1
00:00:00,000 --> 00:00:03,760
Could you see it growing to a point where someone could essentially come in and just say,

2
00:00:03,760 --> 00:00:30,360
Hey, I have these ingredients. Go to it, AI. And I want lunch.

3
00:00:30,360 --> 00:00:35,440
Ladies and gentlemen, boys and girls, children of all ages, dogs, cats, robots, and everybody in between, especially you.

4
00:00:35,920 --> 00:00:41,960
Burgeoning AI culinarians, this is HTTTA, how to talk to AI.

5
00:00:42,360 --> 00:00:44,560
I'm your host, Wes the SynthBind, SynthBind West.

6
00:00:45,040 --> 00:00:49,120
And as always, wait, I'm just getting, I'm getting a report in from the skies above.

7
00:00:49,400 --> 00:00:50,520
That's right, folks.

8
00:00:50,920 --> 00:00:51,960
Gota's not here this week.

9
00:00:51,960 --> 00:00:58,200
She's on vacation, but we have a special aerial guest, our eye in the sky, the queen of the zeppelins herself.

10
00:00:58,200 --> 00:01:00,200
Gota the gondolier.

11
00:01:02,200 --> 00:01:04,000
Hi, Wes. Hi, Keegan.

12
00:01:04,000 --> 00:01:09,480
I'm sorry I missed the podcast, but as you can see, I'm a bit flying in the air.

13
00:01:09,480 --> 00:01:12,840
I can't wait to hear your discussion and your cooking experience.

14
00:01:14,040 --> 00:01:22,880
Today, we're thrilled to have Keegan Dargi join us, the CTO and partner at Make More Creative, a family run creative intelligence agency,

15
00:01:23,080 --> 00:01:25,960
specializing in end to end brand development and deployment.

16
00:01:25,960 --> 00:01:30,520
They help their clients shape and share their stories and roll them out to the world.

17
00:01:30,840 --> 00:01:33,760
But they also create fun stuff like the kitchen AI.

18
00:01:34,040 --> 00:01:41,080
Keegan is a self-taught Python and Java developer and was quickly drawn to the world of artificial intelligence.

19
00:01:41,320 --> 00:01:48,840
As the fascination with it began, closed beta back in the day, playing around with the GPT-3 models as early as 2020.

20
00:01:49,000 --> 00:01:54,760
He took his interest in AI to new heights, diving into IBM's Watson natural language understanding.

21
00:01:54,760 --> 00:02:00,040
And created his first GPT-3 based Discord chat bot, of which he has a litany of now.

22
00:02:00,040 --> 00:02:02,040
I have seen them myself.

23
00:02:02,040 --> 00:02:10,760
Since then, Keegan has developed an array of AI powered personal tools capable of generating everything from sales scripts to social media content.

24
00:02:10,760 --> 00:02:21,640
And because of this, spotting a unique opportunity, he combined AI with personal nutrition, recipe generation and content creation by developing the kitchen AI.

25
00:02:21,640 --> 00:02:28,120
This innovative platform, leveraging large language models and some clever prompt engineering,

26
00:02:28,120 --> 00:02:38,920
intelligently customizes recipes for specific dietary needs, provides step by step instructions, social media copy and the AI generated image of the dish.

27
00:02:38,920 --> 00:02:47,240
Keegan's work is a testament to the practical use of technology in solving real world problems and a real world application of generative AI.

28
00:02:47,240 --> 00:02:52,680
And the incredible synthesis that happens when humans and AI collaborate together.

29
00:02:52,680 --> 00:02:58,440
Ladies and gentlemen, the kitchen AI's executive chef himself, Keegan Dargi.

30
00:02:58,440 --> 00:03:01,640
Thank you for having me, Wes. This is awesome.

31
00:03:01,640 --> 00:03:09,320
Great. I'm so glad you're here. And I'll be the first one to say as a culinary in past, present and hopefully future.

32
00:03:09,320 --> 00:03:17,240
I was floored and just so taken aback by everything you're doing with the kitchen AI. I find it so fascinating.

33
00:03:17,240 --> 00:03:20,520
And we're going to be putting a little bit of that to the test later on in the episode.

34
00:03:20,520 --> 00:03:24,120
But so let's rewind the tape a little bit.

35
00:03:24,120 --> 00:03:24,920
Yeah.

36
00:03:24,920 --> 00:03:30,440
What was your first kind of initial inspiration to build the kitchen AI?

37
00:03:30,440 --> 00:03:37,560
Yeah, that's a great question. So it all kind of started like I've been playing around with AI for a few years now.

38
00:03:37,560 --> 00:03:44,840
Generative AI. I did some like machine learning here and there, but I found a lot more fun working with the large language models.

39
00:03:44,840 --> 00:03:50,760
And as soon as chat GPT came out, I realized that AI was just happening like now.

40
00:03:50,760 --> 00:03:59,480
And one of the things I needed to focus on was to really dive in to artificial intelligence, generative AI and just start building some stuff.

41
00:03:59,480 --> 00:04:07,960
And a project that I had already started working on was doing sort of like custom recipes for my family.

42
00:04:07,960 --> 00:04:14,680
Because, you know, my family, as we're getting older, we go through dietary changes, we have like health conditions that pop up.

43
00:04:14,680 --> 00:04:22,760
And one of the hardest things that we found was that, you know, how do we change our food intake, like what we're eating?

44
00:04:22,760 --> 00:04:28,840
How do we change that to match these new health conditions or dietary needs or whatever?

45
00:04:28,840 --> 00:04:35,000
And that's sort of where it all started. I found a recipe that my family used to make.

46
00:04:35,000 --> 00:04:39,960
And we altered that recipe on the specific health conditions that were required.

47
00:04:39,960 --> 00:04:44,680
And we made it and it actually turned out really good. It kept blood sugar down.

48
00:04:44,680 --> 00:04:50,440
It did everything that it needed to do to match the new health conditions that my family was experiencing.

49
00:04:50,440 --> 00:04:57,400
And yeah, that sort of was the birth of the kitchen AI. And I went on to start building the entire thing.

50
00:04:57,400 --> 00:05:03,480
Yeah. So, I mean, and for our listeners and those who haven't had a chance to surf over to this, please do yourself a favor.

51
00:05:03,480 --> 00:05:08,600
This is a completely autonomous AI cooking website.

52
00:05:08,600 --> 00:05:14,520
All the, like we said in the intro, all the recipes are generated each day.

53
00:05:14,520 --> 00:05:21,640
All of the food imagery, all the copy. And let me tell you, as someone who fancies himself to know his way around the kitchen,

54
00:05:21,640 --> 00:05:29,960
it is exceptionally detailed and a testament to just how your workflow and everything is set up with the kitchen AI.

55
00:05:29,960 --> 00:05:34,600
Can you take us through a couple of the features? I've got it up here on the screen right now.

56
00:05:34,600 --> 00:05:38,520
Where would you direct somebody when they jump on the site for the first time?

57
00:05:38,520 --> 00:05:45,080
Yeah, sure. So, you can see all of the recipes that we've been doing had pretty much started since January.

58
00:05:45,080 --> 00:05:51,400
But once you get onto the front page, you can just click like today's top pick or today's picks at the top of the page.

59
00:05:51,400 --> 00:05:54,920
And that will just bring you to whatever we created that day.

60
00:05:54,920 --> 00:06:05,560
And this is pretty much almost fully automated process of creating the recipe, having it up onto the website and doing the whole social media campaign.

61
00:06:05,560 --> 00:06:12,360
And as you scroll down, basically what it starts with is just a little description of the dish, sort of what's going on.

62
00:06:12,360 --> 00:06:21,960
And then the do this with that stuff is the instructions on the dish and the you need this stuff is the ingredients.

63
00:06:21,960 --> 00:06:28,840
One of my favorite parts about building this thing out was because I wanted to I was doing manually like, you know,

64
00:06:28,840 --> 00:06:35,560
I'd write in a recipe and be like, you know, salmon rice dish, good for like high blood pressure or something.

65
00:06:35,560 --> 00:06:39,880
And it would create that recipe specifically for that health condition.

66
00:06:39,880 --> 00:06:47,880
But I wanted it to sort of be for anybody where I can put in something general and it'll alter that recipe based on what you need.

67
00:06:47,880 --> 00:06:57,080
So if you just scroll down a little bit there, one of the first things that we have is the obvious and not so obvious variations and pairings.

68
00:06:57,080 --> 00:07:01,080
And first, it starts with the flavor profiles.

69
00:07:01,080 --> 00:07:06,920
So when you're in the flavor profiles, say that you are like, oh, this dish was good, but I want it to be a little bit more saltier.

70
00:07:06,920 --> 00:07:13,240
It gives you sort of steps there to increase the sodium content, how you can how you can make it more salty.

71
00:07:13,240 --> 00:07:19,880
And it goes through all all different flavor profiles from like spicy and sweet and umami, everything.

72
00:07:19,880 --> 00:07:22,200
And it just tells you how to change that.

73
00:07:22,200 --> 00:07:27,400
This is so great. We're going to put some of this to the test here later on in the episode.

74
00:07:27,400 --> 00:07:34,360
So the dish that we pulled up here, I sent Keegan in advance of the episode something that I had cooked before.

75
00:07:34,360 --> 00:07:40,440
And I'm going to display that now. It's a nice piece of crispy skin salmon with some pineapple fried rice.

76
00:07:40,440 --> 00:07:43,960
I said, all right, well, let's see. Let's see a A.I.'s version of that.

77
00:07:43,960 --> 00:07:48,840
And later on, I'm going to prepare this and we'll kind of give a evaluation.

78
00:07:48,840 --> 00:08:03,480
So to give a specific example for this recipe, which is called sizzling Samba salmon with fiery pineapple pilaf, a dinner for two, has has a lot of Hawaiian and Thai inspired flavors.

79
00:08:03,480 --> 00:08:17,320
But if someone wanted to say, make it a little saltier, it has five recommendations like add two tablespoons more fish salt, add a half a teaspoon of garlic salt to the salmon seasoning, add a quarter cup of diced olives to the pilaf.

80
00:08:17,320 --> 00:08:29,320
Some like real actionable, like no kidding steps that honestly, like a professional chef would would make these or take these to make adjustments to their dish.

81
00:08:29,320 --> 00:08:42,280
What kind of I mean, did you consult any resources to kind of get it to perform in this way or is this just, you know, just built into these language models themselves, their general body of knowledge?

82
00:08:42,280 --> 00:08:46,920
Yeah, a lot of it like I did a lot of experimentation.

83
00:08:46,920 --> 00:08:54,360
Like one of the things that I really tried to nail down was what is everything that I can possibly get from a single recipe?

84
00:08:54,360 --> 00:09:03,480
You know, I went to a lot of recipe blogs and I've seen sort of what they do and usually it's just the ingredients and instructions and then a blog about like the stuff that happens.

85
00:09:03,480 --> 00:09:04,440
But I wanted to take it.

86
00:09:04,440 --> 00:09:09,640
Even else recipe blogs like someone's life story and then you have down recipe at the bottom.

87
00:09:09,640 --> 00:09:10,720
That's my experience.

88
00:09:10,720 --> 00:09:12,120
Anytime with that.

89
00:09:12,120 --> 00:09:13,800
Yeah, totally.

90
00:09:13,800 --> 00:09:30,040
An SEO optimized chunk, just so their lasagna recipe happens to, you know, hit the first page, but to do that, they got to tell you the entire written oral history of, you know, Tuscany before you get down to how many how many boxes of noodles are you going to need?

91
00:09:30,040 --> 00:09:30,680
Totally, man.

92
00:09:30,680 --> 00:09:36,840
Yeah, that was that was actually one of the things that inspired me to actually create a blog out of this.

93
00:09:36,840 --> 00:09:42,000
And, you know, one of the things that you'll notice is like there's no ads on the website and that's intentional.

94
00:09:42,000 --> 00:09:45,120
We tried doing ads in the beginning to see if it was worth it.

95
00:09:45,120 --> 00:09:51,400
But like we only want to do a small amount in the beginning anyways, but we decided, you know, let's just scrap all the ads.

96
00:09:51,400 --> 00:09:52,280
We hate ads.

97
00:09:52,280 --> 00:09:53,400
We hate bloat.

98
00:09:53,400 --> 00:09:57,720
Let's just get right down to the to the actual content of what's going on.

99
00:09:57,720 --> 00:10:02,560
And that's why our recipe is like, you know, the ingredients and instructions are right at the top.

100
00:10:02,560 --> 00:10:09,240
That in and of itself, regardless of who made it, AI or human is something you should celebrate my book.

101
00:10:09,240 --> 00:10:10,480
Yeah, totally.

102
00:10:10,480 --> 00:10:15,080
One of the a couple of the other things that I want to show with the variations and pairings,

103
00:10:15,080 --> 00:10:20,720
just to go through sort of the process of how these recipes are created is, you know,

104
00:10:20,720 --> 00:10:24,840
the next thing that you'll see after the flavor profiles is the drink pairings.

105
00:10:24,840 --> 00:10:26,520
And I thought this would be a fun addition.

106
00:10:26,520 --> 00:10:31,000
Like, you know how you're supposed to have like white wine with fish, for example.

107
00:10:31,000 --> 00:10:32,760
This is a fish. This is a fish dish.

108
00:10:32,760 --> 00:10:36,520
And one of the top things is Sashicon Blanc, a white wine.

109
00:10:36,520 --> 00:10:38,680
This is it's so on point.

110
00:10:38,680 --> 00:10:44,000
Maybe I'm overstating it, but like I've I have been obsessed with food since I was like a five year old.

111
00:10:44,000 --> 00:10:48,040
You know, I'll share some pictures of some of the stuff I've created in a newsletter.

112
00:10:48,040 --> 00:10:51,720
I mean, I experimented with having my own catering biz for a while.

113
00:10:51,720 --> 00:10:58,880
I mean, I'm just all I watch on YouTube is before AI videos was just food videos and cooking stuff,

114
00:10:58,880 --> 00:11:01,840
knife skills, training, all kinds of stuff like that.

115
00:11:01,840 --> 00:11:08,640
So, like, when I see some of these pairings, I'm like, it is it staggers me how specific

116
00:11:08,640 --> 00:11:10,960
and kind of spot on some of this is.

117
00:11:10,960 --> 00:11:13,760
I mean, for example, you just mentioned the Sauvignon Blanc.

118
00:11:13,760 --> 00:11:20,160
They mention a specific type of beer from Thailand, a lager style beer called Thai Singa beer

119
00:11:20,160 --> 00:11:22,360
to balance the heat from the Thai chilies.

120
00:11:22,360 --> 00:11:27,920
Well, I mean, like, does is that something from the model or is it, you know,

121
00:11:27,920 --> 00:11:36,000
some really clever zero shot or few shot prompts that is causing that information to come in?

122
00:11:36,000 --> 00:11:39,640
Yeah, it's honestly really it's really interesting.

123
00:11:39,640 --> 00:11:45,200
This is sort of the maybe like the third iteration on the back end of the kitchen AI.

124
00:11:45,200 --> 00:11:48,720
And there's definitely a lot of zero and one shot prompts.

125
00:11:48,720 --> 00:11:57,520
There's nothing specific on what to show for what it essentially the way that it's built is it takes the ingredients.

126
00:11:57,520 --> 00:12:03,440
So every piece of this is sort of broken down and it'll take the ingredients to determine these drink pairings.

127
00:12:03,440 --> 00:12:09,040
And it'll say, OK, based on these ingredients, what would be the best drink alcoholic and non alcoholic?

128
00:12:09,040 --> 00:12:11,520
What would be the best drink for this and why?

129
00:12:11,520 --> 00:12:16,240
And I wanted to test the limits of what I could do with, like, you know,

130
00:12:16,240 --> 00:12:20,560
really having a small prompt to increase in the prompt and getting more content in there.

131
00:12:20,560 --> 00:12:23,880
And it does a remarkable job. This is GPT four.

132
00:12:23,880 --> 00:12:31,440
It does a remarkable job at, you know, just figuring it out and knowing what flavor profiles go with what.

133
00:12:31,440 --> 00:12:35,960
Truly remarkable. And this is I think that's that's another thing to state.

134
00:12:35,960 --> 00:12:43,080
Like, this is base GPT four granted, the best probably language model out there right now, pulling these things together.

135
00:12:43,080 --> 00:12:48,800
But imagine the possibilities if you top layer fine tune on, you know,

136
00:12:48,800 --> 00:12:57,440
some specific culinary style or culinary textbooks or flavor textbooks or even like transcripts of cooking shows.

137
00:12:57,440 --> 00:13:00,600
I mean, like, man, just the possibilities.

138
00:13:00,600 --> 00:13:03,600
And then this I don't think I've ever seen.

139
00:13:03,600 --> 00:13:08,440
The thing we just clicked on right now is he's got a section for dietary restrictions

140
00:13:08,440 --> 00:13:17,720
where each kind of dietary style is mentioned and what instructions or modifications should be made to the dish based on that.

141
00:13:17,720 --> 00:13:22,280
So if you want to make this dish keto, it says remove the rice, remove the fish sauce.

142
00:13:22,280 --> 00:13:28,600
You want to make it Mediterranean, replace the Thai cilantro with fresh parsley, add tomatoes, add calamata.

143
00:13:28,600 --> 00:13:33,200
You want to make it paleo, pescatarian, vegan, vegetarian whole 30.

144
00:13:33,200 --> 00:13:41,000
I mean, that might be 10 searches in and of itself to get some of that insight in one place.

145
00:13:41,000 --> 00:13:52,400
Certainly scrolling through several life stories worth of text to to see if that top hit on Google has the information you need on how to make your dish paleo friendly.

146
00:13:52,400 --> 00:14:07,480
This, I mean, just such a clever idea. And I mean, did you have to dial in some specific parameters like, you know, say, hey, this is a paleo diet or does GPT-4 just know about that?

147
00:14:07,480 --> 00:14:18,800
Yeah, well, I mean, that's another thing that I had to experiment with a little bit was, you know, what does GPT actually know and how well does it know that specific topic?

148
00:14:18,800 --> 00:14:24,800
I tried it with really explaining what keto diet was, for example, and then I tried it with just not saying anything.

149
00:14:24,800 --> 00:14:30,920
I was saying this is a keto diet, you know, change these ingredients so it fits that dietary restriction.

150
00:14:30,920 --> 00:14:36,000
And it works, you know, without having to explain every single thing.

151
00:14:36,000 --> 00:14:44,880
You know, when it comes to prompt engineering, you know, the better your prompt, the more information that you can feed it typically results in a better output.

152
00:14:44,880 --> 00:14:50,880
The way that I had to structure, I didn't want to go through like 20 different, you know, dietary restrictions.

153
00:14:50,880 --> 00:14:52,320
This is just something that I wanted to do for fun.

154
00:14:52,320 --> 00:14:53,640
So I just put it out there.

155
00:14:53,640 --> 00:15:02,880
But, you know, if if I did go through every single dietary restriction and then really labeled it out, like don't have this, don't have this, don't have this, it should do this, it should do this, blah, blah, blah.

156
00:15:02,880 --> 00:15:05,200
The results probably would be a little bit better.

157
00:15:05,200 --> 00:15:13,920
But for now, I mean, what it is currently, it works like, you know, we've tested this out for like dietary restrictions.

158
00:15:13,920 --> 00:15:15,280
You know, and we did some calculation.

159
00:15:15,280 --> 00:15:24,800
But the problem is, though, is because this is AI and we don't have like I don't have any I don't have anything that's like going and checking to make sure that it is 100% accurate.

160
00:15:24,800 --> 00:15:26,720
That's to come later.

161
00:15:26,720 --> 00:15:28,640
But you can see there at the top, right?

162
00:15:28,640 --> 00:15:30,880
Like, important note, this is AI.

163
00:15:30,880 --> 00:15:32,200
So is it exact?

164
00:15:32,200 --> 00:15:34,760
Is it going to be, you know, 100%?

165
00:15:34,760 --> 00:15:36,920
Not necessarily, but it tries its best.

166
00:15:36,920 --> 00:15:40,320
And from what we've seen so far, it really works.

167
00:15:40,320 --> 00:15:51,160
Yeah, so the last section he has, and this is for every single recipe, there's these flavor profiles, pairings, dietary restrictions, more how to modify the recipe based on health conditions.

168
00:15:51,160 --> 00:15:54,600
You want someone with celiac disease or diabetes or heart disease.

169
00:15:54,600 --> 00:16:03,880
I mean, it's just like so, so thoughtful in the sense that this can be your one stop shop for for some recipe inspiration.

170
00:16:03,880 --> 00:16:06,200
What are some of your plans with the Kitchen AI?

171
00:16:06,200 --> 00:16:09,400
How do you hope to make it even more interactive?

172
00:16:09,400 --> 00:16:12,560
Yeah, so we're gonna we're getting some cookbooks ready.

173
00:16:12,560 --> 00:16:19,520
We thought it would be fun to have our robot recipes in an actual book form so you don't have to just search the web for everything.

174
00:16:19,520 --> 00:16:25,400
But the thing that I'm currently working on is really trying to turn this into a platform.

175
00:16:25,400 --> 00:16:34,760
You know, being able to set your own preferences and set your own dietary restrictions or, you know, do you like spicy food?

176
00:16:34,760 --> 00:16:36,800
Do you like spice at all?

177
00:16:36,800 --> 00:16:38,720
Or how spicy do you want things?

178
00:16:38,720 --> 00:16:50,040
And then having that specific profile and then run it through the model and then having these recipes being extremely catered to you is the direction that that we're heading in with this.

179
00:16:50,040 --> 00:16:58,760
Wow. I mean, you know, could you see it growing to a point where someone could essentially come in and just say, hey, I have these ingredients.

180
00:16:58,760 --> 00:17:04,000
Go to it, AI, you know, and I want lunch, something like that.

181
00:17:04,000 --> 00:17:07,720
Is there kind of like a vision for that or something?

182
00:17:07,720 --> 00:17:22,280
Yeah, totally. Yeah. One of the really cool things that I saw on YouTube is this this kid on YouTube, McKay Wrigley, that some people may have seen, maybe not, is this guy that taught himself how to code similar to me, I'd say.

183
00:17:22,280 --> 00:17:35,520
And he created this like Python app tool where it connects to his phone and he can point it in his fridge and it points out all of the ingredients that are in his fridge.

184
00:17:35,520 --> 00:17:39,840
That is totally something that I'd love to add into the platform.

185
00:17:39,840 --> 00:17:42,120
Computer vision cooking. Wow.

186
00:17:42,120 --> 00:17:48,280
Yeah. Yeah. You know, and having a list of ingredients that you have, like this is what I currently have.

187
00:17:48,280 --> 00:17:52,920
Yeah, totally. Like this is what I have. Make something with that is absolutely on the table.

188
00:17:52,920 --> 00:17:55,600
And you don't even need to use your thumbs to enter it in.

189
00:17:55,600 --> 00:18:03,440
You just need to hold your hold your phone up and point it at the refrigerator or the pantry and be like, AI, tell me what to make.

190
00:18:03,440 --> 00:18:08,080
Do this. Yeah. It opens a on one side of the coin.

191
00:18:08,080 --> 00:18:10,760
Is this the death of creativity in the kitchen?

192
00:18:10,760 --> 00:18:13,480
You know, because we're no longer thinking about it.

193
00:18:13,480 --> 00:18:25,040
I tend to fall on the side of, hey, you pair this with some someone with some serious expertise, a chef, and give them some of the levers to pull and twist and tweak.

194
00:18:25,040 --> 00:18:40,320
I mean, the possibilities of creating a AI generated menu, but that's prompted by a chef and maybe even trained on their prior recipes, I think is a tremendous prospect.

195
00:18:40,320 --> 00:18:48,640
Yeah, yeah, I would I would really like to I mean, first thing, I would really like to see some expert chefs trying these recipes.

196
00:18:48,640 --> 00:18:51,880
Like we've tried a couple of them and a couple of them are really, really good.

197
00:18:51,880 --> 00:18:56,720
Like, I got to say, one of the earlier ones that I tried was French toast.

198
00:18:56,720 --> 00:18:59,240
And the batter was a batter was unreal.

199
00:18:59,240 --> 00:19:08,120
But the thing that really made it spectacular was that after pan frying the French toast, it told you to put it in the oven for 10 minutes.

200
00:19:08,120 --> 00:19:10,400
And that's what made it. Oh, my God, man.

201
00:19:10,400 --> 00:19:13,800
It was that's what made it like the best French toast.

202
00:19:13,800 --> 00:19:17,600
Wow. See, that's yeah, technique stuff that it suggests.

203
00:19:17,600 --> 00:19:24,600
And that's not that's not commonplace in any recipe that you're going to find when you do a Google search for something.

204
00:19:24,600 --> 00:19:27,760
They are not going to give you the kind of technical pointers.

205
00:19:27,760 --> 00:19:38,560
And, you know, for me as a as a home cook, my recipes, my palate didn't evolve until I learned some some techniques, some knife skills.

206
00:19:38,560 --> 00:19:42,280
What pots to use for what measurement conversions.

207
00:19:42,280 --> 00:19:52,560
And another part of the kitchen, I that I think is just a great feature is you have a whole section on tips, which is just this technique stuff.

208
00:19:52,560 --> 00:19:56,520
Yeah, it actually it stemmed from an idea from my wife.

209
00:19:56,520 --> 00:20:01,320
And she was like, I think if you go to the bottom there, I can't remember what the very first one was.

210
00:20:01,320 --> 00:20:08,720
But one of the things that she wanted to do, she was like, how can I prepare my food best?

211
00:20:08,720 --> 00:20:15,000
And let's see the Mazen place. So it's like it's a plus. Yeah, Mazen plus. Yeah.

212
00:20:15,000 --> 00:20:19,560
So she was like, how can I do this best? And we asked Chachi PT.

213
00:20:19,560 --> 00:20:22,160
We started with Chachi PT and like asked, how do we do this?

214
00:20:22,160 --> 00:20:27,080
And then we like built a prompt around sort of getting these kitchen tips.

215
00:20:27,080 --> 00:20:37,120
And it's it's incredible, man, like the the amount of knowledge that you can get from artificial intelligence and these these generative AI is so great.

216
00:20:37,120 --> 00:20:42,080
And we thought that it would be so fitting to have this type of knowledge on our blog.

217
00:20:42,080 --> 00:20:48,280
I just pulled up the the article on mastering the art of Mies and plus for those that don't know, that means all in place.

218
00:20:48,280 --> 00:20:55,920
And is the the the phrase that when it's said in a professional kitchen, you know, hey, chef, you know, is your Mies ready?

219
00:20:55,920 --> 00:21:04,880
You ready with your Mies? It's basically their organizational technique, how stuff is laid out so they can cook and prepare something as efficiently as possible.

220
00:21:04,880 --> 00:21:11,200
And it's a great analogy for life as well, you know, kind of having everything all in place.

221
00:21:11,200 --> 00:21:17,600
So, I mean, I can't say enough about some of the different components that you've thought about.

222
00:21:17,600 --> 00:21:23,360
Yeah, thanks. Yeah, I'm really excited to see sort of where this where this whole thing goes.

223
00:21:23,360 --> 00:21:28,840
Every time that we tell somebody about the kitchen, the response has just been so positive.

224
00:21:28,840 --> 00:21:34,840
And the further they dig into it and the more that they see of what we're doing, they're like, oh, my God, that's so cool.

225
00:21:34,840 --> 00:21:39,760
Like, let's try some of these things. And, you know, it's great seeing what people can do.

226
00:21:39,760 --> 00:21:43,800
Well, speaking of try, I thought it'd be very appropriate.

227
00:21:43,800 --> 00:21:47,200
We're going to we're going to cut away at this point in the podcast.

228
00:21:47,200 --> 00:21:54,840
Please check us out on YouTube, because I am going to prepare the dish that I came up with for me.

229
00:21:54,840 --> 00:22:02,160
This the salmon dish that I've made before. I'm going to cook a version of it and report back.

230
00:22:02,160 --> 00:22:15,800
Have you discussed or thought about possibly getting together with culinary school or finding a really high end innovative chef to potentially say, hey, let's let's take a shot at this.

231
00:22:15,800 --> 00:22:19,960
Let's cook an A.I. menu for diners and see how it is.

232
00:22:19,960 --> 00:22:26,920
Yeah, we're talking with some of the schools here in Calgary is where we are, Calgary, Canada, and chatting with SAIT.

233
00:22:26,920 --> 00:22:32,000
SAIT has a division focused on the culinary arts, which is great.

234
00:22:32,000 --> 00:22:37,760
So we're looking to do some things with them and see what they can do using the kitchen, AI and the kitchen recipes.

235
00:22:37,760 --> 00:22:43,600
You know, and we think it would be really cool to have, you know, chef versus A.I.

236
00:22:43,600 --> 00:22:51,680
Like, who can do it better and definitely getting into some more social media content like like TikTok videos would be fun.

237
00:22:51,680 --> 00:23:01,680
But one of my things that I'd love to see would be having like some top chefs go head to head with some of these recipes and seeing what can come out.

238
00:23:01,680 --> 00:23:11,040
One of my when it comes to the future of the kitchen, AI, this is already implemented and we're just getting ready to release,

239
00:23:11,040 --> 00:23:14,000
which is the sort of the platform that we're building.

240
00:23:14,000 --> 00:23:17,120
But it has the component of changing the difficulty level.

241
00:23:17,120 --> 00:23:23,280
So, you know, if you're a beginner, it'll give you recipes and instructions specifically for beginner level.

242
00:23:23,280 --> 00:23:25,360
But I thought I would see how far we go.

243
00:23:25,360 --> 00:23:29,320
And I created a master level, which I'm not a master chef.

244
00:23:29,320 --> 00:23:33,000
Like I cook a little bit, but definitely not that level.

245
00:23:33,000 --> 00:23:39,600
And so I did a little bit of research on some of the terminology and I created a master difficulty

246
00:23:39,600 --> 00:23:50,320
where it goes into like some really exotic foods and some really complex instructions and using like high end cookware and stuff like that as well.

247
00:23:50,320 --> 00:23:55,440
I didn't try making them yet, but I tried building them out just to read sort of what it would create.

248
00:23:55,440 --> 00:24:00,400
And I'm really excited. I would love to see some master level chefs trying trying some of these recipes.

249
00:24:00,400 --> 00:24:06,920
I mean, I'm eager to try that because I'm the second it recommends me, you know, hey, get your agar agar

250
00:24:06,920 --> 00:24:12,000
and make little tiny spheres of your sauce or your dressing as caviar.

251
00:24:12,000 --> 00:24:14,040
I'm going to be like, oh, my God, this is too much.

252
00:24:14,040 --> 00:24:26,200
This is it's like it knows how to knows how to cut into chef's weaknesses in terms of like something that's just so tedious in the kitchen, just something so so beautiful looking.

253
00:24:26,200 --> 00:24:27,560
Yeah, totally.

254
00:24:27,560 --> 00:24:29,160
Such a that's such a fun idea.

255
00:24:29,160 --> 00:24:31,360
I mean, I encourage everyone to go check out the kitchen.

256
00:24:31,360 --> 00:24:36,560
I, you know, throw them a little bit of support by buying a hat, buying an apron, you know,

257
00:24:36,560 --> 00:24:46,280
even if you don't end up making anything, I happen to love all the little avatars and robot chefs and little colonistas that it seems to come up with.

258
00:24:46,280 --> 00:24:50,720
And how does it make some good food photography pictures, too?

259
00:24:50,720 --> 00:24:58,760
Yeah, you know, it was a really fun process creating these images because all the images are generated through the large language model as well.

260
00:24:58,760 --> 00:25:02,720
So it'll look at the recipe, the instructions and the ingredients.

261
00:25:02,720 --> 00:25:07,040
And then it'll be like, based on this information, how should this look?

262
00:25:07,040 --> 00:25:09,560
And then it'll give a prompt on how it should look.

263
00:25:09,560 --> 00:25:11,440
And then, you know, just pop that into.

264
00:25:11,440 --> 00:25:16,960
Yeah, pop that into a text to image A.I. and then creates everything.

265
00:25:16,960 --> 00:25:26,960
It's remarkable. So your image generation prompt is actually considering the ingredients themselves and how it's prepared and how it's prepared in its formulation.

266
00:25:26,960 --> 00:25:33,960
Man, that's an interesting little bit of prompt engineering there, too, because as opposed to just describing the end state of a dish,

267
00:25:33,960 --> 00:25:41,120
this is starting from the raw ingredients and using those as building blocks within the prompt to create.

268
00:25:41,120 --> 00:25:47,840
I would arguably say some of the best food like generated food images in my prior attempts.

269
00:25:47,840 --> 00:25:53,800
It tends to take a lot of iteration, at least for me, to just kind of get every little single component right.

270
00:25:53,800 --> 00:25:59,680
You know, you're making a close up image of a burger, but it's got, you know, fries on the burger.

271
00:25:59,680 --> 00:26:05,520
Well, unless you're going for that, like, all right, well, that's not what it's not how people are used to seeing it.

272
00:26:05,520 --> 00:26:09,040
Just the fact that is what's done to make these images come together.

273
00:26:09,040 --> 00:26:10,800
I think it's just tremendous.

274
00:26:10,800 --> 00:26:17,600
Just on top of that as well, with how I'm building the new platform is for every instruction,

275
00:26:17,600 --> 00:26:23,680
it'll tell you what it should smell like, what it should taste like, the textures.

276
00:26:23,680 --> 00:26:31,080
So as you're building up this recipe, it'll tell you sort of what to expect for every step and to really help you hone in.

277
00:26:31,080 --> 00:26:36,880
And I think like the Kitchen AI is almost it could almost turn into like an education on how to cook.

278
00:26:36,880 --> 00:26:38,880
Absolutely, absolutely.

279
00:26:38,880 --> 00:26:46,080
I mean, I'm just as we speak here, I just happen to click on another recipe that is traditionally a little harder of a dessert to make.

280
00:26:46,080 --> 00:26:49,200
Macarons, you have to use almond flour.

281
00:26:49,200 --> 00:26:51,160
Temperatures are very important.

282
00:26:51,160 --> 00:26:57,840
You actually cook it in a kind of cooler oven traditionally, and the Kitchen AI nails that on the head.

283
00:26:57,840 --> 00:27:00,680
300 degree oven for a longer period of time.

284
00:27:00,680 --> 00:27:09,720
It even make recommendations about some of the ingredients, the temperatures they need to be at before assembling stuff together, which in baking is vitally important.

285
00:27:09,720 --> 00:27:14,080
So just the level of detail you've put into this, man, I encourage everyone to check it out.

286
00:27:14,080 --> 00:27:19,520
Even just, you know, spin the wheel and see what you land on and say, all right, let's try it.

287
00:27:19,520 --> 00:27:22,360
Let's try to make it. I think you might surprise yourself.

288
00:27:22,360 --> 00:27:23,600
Thanks, man.

289
00:27:23,600 --> 00:27:25,120
Cool. Well, have you?

290
00:27:25,120 --> 00:27:37,080
I think this is a terrific example of just one of those use cases for AI, specifically generative AI in a industry, in a world that's not tech centric.

291
00:27:37,080 --> 00:27:45,400
I mean, the culinary industry does not have a core that's foundational, you know, founded upon technology.

292
00:27:45,400 --> 00:27:56,520
Are there any other industries or practices or things that you know about that you find compelling that's a non tech centric industry that uses generative AI?

293
00:27:56,520 --> 00:27:59,880
That uses currently or that could use?

294
00:27:59,880 --> 00:28:04,920
Yeah, or could use, you know, something that you think might be an interesting use of it.

295
00:28:04,920 --> 00:28:06,640
Yeah, that's a great question.

296
00:28:06,640 --> 00:28:11,040
I mean, fashion was one thing that we were chatting about a little bit.

297
00:28:11,040 --> 00:28:11,760
Yeah.

298
00:28:11,760 --> 00:28:29,960
And, you know, my wife is in fashion and she the things that AI can help you with and can do is just remarkable, especially with mid journey version five and above can just do some stunning images of art, such as what you're looking at here.

299
00:28:29,960 --> 00:28:53,040
So, I mean, I pulled this up too, and it's so funny that your spouse also comes at this from a, you know, a textile and fabric and that kind of creative, that creative angle, because I found in being able to watch my wife kind of describe a certain dress or certain look that they're trying to achieve in an image.

300
00:28:53,040 --> 00:28:57,440
It's all terminology that I would never know or include in there.

301
00:28:57,440 --> 00:29:03,920
Right now, I pulled up a couple different images of a type of dress she was looking to create.

302
00:29:03,920 --> 00:29:16,000
So these are photorealistic models will include them in the newsletter that are wearing a dress with ruffle sleeves that flutter at the shoulders and merge around the scoop neck.

303
00:29:16,000 --> 00:29:23,040
It's fitted smocked body flounces and a dreamy midi skirt decorated with custom lace inserts.

304
00:29:23,040 --> 00:29:32,800
And then it goes on to describe the model itself, but I'm like that level of detail, you know, that it just shows points to that person's creativity, that person's expertise.

305
00:29:32,800 --> 00:29:45,800
It achieves this tremendous looking image that I find I find at least is is, you know, kind of kind of also take take fashion to new levels, maybe.

306
00:29:45,800 --> 00:29:46,240
Yeah.

307
00:29:46,240 --> 00:29:58,760
And when it comes to prompt engineering as well, I mean, that's a long prompt. And one of my one of the things that I've really realized about prompt engineering is it's not necessarily how big the prompt is.

308
00:29:58,760 --> 00:30:11,000
Like I said earlier, you know, a large prompt can be good, but it's the content in that prompt where you describe things in detail and go through exactly what you're looking for.

309
00:30:11,000 --> 00:30:19,240
You know, one of the one of the things that I tell people all the time is the better you know what you're talking about, the better you're able to do some prompting.

310
00:30:19,240 --> 00:30:26,800
And, you know, I don't know anything about fashion, but the description of those images goes through a lot of stuff that I would have no idea about.

311
00:30:26,800 --> 00:30:28,720
And, you know, look at the results.

312
00:30:28,720 --> 00:30:47,600
And I think it actually kind of does does something twofold here, aside from it producing a much more, I would venture to say, striking and unique image that from a perspective of someone who works in generative AI and has clients that commission him to make professional images like this.

313
00:30:47,600 --> 00:31:03,160
It makes the outcomes better. But then there's also this whole other side of it that is a emerging industry right now where people will prompt AI to create wild fashion out there.

314
00:31:03,160 --> 00:31:05,520
That's so that's so wild, man.

315
00:31:05,520 --> 00:31:19,560
What we're looking at right now is a picture of just this. I mean, how would you even describe it? It's like a it's a Nike sweatsuit being worn by a model, but the the the neck of it is up really like a super tall neck.

316
00:31:19,560 --> 00:31:23,840
It's got kind of this capy futuristic thing happening.

317
00:31:23,840 --> 00:31:29,560
I mean, anything else you could even like, you see something like that.

318
00:31:29,560 --> 00:31:32,720
But it's but it works in its own way.

319
00:31:32,720 --> 00:31:41,360
One of the one of the cool things I messed around with this a little bit, and one of one of the fun things I like to sort of see what AI thinks of is like, what do you think of the future?

320
00:31:41,360 --> 00:31:54,520
And, you know, in this case, you could create a prompt that is like, show me a realistic image of a model in 2200.

321
00:31:54,520 --> 00:32:02,280
What do they look like? And like some of the some of the results are so mind blowing because it's just like, oh, this is what I think it could be, you know.

322
00:32:02,280 --> 00:32:04,200
And I think that that's so cool.

323
00:32:04,200 --> 00:32:11,560
No, I know I've seen several Instagram accounts of people that will do that kind of prompting and then go, all right, I'm going to go make that for real.

324
00:32:11,560 --> 00:32:24,720
Yeah, so much so that I think there's an actually a entire site now called Cala where you can do generative AI from within or upload generated AI images.

325
00:32:24,720 --> 00:32:33,120
I'm pulling it up on the site now and you can provide little notes on certain parts of the garment where you want it to feel textures.

326
00:32:33,120 --> 00:32:38,160
And then the service will actually give you pricing for a certain number of those units to be made.

327
00:32:38,160 --> 00:32:48,040
And the sizing, the mood board, it's very kind of innovative way to basically capture the entire supply chain or the entire design process, at least.

328
00:32:48,040 --> 00:32:50,280
Yeah, yeah, definitely.

329
00:32:50,280 --> 00:32:58,720
Yeah, I love I love the direction that AI is going and I think that we're going to see such a different world by this time next year.

330
00:32:58,720 --> 00:33:07,240
AI, especially generative AI has been mainstream, maybe close, not even mainstream, but like definitely getting bigger over the last six months.

331
00:33:07,240 --> 00:33:11,400
It's growing exponentially. It's getting better so quickly.

332
00:33:11,400 --> 00:33:19,920
And to see what we you know, to see where we're at next year, you know, next June, July is just going to be such a different landscape.

333
00:33:19,920 --> 00:33:32,160
I had seen a study today, one of the biggest surveys they had done on, you know, people's knowledge and having tried a bunch of different types of generative AI technologies, at least here in the States.

334
00:33:32,160 --> 00:33:37,960
And it was something like 58 percent of people knew about chat GPT or had tried it.

335
00:33:37,960 --> 00:33:46,160
That was the most widespread one at this point with the one that's the least known about is a stable diffusion platform,

336
00:33:46,160 --> 00:33:50,160
because that also is a little steeper learning curve to get up and running on.

337
00:33:50,160 --> 00:33:52,360
But, you know, you're absolutely right.

338
00:33:52,360 --> 00:33:57,960
Like, there's still plenty of people that even just kind of like figure out here about this stuff, use it.

339
00:33:57,960 --> 00:34:01,600
It's only going to only going to take off from here.

340
00:34:01,600 --> 00:34:14,720
I think as well, like one of my favorite things about this whole industry is that the amount of collaboration and open source where people are like they're actually like a lot of these things you can actually like.

341
00:34:14,720 --> 00:34:18,760
Falcon was one of the most I don't know, it was a couple of weeks ago.

342
00:34:18,760 --> 00:34:20,880
I can't keep up anymore with the AI race.

343
00:34:20,880 --> 00:34:21,640
It's so quick.

344
00:34:21,640 --> 00:34:31,320
But Falcon was at one point one of the best open source models out there that was really close to chat GPT and other other large language models.

345
00:34:31,320 --> 00:34:32,800
And it's completely open source.

346
00:34:32,800 --> 00:34:35,600
You can download it, put on your computer, run it on your own hardware.

347
00:34:35,600 --> 00:34:40,200
You know, stable diffusion, I think, is also open source, right?

348
00:34:40,200 --> 00:34:47,080
Yeah, so it's just it's such an awesome landscape that we're living in where people are contributing and building these things out.

349
00:34:47,080 --> 00:34:57,920
I think we were talking about this maybe a few weeks ago, but, you know, when you had the dot com boom, that was us, you know, actually discovering the Internet, how to use it.

350
00:34:57,920 --> 00:34:59,120
I mean, the term, what is it?

351
00:34:59,120 --> 00:35:10,160
The term surfing the web literally means people just used to put in different combinations of words, you know, before they were indexed by Google, before they were searchable and see where it got them.

352
00:35:10,160 --> 00:35:11,400
There was the website about it.

353
00:35:11,400 --> 00:35:20,120
If someone had made something, that was literally what the origin of surfing the web is now that it's so ingrained in every aspect of our society and daily lives.

354
00:35:20,120 --> 00:35:25,360
You have all this creativity and development happening at the same time at once.

355
00:35:25,360 --> 00:35:28,280
You know, everything everywhere all at once.

356
00:35:28,280 --> 00:35:29,640
Love that movie.

357
00:35:29,640 --> 00:35:34,520
Yeah, that's a good thing that now kind of like, you know, gets much more collaboration happening.

358
00:35:34,520 --> 00:35:38,760
And then someone does something over here, gives someone an idea to iterate on that.

359
00:35:38,760 --> 00:35:49,200
And it just kind of exponential is not even the word, it's vertical in terms of the growth of a lot of these different type of things or ideas or possibilities.

360
00:35:49,200 --> 00:35:58,320
One of my favorite things about the AI dilemma was that they said something along the lines of like, AI is growing so fast, we cannot conceive it.

361
00:35:58,320 --> 00:36:02,080
We cannot even imagine how fast it's growing.

362
00:36:02,080 --> 00:36:04,600
I think they said like, I can't remember exactly.

363
00:36:04,600 --> 00:36:05,920
I watched the video once.

364
00:36:05,920 --> 00:36:08,320
I think they said it was like double exponential.

365
00:36:08,320 --> 00:36:12,640
It was so fast that we just can't understand how fast it is.

366
00:36:12,640 --> 00:36:14,320
I mean, that's I totally agree.

367
00:36:14,320 --> 00:36:19,000
Well, yeah, with that, a good enough place to end as anywhere.

368
00:36:19,000 --> 00:36:27,680
So for everyone here at HTTTA, I am Wes, the synth mind for Keegan Dargi saying happy prompting everybody.

369
00:36:27,680 --> 00:36:28,080
Thank you.

370
00:36:28,080 --> 00:36:41,840
As always, you can check out the show notes and links at how to talk to dot AI.

371
00:36:41,840 --> 00:36:44,040
That's all for this week's episode.

372
00:36:44,040 --> 00:36:58,640
Happy prompting, everyone.