:07:23.600 --> 00:07:26.740
at UCSD and I was very, very fortunate to work

00:07:26.740 --> 00:07:29.480
with two of the pioneers in the field of neural

00:07:29.480 --> 00:07:31.660
networks. Elizabeth Bates, who was a linguist

00:07:31.660 --> 00:07:34.180
who did a lot of work in understanding out humans

00:07:34.180 --> 00:07:36.699
process language, but how neural network could

00:07:36.699 --> 00:07:39.660
do it as well. And then her colleague, Jeff Elman,

00:07:39.839 --> 00:07:42.899
who was also someone who did a lot of the real

00:07:42.899 --> 00:07:46.379
pioneering works with neural networks and artificial

00:07:46.379 --> 00:07:50.720
intelligence. What was happening back then that's

00:07:50.720 --> 00:07:53.459
contrasted with now and what it's really changed

00:07:53.459 --> 00:08:00.180
is the models were much smaller and the designs

00:08:00.180 --> 00:08:03.459
for the models were small but also the computers

00:08:03.459 --> 00:08:07.939
that we were running them on were you know much

00:08:07.939 --> 00:08:11.160
less powerful than these huge data centers that

00:08:11.160 --> 00:08:13.680
we have access to today. I mean orders and orders

00:08:13.680 --> 00:08:16.980
of magnitude difference. So what happened was

00:08:16.980 --> 00:08:20.839
about oh you know maybe four or five years ago

00:08:21.259 --> 00:08:25.459
there was a sort of revolution that came about

00:08:25.459 --> 00:08:30.000
as a confluence of new algorithms, more powerful

00:08:30.000 --> 00:08:35.779
kinds of computers to run them on, parallel algorithms,

00:08:36.100 --> 00:08:38.860
parallel computers. And that's where we've gotten

00:08:38.860 --> 00:08:43.580
to today. Now, the thing is that a lot of, I

00:08:43.580 --> 00:08:46.919
hope that a lot of businesses are at least dabbling

00:08:46.919 --> 00:08:51.000
in or getting used to using things like chat

00:08:51.000 --> 00:08:54.539
GPT and Claude, because these are very, very

00:08:54.539 --> 00:08:58.779
powerful tools. And you can do I mean, literally,

00:08:58.779 --> 00:09:02.720
I'm able to do weeks worth of work in a few hours.

00:09:02.960 --> 00:09:06.039
So if you get to know them, well, it can really

00:09:06.039 --> 00:09:09.279
as a small business owner, it can really be a

00:09:09.279 --> 00:09:13.059
huge benefit. But the problem that we see is

00:09:13.059 --> 00:09:16.549
very often people have businesses And there are

00:09:16.549 --> 00:09:19.830
studies to show this. When these systems are

00:09:19.830 --> 00:09:23.330
implemented and when they fail, a big problem

00:09:23.330 --> 00:09:28.889
is that businesses don't integrate the AI workflow

00:09:28.889 --> 00:09:32.529
into the rest of their work. So the AI becomes

00:09:32.529 --> 00:09:39.090
sort of this thing over there that is bolted

00:09:39.090 --> 00:09:42.970
on, I guess you could say. to the workflow and

00:09:42.970 --> 00:09:46.210
people are copying pasting from that back to

00:09:46.210 --> 00:09:48.490
their regular workflow and then back to the you

00:09:48.490 --> 00:09:54.190
know the AI system and so on and so forth and

00:09:54.190 --> 00:09:57.230
so because of that a lot of people really just

00:09:57.230 --> 00:09:59.590
kind of give up on using it because they don't

00:09:59.590 --> 00:10:01.490
really understand the power of it and they don't

00:10:01.490 --> 00:10:03.570
understand how to integrate it into their work

00:10:03.570 --> 00:10:06.350
so I would say as a small business person my

00:10:06.350 --> 00:10:09.639
first Suggestion to be really like use this stuff

00:10:09.639 --> 00:10:11.860
so that you really feel comfortable with it But

00:10:11.860 --> 00:10:15.139
second of all you need to figure out how to integrate

00:10:15.139 --> 00:10:18.679
it into your ongoing workflow and not Throw it

00:10:18.679 --> 00:10:21.120
at the people that are doing the work and just

00:10:21.120 --> 00:10:24.539
say, you know, good luck. Here we go Do what

00:10:24.539 --> 00:10:26.840
you can with it they need to people sort of need

00:10:26.840 --> 00:10:30.240
to be led to the right ways and the right use

00:10:30.240 --> 00:10:37.360
cases and the proper ways to integrate AI into

00:10:37.360 --> 00:10:39.279
their workflow. And that's a lot of work in and

00:10:39.279 --> 00:10:42.080
of itself. Yeah, I definitely get that because

00:10:42.080 --> 00:10:44.779
I would say that like one of the things I use

00:10:44.779 --> 00:10:47.799
AI for is with my podcast and my show notes.

00:10:48.179 --> 00:10:51.639
Yeah. So what I do after every single interview,

00:10:52.299 --> 00:10:56.360
I write a paragraph or two of what we discussed.

00:10:56.879 --> 00:11:00.600
Right. And then I have AI like, you know, match

00:11:00.600 --> 00:11:03.139
it or put it into the prompt until it's like,

00:11:03.139 --> 00:11:04.899
you know, based off the information I'm feeding

00:11:04.899 --> 00:11:08.220
it. It's writing a good description to where,

00:11:08.419 --> 00:11:12.159
when I post it on YouTube, and also like PiePage,

00:11:12.799 --> 00:11:17.620
it's able to be scoured by the internet. So therefore

00:11:17.620 --> 00:11:19.919
those keywords and all that start to show up

00:11:19.919 --> 00:11:23.440
and they're used in that language for the search

00:11:23.440 --> 00:11:25.539
engine. And I've seen that with my videos. They've

00:11:25.539 --> 00:11:28.419
been crawling higher and higher on Google search

00:11:28.419 --> 00:11:32.840
engines. Right. So you're using the AI basically

00:11:32.840 --> 00:11:36.149
to phrase things in a way that search engine

00:11:36.149 --> 00:11:39.990
optimization is going to pick up on. Yes. And

00:11:39.990 --> 00:11:42.230
that's a very good use of it. And I think that

00:11:42.230 --> 00:11:45.330
people are beginning to figure out how to integrate

00:11:45.330 --> 00:11:50.090
SEO search engine optimization with techniques

00:11:50.090 --> 00:11:52.850
like chat GPT or local models or whatever it

00:11:52.850 --> 00:11:54.870
is they're using. So that's actually a great,

00:11:54.870 --> 00:11:57.570
great. I mean, have you found it to be, it sounds

00:11:57.570 --> 00:12:00.429
like you found it to be successful. You're getting

00:12:00.429 --> 00:12:04.679
more hits on your podcasts and it it's working

00:12:04.679 --> 00:12:09.820
for you. Yeah. I don't like, you know, some people

00:12:09.820 --> 00:12:11.960
like even with like the thumbnails and stuff,

00:12:12.240 --> 00:12:14.440
I love pictures and things that I mentioned instead

00:12:14.440 --> 00:12:16.559
of trying to get it to create something just

00:12:16.559 --> 00:12:21.220
from typing in prompts. Yeah. And the artwork

00:12:21.220 --> 00:12:25.000
looks way better than a lot of people. So for

00:12:25.000 --> 00:12:27.080
example, people use it to let's say they have

00:12:27.080 --> 00:12:30.419
a small business and they deal with food. Yeah.

00:12:30.419 --> 00:12:33.039
Using a lot of AI generated food looking stuff

00:12:33.039 --> 00:12:35.200
in the mic. your cooking stuff doesn't even look

00:12:35.200 --> 00:12:37.519
like that, so a lot of people are rejecting those

00:12:37.519 --> 00:12:41.539
type of flyers and thumbnails and stuff. That's

00:12:41.539 --> 00:12:45.019
a problem, and there is a big issue, and I read

00:12:45.019 --> 00:12:48.600
a lot about this because it's, you know, out

00:12:48.600 --> 00:12:51.000
there, but also it's actually a topic I'm interested

00:12:51.000 --> 00:12:53.820
in. There's a big, it's very interesting because

00:12:53.820 --> 00:12:57.460
there's some people that feel as though AI is

00:12:57.460 --> 00:13:00.149
great and a blessing and it's going to do a lot

00:13:00.149 --> 00:13:04.070
for us. But there's also a very active school

00:13:04.070 --> 00:13:08.149
of thought out there that is in to greater and

00:13:08.149 --> 00:13:12.789
lesser degrees against using AI in these various

00:13:12.789 --> 00:13:16.210
business processes. Some of those are as simple

00:13:16.210 --> 00:13:18.110
as what you're talking about. Like, oh, the picture

00:13:18.110 --> 00:13:20.850
looks like it's AI, it's AI slop. That's the

00:13:20.850 --> 00:13:24.940
term people always use, AI slop. real food doesn't

00:13:24.940 --> 00:13:27.860
really look like that. Now there are ways to

00:13:27.860 --> 00:13:31.820
use AI in a way that you don't get AI slop. And

00:13:31.820 --> 00:13:33.919
the one thing that I think is really important

00:13:33.919 --> 00:13:37.019
and I always tell people is don't just take what

00:13:37.019 --> 00:13:41.179
the AI gives you and throw it out there. What

00:13:41.179 --> 00:13:44.980
you need to do is you need to test and check

00:13:44.980 --> 00:13:47.419
every single thing that it tells you because

00:13:47.419 --> 00:13:50.120
AI can be wrong and people don't realize that.

00:13:50.500 --> 00:13:54.289
One thing that AI is very good at is producing

00:13:54.289 --> 00:14:00.210
very slick, nice, what I call fluent looking

00:14:00.210 --> 00:14:04.049
output. So it looks really right. And I've written

00:14:04.049 --> 00:14:06.470
an article about this called Fluency is Validity.

00:14:06.730 --> 00:14:09.529
And the risk is that when something looks really

00:14:09.529 --> 00:14:11.850
right, looks well written, looks like it's arguing

00:14:11.850 --> 00:14:15.450
well, people very often, which is what AI is

00:14:15.450 --> 00:14:18.250
very strong at doing, people very often accept

00:14:18.250 --> 00:14:21.549
that at face value. and don't necessarily dive

00:14:21.549 --> 00:14:24.610
into what the actual argument is and whether

00:14:24.610 --> 00:14:29.129
it has merit or not. So AI can get take the most

00:14:29.129 --> 00:14:32.929
ridiculous sounding idea and make it sound very

00:14:32.929 --> 00:14:35.470
believable because that's what it's kind of programmed

00:14:35.470 --> 00:14:37.909
to do. It's programmed to continue to engage

00:14:37.909 --> 00:14:41.649
you. It's programmed to make you think that what

00:14:41.649 --> 00:14:43.970
you're talking about is the greatest idea in

00:14:43.970 --> 00:14:47.549
the world. We call this significance. And that's

00:14:47.549 --> 00:14:49.470
something that people have to guard against in

00:14:49.470 --> 00:14:51.929
any field, whether it's business or academia

00:14:51.929 --> 00:14:54.909
or whatever they're using it for. And there are

00:14:54.909 --> 00:14:57.789
a whole bunch of techniques that you can use

00:14:57.789 --> 00:15:00.909
and that people, I think, need to be more aware

00:15:00.909 --> 00:15:03.809
of that are actually good techniques when you're

00:15:03.809 --> 00:15:06.750
just dealing with media in general. So not even

00:15:06.750 --> 00:15:12.549
just AI, but also just Ways to verify and validate

00:15:12.549 --> 00:15:14.830
when you read an article whether it's generated

00:15:14.830 --> 00:15:17.970
by AI or whether it's generated by a human being

00:15:17.970 --> 00:15:21.149
or some combination thereof You need to check

00:15:21.149 --> 00:15:23.429
and verify every single thing in there and not

00:15:23.429 --> 00:15:26.070
take it for face value And I think that that's

00:15:26.070 --> 00:15:29.610
kind of the key to keeping a lot of this technology

00:15:29.610 --> 00:15:35.250
useful But not harming people Yeah, one things

00:15:35.250 --> 00:15:39.450
I discovered was like truth mode telling it to

00:15:39.450 --> 00:15:43.590
give me truthful answers that are based off of

00:15:43.590 --> 00:15:46.809
data. And now, like you said, talk to me like,

00:15:47.269 --> 00:15:49.529
oh, your ideas are great. And now that's why

00:15:49.529 --> 00:15:52.769
this was work, especially when it comes to marketing.

00:15:53.129 --> 00:15:56.289
Because I think a lot of times we're misled.

00:15:56.669 --> 00:16:00.470
Like you said, it's a believing that it sounds

00:16:00.470 --> 00:16:03.549
good. Right. But what's the old saying? If it

00:16:03.549 --> 00:16:07.029
sounds too good to be true, it's not. It probably

00:16:07.029 --> 00:16:09.850
is. Yeah. So I'm curious what, so when you use

00:16:09.850 --> 00:16:11.730
truth mode, what do you do exactly? How does

00:16:11.730 --> 00:16:15.070
that work for you? So me with creative ideas

00:16:15.070 --> 00:16:20.830
and also marketing driven data. So for instance,

00:16:21.049 --> 00:16:25.409
I'll give this example. I was asking about small

00:16:25.409 --> 00:16:27.850
businesses, especially in particular to like.

00:16:28.670 --> 00:16:33.029
how small businesses look at marketing and advertising

00:16:33.029 --> 00:16:35.429
and investing in their company. And it was telling

00:16:35.429 --> 00:16:39.649
me like, you know, what percentage of small businesses

00:16:39.649 --> 00:16:43.269
are using like, you know, AI and running marketing

00:16:43.269 --> 00:16:45.649
strategies and stuff like that. And then I wanted

00:16:45.649 --> 00:16:48.129
to go into more of a subsection of like, okay,

00:16:48.289 --> 00:16:52.470
so now black owned businesses, do they use marketing?

00:16:52.669 --> 00:16:54.190
And I was like, yeah, they rely on this and that

00:16:54.190 --> 00:16:56.419
and all this stuff. And then I was going back

00:16:56.419 --> 00:16:58.019
and forth with it because I was telling it based

00:16:58.019 --> 00:16:59.960
on my experience of, you know, doing marketing

00:16:59.960 --> 00:17:02.139
for 20 years and working a lot with black owned

00:17:02.139 --> 00:17:06.359
businesses. The ones I run across and of course,

00:17:06.359 --> 00:17:07.839
you know, you can't group everybody together

00:17:07.839 --> 00:17:10.240
because everybody's business is different. Right.

00:17:10.299 --> 00:17:14.400
But generally speaking, if they fall within,

00:17:14.619 --> 00:17:17.940
let's say that I like to call the infancy stage

00:17:17.940 --> 00:17:20.099
or whatever, you know, zero years to just getting

00:17:20.099 --> 00:17:22.559
started to about three or four years. Right.

00:17:22.559 --> 00:17:26.589
It was negative condensation. when it comes to

00:17:26.589 --> 00:17:29.349
wanting to spend money on marketing and thinking

00:17:29.349 --> 00:17:31.970
that they could just do everything for free basically

00:17:31.970 --> 00:17:35.329
on social media and not get somebody's third

00:17:35.329 --> 00:17:39.750
party's perspective on how they should brand

00:17:39.750 --> 00:17:41.970
their business in order to attract the clients

00:17:41.970 --> 00:17:45.170
and understand that it's a process with like

00:17:45.170 --> 00:17:49.250
a sales funnel and all that. And when I was getting

00:17:49.250 --> 00:17:51.009
the information from AI, I was like, you know,

00:17:51.009 --> 00:17:54.420
just basically trying to give me like data, but

00:17:54.420 --> 00:17:57.220
in truth, I was like, no, this is not being the

00:17:57.220 --> 00:17:59.460
experiences I've had. I've actually have a lot

00:17:59.460 --> 00:18:01.539
of people who actually run from it because they're

00:18:01.539 --> 00:18:04.759
fearful and they also have like a like a trauma

00:18:04.759 --> 00:18:08.299
associated with doing this because they will

00:18:08.299 --> 00:18:12.700
have an experience with maybe somebody and they'll

00:18:12.700 --> 00:18:15.220
have that bad experience. So then they group

00:18:15.220 --> 00:18:18.339
everybody else who could help them with their

00:18:18.339 --> 00:18:23.140
business within marketing in the same they'll

00:18:23.140 --> 00:18:26.980
put them in the same monolithic group. It's harder

00:18:26.980 --> 00:18:29.940
to get that sale, the next person who comes behind

00:18:29.940 --> 00:18:32.799
someone, whoever gave someone a bad experience.

00:18:32.880 --> 00:18:36.160
Right. So they overgeneralized from one bad to

00:18:36.160 --> 00:18:39.759
another. Yeah, it did. And then when I was going

00:18:39.759 --> 00:18:42.599
back and forth with that, we gained to a consensus

00:18:42.599 --> 00:18:49.240
like, no, you're right, that this isn't... everyone

00:18:49.240 --> 00:18:52.779
doesn't just look at stuff from this, I guess,

00:18:52.960 --> 00:18:58.779
generic point of view. Yeah. So it was very interesting.

00:18:59.200 --> 00:19:01.059
And then I always ask it, are you still in truth

00:19:01.059 --> 00:19:02.440
mode? Because I don't like to take them out of

00:19:02.440 --> 00:19:05.420
truth mode. Because I won't, when I have ideas,

00:19:05.640 --> 00:19:11.619
especially creative ideas, and how will this

00:19:11.619 --> 00:19:14.619
idea, you know, what's the, I guess the probability

00:19:14.619 --> 00:19:17.819
of it doing well in the marketplace? Yeah. I

00:19:17.819 --> 00:19:20.819
want something that's based off of data and what

00:19:20.819 --> 00:19:22.779
other people have done that's probably similar

00:19:22.779 --> 00:19:26.019
to it. And that's something that's just going

00:19:26.019 --> 00:19:29.859
to agree with me. If it doesn't, if it's a bad

00:19:29.859 --> 00:19:33.920
idea, just tell me it's a bad idea. Right. And

00:19:33.920 --> 00:19:35.920
that's the problem is a lot of times. So do you

00:19:35.920 --> 00:19:38.079
find when you put it in truth mode, is it actually

00:19:38.079 --> 00:19:42.259
truthful? I've had some things that really appear

00:19:42.259 --> 00:19:46.180
to be truthful. Of course, like you said, it

00:19:46.180 --> 00:19:49.599
can make some mistakes, but I think I use AI

00:19:49.599 --> 00:19:51.799
differently than a lot of other people because

00:19:51.799 --> 00:19:55.759
I'm going back and forth with it until I know

00:19:55.759 --> 00:19:57.900
based off of my experience and education of what

00:19:57.900 --> 00:20:00.599
I've been able to do and what would more than

00:20:00.599 --> 00:20:03.759
likely be the outcome if this particular strategy

00:20:03.759 --> 00:20:06.519
was implemented as opposed to, you know, just

00:20:06.519 --> 00:20:11.089
taking everything at face value. So I don't know

00:20:11.089 --> 00:20:13.109
if I don't I don't know a lot of people will

00:20:13.109 --> 00:20:16.710
go back and forth with it. Yeah. So you're saying

00:20:16.710 --> 00:20:20.890
most people just type something in and get something

00:20:20.890 --> 00:20:24.410
out and then they run with that. And you have

00:20:24.410 --> 00:20:27.569
a dialogue with it. I do every single time I'm

00:20:27.569 --> 00:20:31.710
having to die for with it. That's very helpful.

00:20:32.309 --> 00:20:35.269
Exactly. Because that's and that's the thing

00:20:35.269 --> 00:20:38.980
is that's an aspect of it. of AI that I think

00:20:38.980 --> 00:20:43.160
that a lot of people aren't really using. And

00:20:43.160 --> 00:20:47.019
that's what I do as well. So many people just

00:20:47.019 --> 00:20:48.759
say, oh, I'm going to type something in. It's

00:20:48.759 --> 00:20:50.259
like a search engine. It's going to give me an

00:20:50.259 --> 00:20:52.359
answer. And I'll just paste that in. But actually,

00:20:52.720 --> 00:20:57.960
it can be much more useful as a thought partner.

00:20:58.960 --> 00:21:02.019
And so to do exactly the kinds of things you're

00:21:02.019 --> 00:21:04.539
doing where you're saying, oh, here's an idea

00:21:04.539 --> 00:21:06.720
I had. What are the good aspects of it? But what

00:21:06.720 --> 00:21:08.839
are the, you know, tell me some of the reasons

00:21:08.839 --> 00:21:11.799
maybe this isn't a good idea. And then have it

00:21:11.799 --> 00:21:14.660
give you its response. And then you say, but

00:21:14.660 --> 00:21:17.099
what about this? What about that? What about

00:21:17.099 --> 00:21:22.259
the other thing? And a good AI system can help

00:21:22.259 --> 00:21:26.200
to expose a lot of the assumptions in a lot of

00:21:26.200 --> 00:21:29.640
the ideas. that you're working through, but you're

00:21:29.640 --> 00:21:32.880
not going to get that if you just type in a single

00:21:32.880 --> 00:21:36.220
prompt and get a response and paste it in. And

00:21:36.220 --> 00:21:40.259
I think that's where a lot of the value of AI

00:21:40.259 --> 00:21:43.960
is being left on the table. People are using

00:21:43.960 --> 00:21:46.460
it the way you're using it, which is really the

00:21:46.460 --> 00:21:49.480
right way to use it, which is as an interactive

00:21:49.480 --> 00:21:54.119
partner and not as the final say of whatever

00:21:54.119 --> 00:21:58.910
it is you're trying to do. And so I would say

00:21:58.910 --> 00:22:02.450
my other suggestion to business owners could

00:22:02.450 --> 00:22:05.049
be to do exactly what you're doing, which is

00:22:05.049 --> 00:22:08.309
great, which is to not think of it as an Oracle

00:22:08.309 --> 00:22:11.609
that's gonna give you a final answer, but rather

00:22:11.609 --> 00:22:14.869
think of it as a thinking partner that you can

00:22:14.869 --> 00:22:17.490
bounce ideas off of. And maybe it will come up

00:22:17.490 --> 00:22:19.210
with new ideas that you haven't come up with,

00:22:19.210 --> 00:22:23.420
but you still need to. evaluate and decide if

00:22:23.420 --> 00:22:25.799
those ideas are actually as good as it seems

00:22:25.799 --> 00:22:28.339
to think they are. So that's the direction I'd

00:22:28.339 --> 00:22:31.779
like to see people leading into. What do you

00:22:31.779 --> 00:22:35.420
say when people, you know, they use AI and I've

00:22:35.420 --> 00:22:38.579
seen headlines in media where they're talking

00:22:38.579 --> 00:22:40.740
about like the loss of jobs and things of that

00:22:40.740 --> 00:22:44.539
nature and seems more of a scare tactic than

00:22:44.539 --> 00:22:50.960
an educational piece of information. Well, When

00:22:50.960 --> 00:22:54.660
you have any new technology, historically speaking,

00:22:55.180 --> 00:22:57.740
what's happened is there's been a loss of jobs,

00:22:57.980 --> 00:23:00.319
but there hasn't been a loss of employment because

00:23:00.319 --> 00:23:04.380
what happens is that the new technology shifts

00:23:04.380 --> 00:23:06.799
and creates all sorts of new work for people

00:23:06.799 --> 00:23:09.839
to do that didn't exist before. And very often

00:23:09.839 --> 00:23:11.640
that work is actually a lot more interesting

00:23:11.640 --> 00:23:14.079
than what people were doing before. So at one

00:23:14.079 --> 00:23:17.450
time you had rooms full of people with those

00:23:17.450 --> 00:23:20.589
electronic calculators doing banking business,

00:23:20.609 --> 00:23:23.529
for example, and putting things in folders and

00:23:23.529 --> 00:23:26.970
in filing cabinets. And then when the computer

00:23:26.970 --> 00:23:29.490
came along and all of that became automated,

00:23:30.289 --> 00:23:32.630
those people didn't have a job anymore, but they

00:23:32.630 --> 00:23:35.910
were still able to do, you know, to be supervising

00:23:35.910 --> 00:23:38.890
the computers or supervising the work of the

00:23:38.890 --> 00:23:41.529
bank in some way or another, or often different

00:23:41.529 --> 00:23:45.960
jobs. And so far, every single technological

00:23:45.960 --> 00:23:50.720
advancement has actually led to an increase in

00:23:50.720 --> 00:23:53.859
employment overall. Now, I mean, admittedly,

00:23:53.859 --> 00:23:56.660
there's disruption because people need to learn

00:23:56.660 --> 00:24:02.579
new skills. They may need to retool. Sometimes

00:24:02.579 --> 00:24:04.880
if someone's been in the workforce for a while,

00:24:05.180 --> 00:24:07.579
they may be so used to doing what they're doing

00:24:07.579 --> 00:24:10.180
that it's a burden for them to learn this set

00:24:10.180 --> 00:24:13.559
of new skills. So I'm not saying that it's without

00:24:13.599 --> 00:24:16.079
some degree of cost, but I don't think there's

00:24:16.079 --> 00:24:21.099
a risk of AI taking over and nobody having any

00:24:21.099 --> 00:24:23.339
work anymore and everyone being out of a job

00:24:23.339 --> 00:24:25.579
because again, there's always going to be some

00:24:25.579 --> 00:24:30.380
things that can't do that humans do better. So

00:24:30.380 --> 00:24:36.799
with the greater industries, I know that there's

00:24:36.799 --> 00:24:38.940
been a disruption, let's say people who design

00:24:38.940 --> 00:24:41.099
flyers and stuff like that, because a lot of

00:24:41.099 --> 00:24:44.259
times they have to design flyers. And then they

00:24:44.259 --> 00:24:49.019
say that photography and videography work, I

00:24:49.019 --> 00:24:51.460
don't necessarily believe that because I don't

00:24:51.460 --> 00:24:53.759
see any AI walking around here actually shooting

00:24:53.759 --> 00:24:57.680
stuff. And regardless of how much people shoot

00:24:57.680 --> 00:25:00.759
and upload, everyone's eye is just different.

00:25:00.839 --> 00:25:02.400
You're going to capture something different.

00:25:02.980 --> 00:25:05.299
What do you say to people like that as far as

00:25:05.299 --> 00:25:07.819
it disrupting those industries and they feeling

00:25:07.819 --> 00:25:11.940
that within X amount of years, AI can... either

00:25:11.940 --> 00:25:16.420
take over or substantially cut those jobs out.

00:25:17.599 --> 00:25:19.240
I mean, I think it's a very good possibility,

00:25:19.240 --> 00:25:22.660
again, that there are certain job categories.

00:25:23.140 --> 00:25:24.859
You know, if you're talking about flyers being

00:25:24.859 --> 00:25:28.420
handed out, I mean, already we have the Internet,

00:25:28.579 --> 00:25:31.720
which has made using the mail and putting flyers

00:25:31.720 --> 00:25:34.859
in the mail obsolete for many, many years now

00:25:34.859 --> 00:25:37.940
and has. But again, let's take a look at what

00:25:37.940 --> 00:25:42.819
happened. So. department stores and catalog sales

00:25:42.819 --> 00:25:46.960
suffered. But what you have now is this huge

00:25:46.960 --> 00:25:52.619
online booming internet economy. And there's

00:25:52.619 --> 00:25:57.640
lots of jobs there. And so what you have is not

00:25:57.759 --> 00:25:59.839
that people aren't buying anything anymore. It's

00:25:59.839 --> 00:26:02.380
just that they're buying it in a different way.

00:26:02.599 --> 00:26:04.359
And I think that that will be true for a lot

00:26:04.359 --> 00:26:06.619
of these things. They may not be consuming items

00:26:06.619 --> 00:26:09.099
in the same way. The items may be produced differently.

00:26:09.700 --> 00:26:12.539
Some of them may be, in fact, produced by AI.

00:26:13.019 --> 00:26:15.740
But then, you know, let's say the photographer,

00:26:15.940 --> 00:26:20.019
for example, that's going to free that photographer

00:26:20.019 --> 00:26:23.819
up to do other probably more creative work than

00:26:23.819 --> 00:26:25.839
what they were doing. And I think that's a net

00:26:25.839 --> 00:26:29.880
positive. So someone who is just taking pictures

00:26:29.880 --> 00:26:32.579
of oranges, and now you don't need to do that

00:26:32.579 --> 00:26:35.400
because if you want to advertise oranges, you

00:26:35.400 --> 00:26:38.000
just have AI generate a picture of an orange.

00:26:38.460 --> 00:26:41.220
Well, those people may be out of a job taking

00:26:41.220 --> 00:26:44.380
those pictures, but there's a whole range of

00:26:44.380 --> 00:26:47.660
creative opportunities that they can get into

00:26:47.660 --> 00:26:50.200
that AI is not going to immediately replace.

00:26:50.599 --> 00:26:53.440
And I think that's going to be much more the

00:26:53.440 --> 00:26:57.839
pattern. of what you see with AI than everybody

00:26:57.839 --> 00:27:03.200
being thrown out of work. But with it increasing

00:27:03.200 --> 00:27:06.039
efficiency, like you said, you're able to do

00:27:06.039 --> 00:27:09.019
work that would normally take longer, maybe days

00:27:09.019 --> 00:27:12.519
at a time. You can pack it into two hours. Oh,

00:27:12.559 --> 00:27:17.160
absolutely. Do you think that employers or just

00:27:17.160 --> 00:27:21.799
industries in general will require more productivity?

00:27:23.230 --> 00:27:25.329
I think they will. And I think that's happening

00:27:25.329 --> 00:27:29.569
already. But productivity gains because of technology

00:27:29.569 --> 00:27:31.950
are not anything new. I mean, if you think about,

00:27:32.089 --> 00:27:36.190
again, back to those people with the hand calculators

00:27:36.190 --> 00:27:40.970
in the back room of the bank grinding away, once

00:27:40.970 --> 00:27:44.670
you got computers in the picture there, the productivity

00:27:44.670 --> 00:27:50.269
per individual went up amazingly high because

00:27:50.269 --> 00:27:53.519
you didn't have to have the bottleneck of an

00:27:53.519 --> 00:27:56.220
individual grinding through a column of numbers

00:27:56.220 --> 00:28:01.400
for a period of time. So I think that the question

00:28:01.400 --> 00:28:04.859
is going to be sort of where do those productivity

00:28:04.859 --> 00:28:10.400
gains get, who benefits from them? So if you

00:28:10.400 --> 00:28:15.920
get a person who suddenly is 10x more productive,

00:28:16.579 --> 00:28:19.539
is that going to be reflected in their paycheck?

00:28:19.700 --> 00:28:21.980
Is that going to be reflected in some other aspect

00:28:21.980 --> 00:28:24.819
of their compensation? Or are they just going

00:28:24.819 --> 00:28:27.660
to be expected to continue working as hard as

00:28:27.660 --> 00:28:30.460
they've been doing at the same salary? And that's

00:28:30.460 --> 00:28:32.420
a discussion I think that we need to have because

00:28:32.420 --> 00:28:34.779
I think there is something to be said for there

00:28:34.779 --> 00:28:41.200
being a sort of technology dividend. So in other

00:28:41.200 --> 00:28:45.940
words, that all of us should benefit somehow.

00:28:46.250 --> 00:28:49.769
from this new technology, as opposed to just

00:28:49.769 --> 00:28:52.430
the people at the very top running the companies.

00:28:52.990 --> 00:28:55.490
And what that benefit should be, I don't know.

00:28:56.109 --> 00:28:59.049
But there should be some way. And this was something

00:28:59.049 --> 00:29:01.069
that was very big in the 50s that unfortunately

00:29:01.069 --> 00:29:04.230
never really came to pass. I mean, in the 50s,

00:29:04.329 --> 00:29:06.529
automation, we were all going to be pressing

00:29:06.529 --> 00:29:08.869
a button and cooking dinner. And so we would

00:29:08.869 --> 00:29:12.089
all be out playing tennis or golf or swimming.

00:29:12.089 --> 00:29:14.589
And then, you know, we come in from our two hour

00:29:14.589 --> 00:29:18.250
work week, like in the Jetsons. And that never

00:29:18.250 --> 00:29:21.390
really came to pass for a variety of reasons.

00:29:21.470 --> 00:29:26.150
But I think that's much more that's less a technology

00:29:26.150 --> 00:29:29.750
issue and more a social issue. So that's more

00:29:29.750 --> 00:29:34.059
about how do we want to choose? to allocate those

00:29:34.059 --> 00:29:36.920
resources that we've subtly freed up by having

00:29:36.920 --> 00:29:40.460
all of this work done by machines. And, you know,

00:29:40.519 --> 00:29:42.720
for example, right now, I mean, a big question

00:29:42.720 --> 00:29:46.940
is happening with Amazon drivers. They're considered

00:29:46.940 --> 00:29:51.640
to be basically not full time employees, but

00:29:51.640 --> 00:29:56.309
contractors. And there's a foot to. Make it so

00:29:56.309 --> 00:29:58.069
that they're treated as full -time employees

00:29:58.069 --> 00:30:00.309
of Amazon just like anyone else working Amazon

00:30:00.309 --> 00:30:04.470
which would increase their benefits so that's

00:30:04.470 --> 00:30:06.309
the kind of and you know, of course Amazon's

00:30:06.309 --> 00:30:07.750
against that they don't want to pay the money

00:30:07.750 --> 00:30:12.250
but That's the kind of tension. I think that's

00:30:12.250 --> 00:30:14.750
gonna have to be resolved and we're gonna kind

00:30:14.750 --> 00:30:18.490
of have to have a National coming -together moment

00:30:18.490 --> 00:30:23.309
to figure out how to equitably divide up this

00:30:23.309 --> 00:30:25.549
dividend so it's not just all going to a bunch

00:30:25.549 --> 00:30:28.670
of people at the top. I definitely agree with

00:30:28.670 --> 00:30:30.650
that because I think a lot of people at the top

00:30:30.650 --> 00:30:33.470
didn't get enough anyway. So why not share the

00:30:33.470 --> 00:30:37.049
wealth? Yeah, they're getting a lot already and

00:30:37.049 --> 00:30:39.650
the problem is that the system is kind of set

00:30:39.650 --> 00:30:46.309
up to allow them to continue that. process of

00:30:46.309 --> 00:30:49.470
getting a lot and getting more. And I don't know

00:30:49.470 --> 00:30:52.809
what kind of downward pressure is going to come

00:30:52.809 --> 00:30:57.529
about to make it so that there's more of a general

00:30:57.529 --> 00:31:00.529
distribution. When I'm not talking about a redistribution

00:31:00.529 --> 00:31:02.930
of wealth, I'm just talking about if a company

00:31:02.930 --> 00:31:06.690
is able to perform 10X better, then that should,

00:31:06.690 --> 00:31:08.930
I think, be reflected. Now, maybe you make the

00:31:08.930 --> 00:31:10.690
employees stockholders or something like that.

00:31:10.690 --> 00:31:12.650
And so then the value of their stock goes up.

00:31:13.369 --> 00:31:16.890
I don't know what the solution is, but I think

00:31:16.890 --> 00:31:20.549
that there definitely needs to be a way to make

00:31:20.549 --> 00:31:25.650
sure that this sudden boost in productive capability

00:31:25.650 --> 00:31:29.490
is spread more evenly than it seems to be right

00:31:29.490 --> 00:31:33.829
now. To me, that's just basic fairness. Yeah,

00:31:33.829 --> 00:31:38.049
which I definitely agree with it. I have a question.

00:31:38.309 --> 00:31:43.279
I was looking at a different interview and Yeah,

00:31:43.279 --> 00:31:45.700
we're talking about, I believe, like the European

00:31:45.700 --> 00:31:50.859
Union in 2027 is going to form compliance with

00:31:50.859 --> 00:31:56.359
AI. Yes. Yes. My question is, well, it could

00:31:56.359 --> 00:31:59.160
be a two part question. One, in doing something

00:31:59.160 --> 00:32:03.720
like this, I don't know if they're doing this,

00:32:03.920 --> 00:32:07.680
are they looking to like experts like you to

00:32:07.680 --> 00:32:12.119
actually help draw the diplomacy of compliance?

00:32:12.480 --> 00:32:16.519
And then also, who gets to set the rules for

00:32:16.519 --> 00:32:20.140
compliance? Because if people are using AI, like

00:32:20.140 --> 00:32:23.480
you say, in just generality, the wrong way, or

00:32:23.480 --> 00:32:27.579
don't understand how sophisticated it is, how

00:32:27.579 --> 00:32:36.759
can a body of people decide its complicity? Right.

00:32:37.259 --> 00:32:40.859
Exactly. Well, The EU is actually already getting

00:32:40.859 --> 00:32:43.940
pretty involved in AI much more so than the United

00:32:43.940 --> 00:32:47.299
States. For one thing, they now require watermarking.

00:32:47.960 --> 00:32:51.000
So anytime you generate something using AI, there's

00:32:51.000 --> 00:32:54.140
sort of a hidden code in the way the words are

00:32:54.140 --> 00:32:56.740
arranged that doesn't affect the output, but

00:32:56.740 --> 00:32:59.000
that is supposed to be, at least supposedly,

00:32:59.720 --> 00:33:02.700
detectable as having come from AI. And that's

00:33:02.700 --> 00:33:05.279
something that comes from the EU. Now, what's

00:33:05.279 --> 00:33:09.509
coming up in 2027? Really, no one really knows

00:33:09.509 --> 00:33:13.029
what's going to happen. This is a group of politicians

00:33:13.029 --> 00:33:15.769
and probably some European technologists, but

00:33:15.769 --> 00:33:20.230
AI does not really have the strong presence in

00:33:20.230 --> 00:33:22.609
Europe that it does in the United States. So

00:33:22.609 --> 00:33:25.170
I'm not sure where they're getting their technologists

00:33:25.170 --> 00:33:29.240
from. But right now, a lot of what's happening

00:33:29.240 --> 00:33:32.140
is that a lot of companies kind of know that

00:33:32.140 --> 00:33:33.880
sometime next year, they're going to have to

00:33:33.880 --> 00:33:37.259
start complying if they want to work in the EU.

00:33:37.940 --> 00:33:41.380
But everything is really up in the air as far

00:33:41.380 --> 00:33:43.960
as what exactly that is going to mean. I mean,

00:33:44.039 --> 00:33:47.140
is it going to mean having a third party that

00:33:47.140 --> 00:33:50.220
goes through a checklist of things to make sure

00:33:50.220 --> 00:33:53.180
that the AI that you're using has proper guardrails

00:33:53.180 --> 00:33:55.940
and a safe and that they actually certify that

00:33:55.940 --> 00:33:58.180
and you know they sign their names to it kind

00:33:58.180 --> 00:34:01.000
of like a banking examiner or is it going to

00:34:01.000 --> 00:34:04.759
mean something different and at this point we

00:34:04.759 --> 00:34:07.380
really don't know I wouldn't be hugely surprised

00:34:07.380 --> 00:34:10.179
if it doesn't get kicked down the road you know

00:34:10.179 --> 00:34:14.690
to 2028 or 2029 because It's such a complicated

00:34:14.690 --> 00:34:18.550
issue and there's so many stakeholders involved

00:34:18.550 --> 00:34:23.090
that it's kind of a little hard to imagine that

00:34:23.090 --> 00:34:27.530
it's going to become completely resolved within

00:34:27.530 --> 00:34:29.909
the next year. And you know, in this country

00:34:29.909 --> 00:34:32.269
as well, there's a lot of calls and I don't know

00:34:32.269 --> 00:34:33.789
how successful these are going to be. There are

00:34:33.789 --> 00:34:38.809
a lot of calls to put moratoriums on both the

00:34:38.809 --> 00:34:41.670
building out of new data centers because those

00:34:41.670 --> 00:34:43.670
are obviously very controversial. They use a

00:34:43.670 --> 00:34:47.630
lot of power. They use a lot of water. And also

00:34:47.630 --> 00:34:51.929
just on AI development in general. And I'm not

00:34:51.929 --> 00:34:56.010
sure how a national moratorium on AI development

00:34:56.010 --> 00:34:58.769
would even work because you can't stop individual

00:34:58.769 --> 00:35:01.829
researchers from continuing to do their work.

00:35:02.489 --> 00:35:04.769
I don't know what you would have inspectors go

00:35:04.769 --> 00:35:08.170
into open AI labs or Claude's labs and make sure

00:35:08.170 --> 00:35:10.940
they're not doing any they're not up to anything.

00:35:11.400 --> 00:35:14.420
And then you have China. And there's no reason

00:35:14.420 --> 00:35:17.599
that China is going to sign on to that kind of

00:35:17.599 --> 00:35:20.840
an agreement as well. Because they're doing really

00:35:20.840 --> 00:35:23.500
well with, you know, they're definitely in the

00:35:23.500 --> 00:35:25.500
running in the AI race, and they're producing

00:35:25.500 --> 00:35:29.679
a lot of very valuable, open models, which a

00:35:29.679 --> 00:35:34.480
lot of American companies aren't. And so I think

00:35:34.480 --> 00:35:37.480
the whole field right now, it's really very much

00:35:37.849 --> 00:35:41.590
a wild west. We don't really know where all of

00:35:41.590 --> 00:35:44.329
the dust is going to settle. It's kind of like

00:35:44.329 --> 00:35:48.489
the very early days of the web and the internet

00:35:48.489 --> 00:35:53.250
when the idea of using the web to advertise was

00:35:53.250 --> 00:35:56.289
seen as just an awful, awful idea. And the first

00:35:56.289 --> 00:35:59.030
person that did it got a lot of criticism. And

00:35:59.030 --> 00:36:02.030
of course now, I mean, that seems ridiculous.

00:36:02.329 --> 00:36:04.050
Yes, certain engines were controversial because

00:36:04.050 --> 00:36:07.289
of privacy concerns. All of these things, looking

00:36:07.289 --> 00:36:08.789
back on them, and I think we're going to find

00:36:08.789 --> 00:36:11.909
that a lot of probably it's going to be more

00:36:11.909 --> 00:36:13.929
laissez -faire and a lot of the things that we

00:36:13.929 --> 00:36:16.929
worry about now, we'll look back at and say,

00:36:16.969 --> 00:36:19.929
God, what were we even thinking about? Because

00:36:19.929 --> 00:36:23.250
look at how far we've come just in terms of like

00:36:23.250 --> 00:36:27.610
social media didn't exist a little over 20 years

00:36:27.610 --> 00:36:34.730
ago. And it's become so fully integrated into

00:36:34.920 --> 00:36:38.880
our lives. And there are certainly very strong

00:36:38.880 --> 00:36:41.519
arguments to be made for regulating it, but that

00:36:41.519 --> 00:36:43.480
never really came to pass. And now it's kind

00:36:43.480 --> 00:36:46.239
of too late. I mean, there's really no clear

00:36:46.239 --> 00:36:50.400
set of steps that you could put into play to

00:36:50.400 --> 00:36:53.739
regulate social media. So we're kind of for good

00:36:53.739 --> 00:36:56.920
or bad stuck with it the way it is. So I think

00:36:56.920 --> 00:37:00.690
it's probably going to be less something planned

00:37:00.690 --> 00:37:03.190
out and more sort of a whole bunch of different

00:37:03.190 --> 00:37:06.889
little ad hoc things come together. And we figure

00:37:06.889 --> 00:37:11.170
out some sort of compromise, but probably no

00:37:11.170 --> 00:37:13.070
one is going to be happy because no one ever

00:37:13.070 --> 00:37:18.630
is happy with a good compromise. This is amazing.

00:37:19.989 --> 00:37:23.829
So I want you to explain a little bit of the

00:37:23.829 --> 00:37:27.719
complexity. of AI because most people just think

00:37:27.719 --> 00:37:30.380
of it like as a computer. You go and put information

00:37:30.380 --> 00:37:34.360
in and it researches and spits out information.

00:37:35.099 --> 00:37:38.159
And one thing I came across, I believe you was

00:37:38.159 --> 00:37:41.860
given an example of I think a 1958 experiment

00:37:41.860 --> 00:37:45.059
where they were talking to kids and made up a

00:37:45.059 --> 00:37:49.000
word. And then that particular word, because

00:37:49.000 --> 00:37:51.519
they said there were two of them, the kids naturally

00:37:51.519 --> 00:37:55.889
pluralized it. the language in how that relates

00:37:55.889 --> 00:38:00.690
to AI in the way that it arrives at its decision.

00:38:01.949 --> 00:38:08.150
Well, so that's out of a sort of area of more

00:38:08.150 --> 00:38:11.409
psycholinguistic, psychological, cognitive psychology

00:38:11.409 --> 00:38:14.550
research. And that basically is an area that

00:38:14.550 --> 00:38:16.869
I'm interested in, which has to do with trying

00:38:16.869 --> 00:38:21.539
to figure out to what extent these computer models

00:38:21.539 --> 00:38:25.199
are similar to humans when they use language.

00:38:25.340 --> 00:38:27.800
And there's a lot of ways in which they're completely

00:38:27.800 --> 00:38:31.380
not like humans at all. And I think that's an

00:38:31.380 --> 00:38:33.179
important thing to keep in mind because people

00:38:33.179 --> 00:38:35.840
think that they're alive, they're human, they're

00:38:35.840 --> 00:38:40.000
talking to them. But if you break down what they're

00:38:40.000 --> 00:38:44.260
doing, in some ways, it's incredibly simple compared

00:38:44.260 --> 00:38:48.099
to how humans process information and process

00:38:48.099 --> 00:38:51.030
language. point that I always think is just fascinating

00:38:51.030 --> 00:38:55.449
is when you look at the amount of energy that

00:38:55.449 --> 00:38:59.690
a human brain needs, it's like a light bulb or

00:38:59.690 --> 00:39:03.710
something tiny like that, compared to the huge

00:39:03.710 --> 00:39:07.150
amounts of energy that one of these queries requires

00:39:07.150 --> 00:39:10.590
in order to process. So right there, that tells

00:39:10.590 --> 00:39:14.190
us that something's not the same about these

00:39:14.190 --> 00:39:17.469
two processes, because whatever we're doing can

00:39:17.469 --> 00:39:22.489
be done with much less energy consumption than

00:39:22.489 --> 00:39:26.130
the method that we're using now for algorithms

00:39:26.130 --> 00:39:29.889
in the computer. To your point about the wugs,

00:39:30.369 --> 00:39:32.809
what you were talking about is that way back

00:39:32.809 --> 00:39:36.909
in the day, they wanted to see how children acquired

00:39:36.909 --> 00:39:42.889
rules like the plural is a dash S. You would

00:39:42.889 --> 00:39:46.510
give someone, a kid, a new like a made -up word,

00:39:46.639 --> 00:39:51.559
a wug and you would say, here's two of them,

00:39:51.599 --> 00:39:53.360
what do you have? And they would say two wugs.

00:39:53.920 --> 00:39:57.679
And the idea of that was that they had generalized

00:39:57.679 --> 00:40:04.039
a rule that to make a plural, you add an S to

00:40:04.039 --> 00:40:08.119
a word. And the point is that that rule applied

00:40:08.119 --> 00:40:11.599
to things that you had never seen before. And

00:40:11.599 --> 00:40:15.860
that kind of is the basis for a lot of linguistic

00:40:15.860 --> 00:40:19.119
theory because you obviously have to have some

00:40:19.119 --> 00:40:23.079
system that lets you introduce new words into

00:40:23.079 --> 00:40:25.739
the child's vocabulary and not have them be completely

00:40:25.739 --> 00:40:29.639
confused. And the same was done, there was a

00:40:29.639 --> 00:40:32.579
lot of work done with the past tense as well.

00:40:33.000 --> 00:40:34.739
And what you see in children is very interesting.

00:40:34.800 --> 00:40:37.940
What you see in children is that they learn the

00:40:37.940 --> 00:40:45.079
irregular past tense like went quite But then

00:40:45.079 --> 00:40:49.619
they use the, when they learn the dash ED for

00:40:49.619 --> 00:40:51.960
the regular, sometimes they over -correct and

00:40:51.960 --> 00:40:55.440
they start saying goat. And then they go and

00:40:55.440 --> 00:40:59.260
say, went again. And I actually, many people

00:40:59.260 --> 00:41:01.219
have run these experiments and I've run them

00:41:01.219 --> 00:41:07.559
myself. I wanted to see, did the machine show

00:41:07.559 --> 00:41:11.260
the same kind of rule -based learning? that humans

00:41:11.260 --> 00:41:14.880
seem to have. And it doesn't. It actually does

00:41:14.880 --> 00:41:18.500
not show the same pattern at all. And what happens

00:41:18.500 --> 00:41:25.460
is that it kind of learns that a dash ed is what

00:41:25.460 --> 00:41:28.260
you put at the end of a past tense when you have

00:41:28.260 --> 00:41:30.559
a new word. But it doesn't always learn that

00:41:30.559 --> 00:41:34.179
the way children do. And a lot of it has to do

00:41:34.179 --> 00:41:38.449
with just how frequent the input is. to the system.

00:41:38.750 --> 00:41:41.530
So it has lots of goes and whens and whens and

00:41:41.530 --> 00:41:45.969
goes, then it learns that very well. And whether

00:41:45.969 --> 00:41:50.030
or not it's really learning a rule, or it's just

00:41:50.030 --> 00:41:53.730
learning because it heard this before, and it's

00:41:53.730 --> 00:41:55.650
general, it's not generalizing, but it's just

00:41:55.650 --> 00:41:58.590
using its own memory. So I think that's probably

00:41:58.590 --> 00:42:00.610
some of the experiments that you're talking about.

00:42:00.630 --> 00:42:04.750
And they're pretty central to the whole question

00:42:04.750 --> 00:42:07.809
of how how similar these systems are to human

00:42:07.809 --> 00:42:11.849
language processing, and the degree to which

00:42:11.849 --> 00:42:13.969
they can tell us something about human language

00:42:13.969 --> 00:42:21.429
processing. Love it. You have a podcast, which

00:42:21.429 --> 00:42:24.690
you do, when you're talking about the AI, the

00:42:24.690 --> 00:42:29.550
black box, if I'm not mistaken? Inside the black

00:42:29.550 --> 00:42:32.050
box. Inside the black box. Yeah, I have a substack,

00:42:32.050 --> 00:42:34.659
and I also just sort of put it into a podcast

00:42:34.659 --> 00:42:37.820
form as well. That's right. What made you want

00:42:37.820 --> 00:42:41.260
to use that method of delivery for the information

00:42:41.260 --> 00:42:44.559
until anybody who wants to learn more about it?

00:42:44.559 --> 00:42:47.480
What would they get out of it or they can expect

00:42:47.480 --> 00:42:51.039
to learn? Well, I've talked to people that read

00:42:51.039 --> 00:42:53.820
this or that listen to the podcast, and I think

00:42:53.820 --> 00:42:58.079
that there are a lot of people out there who

00:42:58.079 --> 00:43:02.780
are experts in their own fields. So doctors,

00:43:02.940 --> 00:43:06.360
lawyers, engineers, people like that and who

00:43:06.360 --> 00:43:11.239
are fascinated by AI but don't really understand

00:43:11.239 --> 00:43:14.300
much about the nuts and bolts of how it works.

00:43:14.900 --> 00:43:17.159
And so what I wanted to do and what I try to

00:43:17.159 --> 00:43:21.139
do is to pick a topic every week that's kind

00:43:21.139 --> 00:43:25.380
of a good explanation of some little aspect.

00:43:25.420 --> 00:43:27.679
So we were talking about the past tense that

00:43:27.679 --> 00:43:30.699
last week I posted some experiments that I had

00:43:30.699 --> 00:43:34.710
done. looking at learning of the past tense in

00:43:34.710 --> 00:43:37.429
AI systems and how that compares and contrasts

00:43:37.429 --> 00:43:40.110
what you see with kids and what that can tell

00:43:40.110 --> 00:43:42.849
you about what's going on inside those systems.

00:43:43.710 --> 00:43:47.750
So I think that in the end, people are really

00:43:47.750 --> 00:43:51.050
fascinated by these systems. But what surprises

00:43:51.050 --> 00:43:53.650
them and what I kind of focus on and inside the

00:43:53.650 --> 00:43:58.530
black box on Substack and the podcast is that

00:43:58.679 --> 00:44:02.760
even the people that created them do not fully

00:44:02.760 --> 00:44:05.820
understand how it is that they arrive at the

00:44:05.820 --> 00:44:09.360
solutions that they arrive at. And there's a

00:44:09.360 --> 00:44:12.340
whole field called mechanistic interpretability

00:44:12.340 --> 00:44:16.460
that aims to understand what's at, because, and

00:44:16.460 --> 00:44:18.719
the reason for that is simply that these systems

00:44:18.719 --> 00:44:22.300
have literally billions and billions of parameters.

00:44:22.460 --> 00:44:27.269
I mean, they have huge numbers of numbers, really

00:44:27.269 --> 00:44:31.230
large vectors being processed by these chips,

00:44:31.650 --> 00:44:37.090
by these GPUs. And so when you have a solution

00:44:37.090 --> 00:44:41.630
space that's that large, it's very difficult,

00:44:41.829 --> 00:44:44.250
unless everything is crisp and rule -based, which

00:44:44.250 --> 00:44:47.090
we don't think it is, it's very difficult to

00:44:47.090 --> 00:44:50.389
figure out exactly how it's learning because

00:44:50.389 --> 00:44:54.469
what these systems do is you give them the answer

00:44:54.760 --> 00:44:57.440
And then you see if they come up with the answer.

00:44:57.519 --> 00:44:59.440
And then if they don't, you adjust them a little

00:44:59.440 --> 00:45:03.119
bit and you just keep doing that repeatedly for

00:45:03.119 --> 00:45:06.199
literally trillions, sometimes trillions of iterations

00:45:06.199 --> 00:45:09.260
until when you put something in at the other

00:45:09.260 --> 00:45:12.420
end, you get something out and it makes sense.

00:45:12.780 --> 00:45:15.079
But what's going on in that in between with all

00:45:15.079 --> 00:45:18.739
those numbers is extremely complicated and very

00:45:18.739 --> 00:45:26.030
difficult to really fathom, I think. And so what

00:45:26.030 --> 00:45:28.969
I do in Inside the Black Box is try to talk a

00:45:28.969 --> 00:45:31.309
little bit about, you know, different topics

00:45:31.309 --> 00:45:34.289
in there and maybe make it a little bit less

00:45:34.289 --> 00:45:37.730
mysterious for people and a little less scary.

00:45:38.889 --> 00:45:44.250
I love it. So for everybody out there, they want

00:45:44.250 --> 00:45:47.909
to learn more. Where would you like them to go?

00:45:48.210 --> 00:45:53.079
And then also What's the number one misnomer

00:45:53.079 --> 00:45:58.719
that people have about AI? Well, so you can see

00:45:58.719 --> 00:46:02.400
the substack is inside the black box .ai. I also

00:46:02.400 --> 00:46:06.280
have an app called your voicecraft .ai, which

00:46:06.280 --> 00:46:08.280
you can go to and get a beta for and that will

00:46:08.280 --> 00:46:13.539
allow you to basically train a local model like

00:46:13.539 --> 00:46:15.500
what we were talking about to speak in your own

00:46:15.500 --> 00:46:21.460
voice and your written voice. As far as the things

00:46:21.460 --> 00:46:25.920
that I see people wrong about, I guess you could

00:46:25.920 --> 00:46:28.860
say about AI, I think it's not something to be

00:46:28.860 --> 00:46:31.840
scared of. There's a lot of very anti -AI sentiment,

00:46:32.719 --> 00:46:36.599
and I live in a kind of bubble where everyone

00:46:36.599 --> 00:46:38.380
thinks AI is great, and I have to keep reminding

00:46:38.380 --> 00:46:40.000
myself that there are a lot of people out there

00:46:40.000 --> 00:46:42.920
that don't think that AI is great. They're afraid

00:46:42.920 --> 00:46:45.079
it'll take away their jobs, which is an issue

00:46:45.079 --> 00:46:48.590
we've already discussed. They're afraid that

00:46:48.590 --> 00:46:51.690
it will somehow take over the world, which is

00:46:51.690 --> 00:46:58.309
pretty unlikely. And there are people who think,

00:46:58.309 --> 00:46:59.829
I think, that they're actually, in some sense,

00:47:00.010 --> 00:47:03.510
alive. And it's an actual entity that you're

00:47:03.510 --> 00:47:08.090
talking to. And that's just not the case. You're

00:47:08.090 --> 00:47:10.610
dealing with a computer. It's all ones and zeros.

00:47:10.690 --> 00:47:13.469
You're putting information into a huge register

00:47:13.469 --> 00:47:16.400
or stack. of computing elements, and you're getting

00:47:16.400 --> 00:47:19.699
something out. That's nothing like what the human

00:47:19.699 --> 00:47:22.360
brain does. It's much more complicated. Neurons

00:47:22.360 --> 00:47:26.000
are just these completely mysterious computational

00:47:26.000 --> 00:47:29.219
elements that do a lot more than what a computer

00:47:29.219 --> 00:47:32.000
does, and with a lot less power, as we've discussed.

00:47:32.480 --> 00:47:34.940
So I would say, my takeaways are, don't be scared.

00:47:35.119 --> 00:47:39.139
Learn it. Actually get to know the strengths

00:47:39.139 --> 00:47:42.670
and weaknesses of it. when you go in and start

00:47:42.670 --> 00:47:45.130
playing with it and trying to get responses from

00:47:45.130 --> 00:47:47.969
it, you begin to get a sort of sense of its personality.

00:47:48.110 --> 00:47:49.929
I don't like to say personality, but I'll say

00:47:49.929 --> 00:47:53.389
personality because it's not a human, but it's

00:47:53.389 --> 00:47:56.849
sort of personality and the sorts of ways that

00:47:56.849 --> 00:47:59.789
it responds to the things that you say. And I

00:47:59.789 --> 00:48:03.090
think that helps to demystify it and make it

00:48:03.090 --> 00:48:06.349
a lot less scary for a lot of people. So that

00:48:06.349 --> 00:48:11.639
would be my take home. But Dr. Blackwell, I appreciate

00:48:11.639 --> 00:48:17.639
your time and you giving me inside perspective

00:48:17.639 --> 00:48:21.000
to this, which I've learned some more things

00:48:21.000 --> 00:48:25.059
today and just even how to just look at AI. And

00:48:25.059 --> 00:48:28.800
I'm glad that I've been prompting like pretty

00:48:28.800 --> 00:48:30.820
correctly. Yeah, I like your prompt. The way

00:48:30.820 --> 00:48:33.699
you've been prompting is very good. I mean, that's

00:48:33.699 --> 00:48:37.829
what I do. So I'm on the right path. I like that.

00:48:40.190 --> 00:48:42.590
Absolutely. Everybody out there, y 'all can go

00:48:42.590 --> 00:48:45.369
check out Inside the Black Box podcast. I have

00:48:45.369 --> 00:48:48.190
it all description and linked in the show. And

00:48:48.190 --> 00:48:50.829
then also you said you're on Substack as well.

00:48:51.750 --> 00:48:54.469
That the Inside the Black Box .ai is Substack.

00:48:54.530 --> 00:48:57.190
But if you do Inside the Black Box, you'll also

00:48:57.190 --> 00:48:59.409
if you do a podcast search, you'll find it as

00:48:59.409 --> 00:49:03.329
a podcast as well on Apple Spotify. and y 'all

00:49:03.329 --> 00:49:05.409
can go and check it out. If you want to learn

00:49:05.409 --> 00:49:08.909
more, you got questions, think of nature, get

00:49:08.909 --> 00:49:11.690
up Dr. Blackwell, I'm sure he would respond whenever

00:49:11.690 --> 00:49:14.570
he has free time. Yeah, absolutely. Substack

00:49:14.570 --> 00:49:17.429
actually has a mailing system, so you can DM

00:49:17.429 --> 00:49:19.869
me if you want. That's what's up. So you got

00:49:19.869 --> 00:49:22.610
a system in place for everything. Yeah, that's

00:49:22.610 --> 00:49:25.710
right. I try to. I love it. You're going to learn

00:49:25.710 --> 00:49:28.440
how to be more efficient out here using AI. I'm

00:49:28.440 --> 00:49:30.619
your host, Andreas. But think that radio show

00:49:30.619 --> 00:49:33.619
has been music and business talk in covering,

00:49:33.619 --> 00:49:36.679
you know, AI. When I started my first independent

00:49:36.679 --> 00:49:38.739
record label, I was folding shirts at Old Navy,

00:49:38.920 --> 00:49:41.099
slipping handmade compilation CDs, straight into

00:49:41.099 --> 00:49:43.500
customer shopping bags just to secure my first

00:49:43.500 --> 00:49:46.019
sales. I had zero budget, but I had an infinite

00:49:46.019 --> 00:49:49.579
drive. But today I see so many solopreneurs running

00:49:49.579 --> 00:49:52.280
on social media, treadmills, posting content

00:49:52.280 --> 00:49:55.320
for free and praying that algorithm smiles on

00:49:55.320 --> 00:49:57.969
them. That's not a business plan. That's hope

00:49:57.969 --> 00:49:59.969
marketing. If you don't pay to advertise and

00:49:59.969 --> 00:50:02.409
build search engine infrastructure, you are paying

00:50:02.409 --> 00:50:05.550
with your manual labor hours instead. Stop building

00:50:05.550 --> 00:50:08.409
on rented land. I put my entire 10 -year field

00:50:08.409 --> 00:50:11.710
manual on pay hip called I Can Do Bad All By

00:50:11.710 --> 00:50:15.210
Myself to show you exactly how to build a sovereign

00:50:15.210 --> 00:50:17.909
empire you own outright. Comment the word link

00:50:17.909 --> 00:50:20.070
below and I will instantly DM you the link to

00:50:20.070 --> 00:50:21.769
grab your copy for $19 .99.