1
00:00:01,080 --> 00:00:02,960
Speaker 1: How'd you like to listen to dot net rocks with

2
00:00:03,040 --> 00:00:03,960
no ads?

3
00:00:04,440 --> 00:00:04,839
Speaker 2: Easy?

4
00:00:05,360 --> 00:00:08,560
Speaker 1: Become a patron For just five dollars a month, you

5
00:00:08,599 --> 00:00:11,320
get access to a private RSS feed where all the

6
00:00:11,359 --> 00:00:14,599
shows have no ads. Twenty dollars a month will get

7
00:00:14,599 --> 00:00:17,679
you that and a special dot net Rocks patron mug.

8
00:00:18,160 --> 00:00:34,399
Sign up now at Patreon dot dot NetRocks dot com.

9
00:00:34,439 --> 00:00:37,520
Hey guess what, it's dot net rocks from Bill. This

10
00:00:37,560 --> 00:00:39,920
is the last show we were recording. I'm Carl Franklin.

11
00:00:40,600 --> 00:00:44,840
We're here with Sef Juarez and we'll be talking to.

12
00:00:44,840 --> 00:00:48,119
Speaker 2: Get that out of there. Get a little fam Sorry,

13
00:00:48,200 --> 00:00:50,560
you gotta get it out. You know, we're a little punchy.

14
00:00:50,600 --> 00:00:52,159
This is not only the last show of the day,

15
00:00:52,200 --> 00:00:54,920
but it's the last show up. Oh, if you got

16
00:00:54,960 --> 00:00:57,759
a punchy a show with a punchy dude, it's gonna

17
00:00:57,759 --> 00:00:58,600
be great.

18
00:00:58,759 --> 00:01:01,640
Speaker 1: It's gonna be You've been working hard, that's right. I

19
00:01:01,719 --> 00:01:03,560
just want to start talking to Seth. Do we really

20
00:01:03,600 --> 00:01:05,719
have to do the intro stuff? You should do the interesting?

21
00:01:06,040 --> 00:01:07,840
Speaker 2: We gotta do it. You got to pay so you

22
00:01:07,920 --> 00:01:11,480
know it's episode nineteen fifty seven. Wow, that's the year

23
00:01:11,519 --> 00:01:14,079
I was born. It's not true. It could be in

24
00:01:14,120 --> 00:01:16,519
a different universe very close. It could be in a

25
00:01:16,560 --> 00:01:19,439
parallel universe. If you know quantum theory, the many worlds

26
00:01:20,120 --> 00:01:23,319
as you know, Richard In posits that perhaps there's another

27
00:01:23,400 --> 00:01:25,959
universe where I was born in nineteenfty seven. There's a

28
00:01:26,040 --> 00:01:30,640
million universes. That happened, sixty seven year old set. I

29
00:01:30,719 --> 00:01:32,959
just wanted to use the word posits in the thing.

30
00:01:33,040 --> 00:01:35,959
And so now my job is I'm done. You must

31
00:01:35,959 --> 00:01:37,319
have a master's degree.

32
00:01:37,680 --> 00:01:41,000
Speaker 1: I have something. So do you want to do the

33
00:01:41,079 --> 00:01:44,040
history first? You want to do I didn't do the history.

34
00:01:44,120 --> 00:01:46,439
So what happened in nineteen fifty spot Nick one?

35
00:01:46,840 --> 00:01:50,640
Speaker 2: Okay, the first the first artificial salarate over the world,

36
00:01:50,680 --> 00:01:53,200
and then sput Nick two. They put a dog in

37
00:01:53,280 --> 00:01:56,640
it and she died. No, actually no, she was in

38
00:01:56,680 --> 00:02:02,799
the Guardians of the Galaxies. Right, wait, that's not real.

39
00:02:03,719 --> 00:02:07,159
We want the historical documents and it's worse than any think.

40
00:02:07,200 --> 00:02:09,680
She didn't she didn't die and re entry, she didn't suffocate.

41
00:02:10,000 --> 00:02:14,599
She overheated. There wasn't temperature controls oh thanks.

42
00:02:15,319 --> 00:02:19,199
Speaker 1: Sometime around here in the nineteen late nineteen fifties, electric

43
00:02:19,280 --> 00:02:24,360
boat from General Dynamics was established in the Nautilus.

44
00:02:23,719 --> 00:02:27,039
Speaker 2: That was no electric boat? Is that? Is that a

45
00:02:27,280 --> 00:02:29,719
like a line dance? What is fifty five? Isn't it?

46
00:02:29,800 --> 00:02:33,800
The electric boat? Is that like the electric slide? But no, no,

47
00:02:33,800 --> 00:02:37,560
no waste? The nuclear submarine? Oh yeah, the first nuclear summer, theatlis.

48
00:02:37,280 --> 00:02:42,039
Speaker 1: That's right, Okay, I'm sorry. I had a forward memory of.

49
00:02:42,199 --> 00:02:46,719
Speaker 2: I also usually referenced a computer, and in nineteen fifty seven,

50
00:02:46,800 --> 00:02:50,759
IBM released a compact computer, the six', ten only eight hundred,

51
00:02:50,759 --> 00:02:54,240
pounds much smaller than the previous. Ones it really was

52
00:02:54,280 --> 00:02:56,000
an improvement there you. GO i, mean if you want

53
00:02:56,000 --> 00:02:58,919
to get, fit you get one of those commonract eight

54
00:02:59,000 --> 00:03:03,800
hundred pounds. Computers also released the first compiler for. Fortrend hey, yo,

55
00:03:03,919 --> 00:03:07,439
wow still running some code out. There there's as for

56
00:03:07,599 --> 00:03:10,319
trend dot hold. On So i'm not. Kidding there are

57
00:03:10,400 --> 00:03:13,800
some linear algebra operations that were written In fortrand that

58
00:03:13,800 --> 00:03:17,000
people were, like let's just not do that. AGAIN i

59
00:03:17,000 --> 00:03:19,879
think it's like some WEIRD langsos method for, computing like

60
00:03:19,960 --> 00:03:23,360
singular value, decomposition that they're, like let's just not do that.

61
00:03:23,400 --> 00:03:26,400
Again somebody did, that and we're, gonna we're, gonna we're

62
00:03:26,439 --> 00:03:30,879
gonna honor them and we're just gonna leave that. Alone. Wow.

63
00:03:32,080 --> 00:03:33,599
WOW i actually contributed this.

64
00:03:33,639 --> 00:03:36,919
Speaker 1: SOMEHOW i actually remember taking A fortran. Class but it

65
00:03:36,960 --> 00:03:41,039
was like in the. Eighties, yeah a little further, on yeah,

66
00:03:41,960 --> 00:03:44,639
too it was like in THE i did fourtran.

67
00:03:44,759 --> 00:03:52,199
Speaker 2: Three oh is there a sound? Effect weing? At BUT

68
00:03:53,800 --> 00:03:57,080
i love? It all? Right, well, anyway anything else in

69
00:03:57,159 --> 00:03:59,120
nineteen fifty there was a lot Of there was a

70
00:03:59,159 --> 00:04:00,960
lot of. Things there's new clear. Test it's all kinds

71
00:04:00,960 --> 00:04:03,400
of horrible stuff listed with, computers all, right and speaking

72
00:04:03,439 --> 00:04:08,000
of atomic, fires what are we talking about? Today, well

73
00:04:07,560 --> 00:04:12,400
we gotta finished the fish such a. Hurry i'm not

74
00:04:12,520 --> 00:04:12,680
in a.

75
00:04:12,759 --> 00:04:16,879
Speaker 1: Hurry, okay all, right let's do the better note of

76
00:04:16,959 --> 00:04:28,720
framework roll the, music all. Right, so as you, Know

77
00:04:28,920 --> 00:04:31,879
Jeff fritz AND I Jeff frits For microsoft And big

78
00:04:31,920 --> 00:04:34,120
INTO ai And. Blazer we do This blazer puzzle.

79
00:04:34,160 --> 00:04:35,560
Speaker 2: Show AND i was just.

80
00:04:35,519 --> 00:04:38,319
Speaker 1: Talking to him about all of the cool THINGS i

81
00:04:38,360 --> 00:04:40,680
saw at the, keynote including, you my friend oh hey

82
00:04:41,040 --> 00:04:43,120
uh at the keynote it build and of course he's

83
00:04:43,160 --> 00:04:45,680
hipped to all of this, stuff and we were kind

84
00:04:45,680 --> 00:04:48,079
of talking about how and we talked about this with

85
00:04:48,120 --> 00:04:52,759
some of our guests this. Week about how the prompts

86
00:04:52,839 --> 00:04:57,319
are becoming code and instead of having code, libraries people

87
00:04:57,360 --> 00:05:01,519
are going to start sharing prompt. Libraries Prompt so what

88
00:05:01,600 --> 00:05:05,920
Did fritz point me? To he pointed me to sharp,

89
00:05:06,000 --> 00:05:11,959
site which is all these prompts for Doing blazer applications

90
00:05:12,959 --> 00:05:13,240
with the.

91
00:05:13,279 --> 00:05:15,480
Speaker 2: Agent oh, interesting do they? Work?

92
00:05:15,680 --> 00:05:19,399
Speaker 1: Yeah in good hub co pilot so code style and,

93
00:05:19,480 --> 00:05:25,879
structure naming, conventions dot net specific, guidelines sho air handling and,

94
00:05:25,959 --> 00:05:29,680
validation like all of these great prompts to help you

95
00:05:31,160 --> 00:05:32,879
create Great blazer.

96
00:05:32,879 --> 00:05:36,519
Speaker 2: Applications that's all the. Agent how about? THAT i think that's.

97
00:05:36,560 --> 00:05:38,839
Great anything that we can do to further the work OF,

98
00:05:38,879 --> 00:05:41,639
ai because you, Know i've BEEN i was like on

99
00:05:41,720 --> 00:05:44,920
show like, SEVENTEEN i don't, know something like that eighteen

100
00:05:45,360 --> 00:05:47,839
AND i was, like machine learning is the future AND

101
00:05:47,839 --> 00:05:49,920
i was like What and we didn't understand. It then

102
00:05:49,959 --> 00:05:52,199
we still, don't but at least it's. Working but it's,

103
00:05:52,240 --> 00:05:55,480
true and you just still understand. It we just, believe you,

104
00:05:55,519 --> 00:05:58,639
KNOW i really excited that people are, like let's use

105
00:05:58,680 --> 00:06:03,000
this stuff right five seventy Six july of twenty. Ten

106
00:06:03,160 --> 00:06:07,279
oh my, gosh aires on machines in eighteen years AND

107
00:06:07,319 --> 00:06:09,560
i just get more. Beautiful i'm glad this is. True

108
00:06:09,720 --> 00:06:12,639
in this, medium it's really hard to, know but it's for.

109
00:06:12,680 --> 00:06:14,240
SURE i want you to imagine, it all.

110
00:06:14,319 --> 00:06:17,720
Speaker 1: Right richard Save, us who's talking to us, today.

111
00:06:17,560 --> 00:06:19,879
Speaker 2: Grabbed to comment off a show nineteen, nineteen which you

112
00:06:19,879 --> 00:06:21,759
did in twenty twenty, four a little more recent In

113
00:06:21,800 --> 00:06:26,199
october With Prascent boy here he's AN mvp DOING ai

114
00:06:26,279 --> 00:06:28,240
related stuff that he wanted to talk, about and we

115
00:06:28,240 --> 00:06:31,040
did talk About Copilot, studio WHICH i believe has now

116
00:06:31,040 --> 00:06:35,000
been renamed because. Reasons, yeah And John tallon had this

117
00:06:35,079 --> 00:06:36,680
great comments from a few months. Ago he, SAID i

118
00:06:36,759 --> 00:06:39,000
enjoyed the. Show chat BASED ai apps are quick to stand,

119
00:06:39,120 --> 00:06:41,839
up but tough to get right every time a man

120
00:06:42,480 --> 00:06:46,439
rag dazzles that. First but yet it's limited without deep. Engineering.

121
00:06:46,560 --> 00:06:49,240
Ooh in real, conversations not everything is a, question and

122
00:06:49,279 --> 00:06:52,360
bought led chats are no. Different we must parse the user's,

123
00:06:52,399 --> 00:06:54,879
intent evaluate the quality of their, input then apply a

124
00:06:54,959 --> 00:06:58,680
chain of erication to fact checkbox. Responses i'm a practice

125
00:06:58,680 --> 00:07:01,279
director in data AND II, ai And i've worked in Gen,

126
00:07:01,360 --> 00:07:04,160
Ai AZURE Ai studio and Co Pilot, studio and despite

127
00:07:04,160 --> 00:07:08,519
the Marketing Copilot studio isn't for your Quote jim From.

128
00:07:08,560 --> 00:07:11,839
Accounts it's for seasoned consultants Like brashant who know how

129
00:07:11,879 --> 00:07:14,879
to use the. Platform using it for complex, interactions like

130
00:07:14,879 --> 00:07:16,720
a rocket launcher on a. Fly, yeah it, works but

131
00:07:16,720 --> 00:07:21,279
it's a little over. Complicated rocket launcher on a. Fly it's,

132
00:07:21,399 --> 00:07:23,600
like use. It, yeah let me say that. Again using

133
00:07:23,600 --> 00:07:26,120
it for complex interactions is like using a rocket launcher

134
00:07:26,319 --> 00:07:26,680
on a.

135
00:07:26,680 --> 00:07:30,279
Speaker 1: Fly it's kind of taken a nod from my term

136
00:07:30,399 --> 00:07:31,480
swatting a fly with a.

137
00:07:31,560 --> 00:07:35,319
Speaker 2: Buick we had a term swatting a fly with a,

138
00:07:35,360 --> 00:07:37,560
shotgun and my algorithms professor was, like we have an

139
00:07:37,600 --> 00:07:40,639
algorithm for sorting that's faster than en log, in but

140
00:07:40,920 --> 00:07:44,639
only with giant data. Sets it's like swatting a fly

141
00:07:44,759 --> 00:07:47,439
with a. Shotgun, yes, yeah. Nice we should only explose

142
00:07:47,519 --> 00:07:49,959
enough complexity to get the job. Done letting use His

143
00:07:50,000 --> 00:07:53,519
harner's AI's potential without feeling. Overwhelmed after, all the goal

144
00:07:53,600 --> 00:07:56,879
is to DEMOCRATIZE, ai not. Intimidate what year was? That

145
00:07:56,879 --> 00:07:59,199
that was last? Week, yeah this is like a few months.

146
00:07:59,240 --> 00:08:03,240
Ago the answer is you just, wait wait till next.

147
00:08:03,439 --> 00:08:05,000
Years you. Know one of the things that CAME i

148
00:08:05,040 --> 00:08:07,279
felt came out of build was this sense that nobody's

149
00:08:07,319 --> 00:08:10,399
dominating in the space at. All, yeah there's lots of different,

150
00:08:10,439 --> 00:08:12,759
models there's lots of different. Tools the fact that everybody

151
00:08:12,839 --> 00:08:15,839
jumped ON mcp is. Clear nobody has the confidence to, go,

152
00:08:15,879 --> 00:08:18,079
oh we'll go it. Alone we're all trying to find

153
00:08:18,120 --> 00:08:20,560
standards in inter interapt with each. Other. Yeah BUT i

154
00:08:20,600 --> 00:08:23,959
mean even EVEN mcp has. Limitations AND i have thoughts on,

155
00:08:24,000 --> 00:08:26,319
that And i'm tired And i'm just gonna say WHAT i.

156
00:08:26,360 --> 00:08:30,000
Think Oh, NO mtcp definitely has some. Issues there's no

157
00:08:30,040 --> 00:08:32,120
two ways about. That but let me wrap up With john. Here,

158
00:08:32,159 --> 00:08:33,480
john thanks you so much for your, comment and the

159
00:08:33,559 --> 00:08:35,039
copy of Music coba is on its way to. You

160
00:08:35,039 --> 00:08:36,519
and if you'd like a copy Of Music, COBE i

161
00:08:36,519 --> 00:08:38,519
write a comment on the website at dot NetRocks dot

162
00:08:38,559 --> 00:08:40,559
com or on the. Facebooks we publish every show, there

163
00:08:40,559 --> 00:08:41,759
and if you comment there and every read it on the,

164
00:08:41,759 --> 00:08:44,600
show we'll send you a copy Of Music. Koba And music.

165
00:08:44,320 --> 00:08:46,840
Speaker 1: To Code by is still going. Strong twenty two tracks

166
00:08:46,840 --> 00:08:50,480
to help you get in the zone and focus twenty

167
00:08:50,480 --> 00:08:53,639
five minutes each to coincide with The pomodor, method and

168
00:08:53,720 --> 00:08:54,240
they will help.

169
00:08:54,279 --> 00:08:57,080
Speaker 2: YOU i actually bought the first. Ones yeah a long

170
00:08:57,080 --> 00:08:57,600
times years.

171
00:08:57,600 --> 00:09:00,879
Speaker 1: Ago, yeah it was only three AND i did A,

172
00:09:00,960 --> 00:09:02,159
KICKSTARTER i did.

173
00:09:02,159 --> 00:09:05,360
Speaker 2: THREE i got the. THREE i still have THE mp

174
00:09:05,480 --> 00:09:08,559
three is somewhere in my junk drawer of one.

175
00:09:08,639 --> 00:09:10,919
Speaker 1: Drive, well now you can get the whole collection IN

176
00:09:11,080 --> 00:09:14,080
mp three flak en. Wave so you can go to

177
00:09:14,200 --> 00:09:16,320
music To code by dot net for that or you,

178
00:09:16,360 --> 00:09:18,279
know you could just send us a comment or leave a.

179
00:09:18,320 --> 00:09:20,559
Speaker 2: Comment, yeah, absolutely and reading on the show you'll get

180
00:09:20,559 --> 00:09:21,240
a copy of MUSIC o.

181
00:09:21,320 --> 00:09:24,080
Speaker 1: Bye, yeah all, Right So i'm going to introduce.

182
00:09:23,679 --> 00:09:27,600
Speaker 2: You for real areas here we go That i'm introducing.

183
00:09:27,639 --> 00:09:30,200
Speaker 1: It, yeah that voice that you've been listening to the

184
00:09:30,759 --> 00:09:33,600
can who it is Is Seph wares and he works

185
00:09:33,600 --> 00:09:37,200
in THE ai platform group At. Microsoft he received his bachelor's.

186
00:09:37,200 --> 00:09:39,159
Degree that's, it that's your. Title Now i'm going to

187
00:09:39,159 --> 00:09:41,720
talk about your education because that's the part of your.

188
00:09:41,759 --> 00:09:45,600
Bio we see his bachelors and computer science at you

189
00:09:46,039 --> 00:09:49,879
n L v saying place that Guy fieri got his.

190
00:09:49,960 --> 00:09:54,440
Speaker 2: Degree we were. Partners he's a he's a machine learning master.

191
00:09:54,440 --> 00:09:58,399
Speaker 1: Absolutely with a minor in mathematics and complete his master's

192
00:09:58,519 --> 00:10:00,360
great at The university Of utah and the field of

193
00:10:00,399 --> 00:10:04,080
computer science And. SETH i have a picture that BECAUSE

194
00:10:04,080 --> 00:10:08,200
i was in the front row at the build AND

195
00:10:08,240 --> 00:10:09,240
i took a picture of.

196
00:10:09,279 --> 00:10:12,000
Speaker 2: You oh what WAS i? Doing? Well you you opened the.

197
00:10:12,080 --> 00:10:15,559
Speaker 1: Show you came out and, SAID i Am seth and you,

198
00:10:15,600 --> 00:10:18,600
know with with somebody, else AND i DON'T i didn't recognize.

199
00:10:18,639 --> 00:10:20,960
Her BUT i took a picture of you two and

200
00:10:21,039 --> 00:10:24,240
here it. Is oh my, gosh That's. Kadasha, yeah she

201
00:10:24,440 --> 00:10:27,279
is delightful human. Beings SO i sent this To, kelly my,

202
00:10:27,360 --> 00:10:29,799
wife you, know, yes of course she actually met you

203
00:10:29,879 --> 00:10:30,600
BEFORE i met.

204
00:10:30,639 --> 00:10:33,519
Speaker 2: You, yes she did, so and, she.

205
00:10:33,519 --> 00:10:35,279
Speaker 1: You, know she says back to me and she, says

206
00:10:35,960 --> 00:10:38,600
that's a great. Picture but who is that black lady

207
00:10:38,600 --> 00:10:39,039
With Ricky?

208
00:10:39,120 --> 00:10:47,320
Speaker 2: Martin but, WOW i do get more. Beautiful AND i, said,

209
00:10:47,440 --> 00:10:49,360
again this is the medium for me to say these

210
00:10:49,440 --> 00:10:53,360
kinds of. THINGS i am getting more. Beautiful look at the.

211
00:10:53,399 --> 00:10:56,759
Speaker 1: Picture your name is not on the, screen it's it's her. Last,

212
00:10:57,360 --> 00:11:00,639
yeah so she wouldn't know and she didn't recognize.

213
00:11:00,679 --> 00:11:02,759
Speaker 2: You that is so. Funny then WHEN.

214
00:11:02,679 --> 00:11:05,240
Speaker 1: I told, her she, says, oh, yes that is.

215
00:11:05,720 --> 00:11:08,519
Speaker 2: YEAH i actually owe her a. Call she called me

216
00:11:08,559 --> 00:11:11,039
the other day AND i haven't called. Back, oh, okay she's,

217
00:11:11,120 --> 00:11:14,759
delightful she. Called she checks up on. Me she does. That, yeah,

218
00:11:14,960 --> 00:11:17,559
yeah so. Nice well you do need. Supervision it is.

219
00:11:17,639 --> 00:11:20,519
True my wife has had me now for twenty two ish,

220
00:11:20,559 --> 00:11:23,799
years and so during THE covid she may be. Tired

221
00:11:24,000 --> 00:11:24,399
during THE.

222
00:11:24,360 --> 00:11:27,120
Speaker 1: Covid, times she reached out to friends of mine and

223
00:11:27,159 --> 00:11:29,159
hers that she hadn't talked to in years and, Said,

224
00:11:29,159 --> 00:11:32,200
hey you know, What i'm just calling BECAUSE i want

225
00:11:32,240 --> 00:11:32,679
to check.

226
00:11:32,720 --> 00:11:34,320
Speaker 2: In it was.

227
00:11:34,320 --> 00:11:37,200
Speaker 1: GREAT i don't want, anything you, KNOW i just wanted to, Say, Hi.

228
00:11:37,639 --> 00:11:40,399
Speaker 2: GOSH i wish the niceness would rub off on other. People,

229
00:11:40,759 --> 00:11:45,759
yeah it's not working. Right all.

230
00:11:45,840 --> 00:11:49,759
Speaker 1: Right so we're talking about agentic, hey AND i got

231
00:11:49,759 --> 00:11:52,279
to tell, You i'm blown away by WHAT i saw.

232
00:11:52,320 --> 00:11:55,360
HERE i, MEAN i know that there have been other

233
00:11:55,559 --> 00:11:58,600
agents and products that work with agency and all that,

234
00:11:58,639 --> 00:12:01,639
stuff and the guys From manet have been into, it,

235
00:12:01,960 --> 00:12:04,559
right but they all said the same. Thing it's, Like,

236
00:12:04,720 --> 00:12:07,600
Wow microsoft is kind of catching up to what these

237
00:12:07,639 --> 00:12:11,080
guys have been. Doing AND i hope they dominate because

238
00:12:11,720 --> 00:12:14,320
they have the, power, Right microsoft has the power Of

239
00:12:14,360 --> 00:12:18,600
azure and all of these. Things and WHEN i learned ABOUT,

240
00:12:18,679 --> 00:12:21,440
mcp we've been talking about anybody who's listened to the

241
00:12:21,519 --> 00:12:26,399
last six, shows oh, yeah knows about this about the,

242
00:12:28,159 --> 00:12:29,120
okay is it.

243
00:12:29,440 --> 00:12:31,720
Speaker 2: Model model context context.

244
00:12:31,759 --> 00:12:34,759
Speaker 1: Protocol so this is the protocol by which agents can

245
00:12:34,799 --> 00:12:38,879
talk to your resources to do things like talk to

246
00:12:38,879 --> 00:12:41,240
YOUR sql, server talk To azure resources and all that.

247
00:12:41,240 --> 00:12:43,200
Stuff and you can just tell it to go do, things,

248
00:12:43,279 --> 00:12:44,639
yes and it will figure them.

249
00:12:44,639 --> 00:12:44,840
Speaker 2: Out.

250
00:12:44,960 --> 00:12:48,399
Speaker 1: Yes the combination of AN llm like get Hub copilot

251
00:12:48,720 --> 00:12:53,720
AND mcp connections to other resources just blows my. Mind

252
00:12:53,799 --> 00:12:54,519
it changes the.

253
00:12:54,559 --> 00:12:57,759
Speaker 2: Game, Yeah i'd love to get more low level, though

254
00:12:57,840 --> 00:13:01,759
so folks have an understanding of what an agent actually

255
00:13:01,919 --> 00:13:06,519
is because there is This what happened was and you

256
00:13:06,600 --> 00:13:09,720
were there with me for fifteen. Years i've been in

257
00:13:09,720 --> 00:13:13,399
this space for fifteen. Years nobody, cared and then we

258
00:13:13,480 --> 00:13:16,960
flip from nobody cared to this stuff is like jungle,

259
00:13:17,039 --> 00:13:19,360
magic or this stuff is like going to throw you

260
00:13:19,399 --> 00:13:20,480
into outer space or.

261
00:13:20,759 --> 00:13:22,759
Speaker 1: So we didn't have to create the. Models, yeah that

262
00:13:22,879 --> 00:13:23,480
was the hard.

263
00:13:23,279 --> 00:13:25,919
Speaker 2: Thing it's. True but but what happened is we kind

264
00:13:25,919 --> 00:13:28,759
of missed the in between part Of, hey let's explain

265
00:13:28,799 --> 00:13:32,120
what these things are doing and why it's actually. Useful

266
00:13:32,840 --> 00:13:36,039
because there's a THERE'S i don't want. To as you,

267
00:13:36,159 --> 00:13:39,759
know we had SEVERAL ai winters because of the over,

268
00:13:39,960 --> 00:13:43,559
hype AND i AM i am concerned that if we

269
00:13:43,720 --> 00:13:46,519
overhype too much that there's going to be a correction

270
00:13:46,840 --> 00:13:48,960
when there really doesn't need to be because it actually

271
00:13:49,000 --> 00:13:54,039
this time does do qualitatively cool. Stuff, yes and that's

272
00:13:54,039 --> 00:13:56,600
What i'm worried. ABOUT i, mean the winters were mostly

273
00:13:56,600 --> 00:13:58,720
induced by the fact that there was a limited utility

274
00:13:58,759 --> 00:14:01,879
what was being made, exactly and there was a limited

275
00:14:02,039 --> 00:14:05,080
of computational power to actually make the model the thing

276
00:14:05,120 --> 00:14:07,960
we wanted them. To hinton's famous papers actually the paper

277
00:14:08,480 --> 00:14:11,080
talking about backpropagation SAYING i can't do with this equipment

278
00:14:11,120 --> 00:14:13,519
in like nineteen eighty four mm. Hmmm and then In

279
00:14:13,600 --> 00:14:17,159
suskovar does off that paper in two thousand and nine

280
00:14:17,279 --> 00:14:19,759
and they start working on the image net. Problem, yes

281
00:14:20,080 --> 00:14:23,240
and those models are quite simple as compared to the

282
00:14:23,240 --> 00:14:25,440
ones that are now FOR. LMS i, MEAN i can

283
00:14:25,480 --> 00:14:28,440
get into the architecture and stuff of. It they're just much.

284
00:14:28,480 --> 00:14:31,639
Bigger and the thing about the thing about neural networks

285
00:14:31,639 --> 00:14:36,919
of that size Is i've spoken To Turing award winners about,

286
00:14:36,960 --> 00:14:39,679
like how do you shape these actual? Models to do

287
00:14:39,720 --> 00:14:42,120
the things you, Want and the reality is we don't

288
00:14:42,159 --> 00:14:46,960
really have a good understanding or. Explanation but what happens

289
00:14:47,039 --> 00:14:51,320
is they're innovations that come out like a transformer model

290
00:14:51,360 --> 00:14:55,720
with attention that was really designed to do translation that,

291
00:14:55,919 --> 00:15:02,000
somehow when you made it, bigger learned language, well presented

292
00:15:02,080 --> 00:15:05,200
language in a form that you could anthropomorphize. Correct and

293
00:15:05,240 --> 00:15:09,759
that's that's the Part richard does not like at. All

294
00:15:10,120 --> 00:15:12,360
the thing about it is is WHEN i picture these

295
00:15:12,399 --> 00:15:15,360
things in my, head it's like the matrix for, me

296
00:15:15,399 --> 00:15:19,879
Where i'm picturing like there's, weights multiplying vectors and then

297
00:15:20,000 --> 00:15:23,720
going over some kind of smoothing function to, remove you,

298
00:15:23,720 --> 00:15:27,679
know the linearities so that these scenes can predict complex.

299
00:15:27,759 --> 00:15:30,440
Things but in the, end all these things are doing

300
00:15:30,840 --> 00:15:34,919
is you're getting a string of words that are broken

301
00:15:35,000 --> 00:15:38,120
up into vectors that get passed in and multiplied by

302
00:15:38,159 --> 00:15:40,720
the transformer model to figure out how these things are,

303
00:15:40,759 --> 00:15:45,840
related pushed through a giant neural network multiple, times and

304
00:15:45,919 --> 00:15:49,279
the output is a giant vector of the entire space

305
00:15:49,600 --> 00:15:52,759
with a probably distribution that says which token should you

306
00:15:52,799 --> 00:15:55,600
put out? Next so when it looks like it's, typing

307
00:15:55,720 --> 00:15:59,200
it isn't it's just going back through the, model and

308
00:15:59,240 --> 00:16:02,399
that's what these things are. Doing and as, programmers the

309
00:16:02,440 --> 00:16:05,759
only thing we can control is what goes in the

310
00:16:05,799 --> 00:16:10,480
front the. Prompt and that's the beefy, bit right that

311
00:16:10,559 --> 00:16:13,759
people don't. Understand at that, point when we first started

312
00:16:13,799 --> 00:16:17,919
doing this, stuff it was magical for. Chat, Yeah but

313
00:16:17,960 --> 00:16:21,840
then early, on about maybe like six months into, it

314
00:16:21,919 --> 00:16:24,600
folks were starting to use these things for something that

315
00:16:24,639 --> 00:16:27,279
we liked to THAT i like to call intent. Mapping

316
00:16:27,759 --> 00:16:30,360
Cassie brevy AND i worked together for a couple of

317
00:16:30,440 --> 00:16:32,600
years and that's what we called, it where we would

318
00:16:32,600 --> 00:16:35,480
make THE lm like look at a block of text and,

319
00:16:35,519 --> 00:16:39,720
say is this question a support or a sales? Question

320
00:16:40,639 --> 00:16:43,600
and it was at that point that these, things because

321
00:16:43,639 --> 00:16:47,159
we allowed them to decide is it a support or

322
00:16:47,159 --> 00:16:50,879
a sales? Question when the code, branched those things all

323
00:16:50,919 --> 00:16:53,679
of a sudden were imbued with this notion OF i

324
00:16:53,919 --> 00:16:57,360
can make a CHOICE lm on what this thing. Is

325
00:16:57,679 --> 00:17:02,559
and at that point those things became, agents, yes and

326
00:17:02,559 --> 00:17:05,720
and and because we trusted them to answer our. Questions. Yeah.

327
00:17:06,839 --> 00:17:09,400
Yeah but but but nottice that WHEN i was doing this,

328
00:17:09,480 --> 00:17:13,720
ENGINEERING i narrowed the task to very specific you will

329
00:17:13,799 --> 00:17:18,920
say either support or, sales and that's. It and what

330
00:17:19,000 --> 00:17:21,359
that did is it allowed me to create a prompt

331
00:17:21,680 --> 00:17:24,119
that made it so that any user input that we

332
00:17:24,240 --> 00:17:27,400
put in, it it would tell me support or sales every.

333
00:17:27,440 --> 00:17:31,680
Time and my code had an if statement if support do,

334
00:17:31,799 --> 00:17:34,920
this if sales do, this, right and that at that,

335
00:17:34,960 --> 00:17:38,640
point that is the most simplistic way of thinking about an.

336
00:17:38,720 --> 00:17:42,319
Agent when AN lm has the has the ability to

337
00:17:42,440 --> 00:17:45,039
choose reason overset of data and then act on the

338
00:17:45,039 --> 00:17:50,240
reasoning you've. Generated and effectively that became a squishy programming

339
00:17:50,359 --> 00:17:56,160
control structure and that's what an agent is. Fundamentally but you're,

340
00:17:56,200 --> 00:17:58,039
like but it's got to be more than. That seth

341
00:17:58,119 --> 00:18:00,680
what about the tools? Stuff at that people are, like,

342
00:18:00,720 --> 00:18:04,680
well let's do structured output so that it could output adjacent.

343
00:18:04,759 --> 00:18:07,960
Thing and if we can output adjacent, thing why can't

344
00:18:07,960 --> 00:18:12,319
we output adjacent Thing that says this function with these.

345
00:18:12,359 --> 00:18:16,359
Parameters and at that point those things became tools and

346
00:18:16,400 --> 00:18:20,200
now instead of supporter, sales it, said run this function

347
00:18:20,839 --> 00:18:25,720
with these. Parameters and that is the extent of all

348
00:18:25,799 --> 00:18:28,519
of the, magic sauce of all of anything.

349
00:18:28,559 --> 00:18:31,720
Speaker 1: Agentic NOW i want to say something Here, seth you

350
00:18:32,000 --> 00:18:35,359
are among some of the most brilliant people That i've ever.

351
00:18:35,400 --> 00:18:36,240
Speaker 2: Met, oh thank.

352
00:18:36,279 --> 00:18:38,359
Speaker 1: You richard is one of. Them you are one of.

353
00:18:38,400 --> 00:18:40,680
Them but the other people are like dev express. People,

354
00:18:40,759 --> 00:18:44,720
Yeah Mark miller And Richard morris like these guys and. You,

355
00:18:44,960 --> 00:18:47,279
NOW i want to tell the listeners if you don't

356
00:18:47,279 --> 00:18:50,880
know Who seth. Is, seth you had an open SOURCE

357
00:18:51,079 --> 00:18:57,240
AI mL library Before microsoft hired. You, yeah before they

358
00:18:57,319 --> 00:19:01,319
Had AZURE. Ai you had this open source. Library we

359
00:19:01,400 --> 00:19:03,240
did a thing about it on D NR tv that

360
00:19:03,279 --> 00:19:03,680
showed it to.

361
00:19:03,680 --> 00:19:05,920
Speaker 2: ME i didn't understand. It, yeah of COURSE i kind of.

362
00:19:05,920 --> 00:19:08,880
Did but but you were doing.

363
00:19:08,960 --> 00:19:12,400
Speaker 1: That you've been doing this longer Than microsoft has been doing.

364
00:19:12,200 --> 00:19:14,400
Speaker 2: It way before it was. COOL i was doing before

365
00:19:14,400 --> 00:19:18,279
it was. Cool the secret, sauce, though is machine learning

366
00:19:18,319 --> 00:19:22,680
has been happening since nineteen fifty. Six Frank rosenblatt The,

367
00:19:22,680 --> 00:19:24,960
perceptron which is the basic foundation OF.

368
00:19:25,119 --> 00:19:28,440
Speaker 1: Ski so that was the last, show righteen Fifty.

369
00:19:28,359 --> 00:19:32,839
Speaker 2: Oh, hey, yeah let's call back To Yesterye, frank we've missed.

370
00:19:32,839 --> 00:19:35,839
Now but the reality is That microsoft's been DOING ai

371
00:19:36,240 --> 00:19:38,759
in the form of machine learning for a very long.

372
00:19:38,799 --> 00:19:41,720
Time WHAT i was doing IS i was trying to

373
00:19:41,720 --> 00:19:45,799
bring it to the dot net community BECAUSE i recognize

374
00:19:45,880 --> 00:19:49,720
that there is a space for algorithms which are learned

375
00:19:49,799 --> 00:19:52,960
instead of, programmed AND i thought that IF i could

376
00:19:53,000 --> 00:19:55,160
do a. Thing OBVIOUSLY i haven't kept up on that

377
00:19:55,319 --> 00:19:57,319
for a while because it was just. Me BUT i

378
00:19:57,319 --> 00:19:59,680
implement a whole set of stuff to make that. Easier

379
00:20:00,160 --> 00:20:02,519
but my goal has been for the last you, know

380
00:20:02,599 --> 00:20:05,519
fifteen twenty, YEARS i want people to know exactly what

381
00:20:05,559 --> 00:20:08,440
these things are, doing so that with, confidence you can

382
00:20:08,480 --> 00:20:09,799
make them do even cooler.

383
00:20:09,839 --> 00:20:13,960
Speaker 1: Things, right, yeah, Yeah and so now you're At, microsoft

384
00:20:13,960 --> 00:20:15,079
you're working on THE ai.

385
00:20:15,160 --> 00:20:19,400
Speaker 2: Stuff that's The azure. Stuff that's. Right The Italians channel,

386
00:20:19,480 --> 00:20:22,200
Nine oh, yeah, No channel nine was. Great you had

387
00:20:22,200 --> 00:20:24,000
such a good. TIME i really. GOOD i love. That

388
00:20:24,039 --> 00:20:25,799
and then somebody figured out actually knew them all, stuff

389
00:20:25,799 --> 00:20:27,440
and it was. Important it's, like you come with, me you,

390
00:20:27,480 --> 00:20:29,720
know CAN i be? HONEST i ACTUALLY I actually it

391
00:20:29,799 --> 00:20:32,319
was different than that because AS i started doing The

392
00:20:32,359 --> 00:20:36,160
channel nine, stuff the reason WHY i knew what questions

393
00:20:36,279 --> 00:20:38,519
asked is BECAUSE i already knew their content AND i

394
00:20:38,559 --> 00:20:40,920
knew how to make them talk about and make things.

395
00:20:40,960 --> 00:20:44,519
Sing but people just started saying stuff to me, Like,

396
00:20:44,799 --> 00:20:46,720
seth don't you ever get tired of all this technical?

397
00:20:46,799 --> 00:20:49,880
Stuff And i'm, like oh, man MAYBE i should move

398
00:20:49,920 --> 00:20:53,640
into something more. Technical maybe. Maybe so THEN i moved

399
00:20:53,640 --> 00:20:57,880
into DevRel FOR aiml and Then John, montgomery who was

400
00:20:57,880 --> 00:21:01,480
A cvp Of young he was, Like, seth come work

401
00:21:01,559 --> 00:21:03,519
for me in THE ai core.

402
00:21:03,559 --> 00:21:06,920
Speaker 1: Group so that's what you obviously knew your, pedigree, right

403
00:21:06,960 --> 00:21:07,559
they knew they.

404
00:21:07,640 --> 00:21:10,240
Speaker 2: Did nobody, knew LIKE i THINK i think people record

405
00:21:10,279 --> 00:21:14,240
like see me and like Because i'm pretty jovial and

406
00:21:14,440 --> 00:21:17,440
and BUT I i.

407
00:21:16,680 --> 00:21:18,440
Speaker 1: Think you don't come across as AN ai.

408
00:21:18,559 --> 00:21:21,720
Speaker 2: Scientist, NO i KNOW i do. Not and but if

409
00:21:21,720 --> 00:21:24,000
you looked at, me obviously we've said this many, Times

410
00:21:24,039 --> 00:21:27,440
i'm very beautiful in, appearance and people would not Like mark.

411
00:21:27,720 --> 00:21:31,039
Exactly people wouldn't, know BUT I i really have been

412
00:21:31,079 --> 00:21:34,599
doing machine learning. Forever and, again my primary goal is

413
00:21:34,599 --> 00:21:38,000
for people to understand at a profound level what these things.

414
00:21:38,039 --> 00:21:41,559
Are and to, me they are basically a new programming

415
00:21:41,599 --> 00:21:45,200
control structure which can Take english and point you to

416
00:21:45,359 --> 00:21:50,200
what to run. Next, yeah fairly still Constrained, english, yes, well,

417
00:21:50,240 --> 00:21:52,880
no even Regular english with AN, lm because if you

418
00:21:52,880 --> 00:21:55,079
can you can type into AN ll m like Bad.

419
00:21:55,119 --> 00:21:58,279
English and for, EXAMPLE i did a demo at my

420
00:21:58,440 --> 00:22:01,519
talk two days ago WHERE i showed him the supporter

421
00:22:01,640 --> 00:22:04,519
sales thing AND i, said my tent is, broken, yo

422
00:22:04,920 --> 00:22:08,079
AND i spelled it wrong and it said support AND

423
00:22:08,119 --> 00:22:10,880
i was, like what's the borst tent in the, world?

424
00:22:11,000 --> 00:22:13,480
Yo and it was like burst and not b e

425
00:22:13,640 --> 00:22:16,200
r S t BECAUSE i fat fingered it. Right and

426
00:22:16,240 --> 00:22:21,680
then it was, like, uh. Sales notice that because it's

427
00:22:21,759 --> 00:22:25,039
a large language, model it's able to reason over the

428
00:22:25,079 --> 00:22:29,599
text and then create the output that we. Want and

429
00:22:29,680 --> 00:22:31,920
if that output is a, function then it tells you

430
00:22:31,920 --> 00:22:33,599
what to run. Next so it.

431
00:22:33,559 --> 00:22:37,039
Speaker 1: Didn't understand, bursts so it figured that you really meant worst.

432
00:22:36,960 --> 00:22:40,599
Speaker 2: A best, Best, yeah, okay best b e r S.

433
00:22:40,680 --> 00:22:43,440
T and it was, like even if it was, worst

434
00:22:43,559 --> 00:22:47,440
it's still a sales and it still was able to

435
00:22:47,440 --> 00:22:50,519
figure it. Out and and that was with me, again

436
00:22:50,759 --> 00:22:53,839
and THEN i. CAN'T i don't know how to emphasize

437
00:22:53,839 --> 00:22:56,799
this is. Enough, basically you get a control structure that

438
00:22:56,839 --> 00:22:58,799
gives You english and tells you what things run. Next

439
00:22:59,039 --> 00:23:02,200
WHAT mcp does and whether it does it well or

440
00:23:02,279 --> 00:23:04,799
not is neither here nor. THERE i don't think it's

441
00:23:04,839 --> 00:23:09,000
Necessary what it does is it allows you to query

442
00:23:10,240 --> 00:23:14,000
what functions can you? Call and what it does is

443
00:23:14,039 --> 00:23:17,400
it maps those into the tool calls that are available

444
00:23:17,640 --> 00:23:19,440
to the L, m and then it can call. It

445
00:23:19,519 --> 00:23:21,839
but you can do that With swagger AND. Rest, sure,

446
00:23:22,119 --> 00:23:23,039
yeah but it's.

447
00:23:23,079 --> 00:23:27,079
Speaker 1: Not it doesn't have the the almost it does.

448
00:23:27,119 --> 00:23:30,680
Speaker 2: Reasoning it, does but it does because the reasoning happens

449
00:23:30,680 --> 00:23:33,799
in THE. Lm and what it outputs is it outputs

450
00:23:33,839 --> 00:23:36,400
what functions Are swagger AND rest don't have that they,

451
00:23:36,440 --> 00:23:38,240
don't but you don't need it BECAUSE, mcp if you

452
00:23:38,279 --> 00:23:40,799
look at, it is basically Like. Swagger it tells you

453
00:23:40,839 --> 00:23:43,759
what tools are, available and it's LIKE, rest and it

454
00:23:43,759 --> 00:23:45,680
tells you how to call. Them, Right and the thing

455
00:23:45,720 --> 00:23:47,920
THAT i like about like those kind of and OPEN,

456
00:23:47,960 --> 00:23:50,880
api which has been NOW i think the Next swagger is, that,

457
00:23:50,960 --> 00:23:54,400
RIGHT i don't. Know but with OPEN api and WITH

458
00:23:54,640 --> 00:23:57,359
rest you're able to do the same things but with

459
00:23:57,440 --> 00:24:01,359
the benefit of like twenty years of security, handling with,

460
00:24:01,440 --> 00:24:06,880
headers WITH, ooth with token, negotiation et. Cetera and so that's,

461
00:24:06,920 --> 00:24:09,920
why like IF mcp gets you, going like WHEN i

462
00:24:10,000 --> 00:24:13,039
USE vb, six, right it got me. GOING i loved

463
00:24:13,039 --> 00:24:14,720
it because it was able to. Get but then when

464
00:24:14,720 --> 00:24:16,640
you were, LIKE i need to be more memory, CONSCIOUS

465
00:24:16,759 --> 00:24:19,599
i needed to be more. Careful then you can skip

466
00:24:19,720 --> 00:24:22,759
down a layer if you if you want. To but,

467
00:24:23,200 --> 00:24:25,799
AGAIN i love THAT mcp has gotten people thinking about.

468
00:24:25,799 --> 00:24:29,279
It but the reasoning doesn't happen WITH. Mcp, no it's,

469
00:24:30,119 --> 00:24:32,359
yeah it just tells you what function are. Next but

470
00:24:32,400 --> 00:24:35,119
the magic that makes it look good is once it

471
00:24:35,200 --> 00:24:38,000
runs the, function there's a thread that has like what

472
00:24:38,119 --> 00:24:40,599
the QUESTION i? Asked it asks for a Function i'm

473
00:24:40,599 --> 00:24:43,319
like pointing like people can. See and then when it executes,

474
00:24:43,359 --> 00:24:45,920
out it puts that function return call back on the

475
00:24:45,960 --> 00:24:48,640
thread and THE lm is called again with that on its,

476
00:24:48,680 --> 00:24:51,000
thread and then it comes with the. Answer and so

477
00:24:51,279 --> 00:24:56,039
that actually going out from THE ai space is effectively

478
00:24:56,119 --> 00:25:01,559
bridging the gap FROM ai reasoning to perative or functional. Code.

479
00:25:01,640 --> 00:25:04,160
Right how we MAKE sky? IT i get? It, NO

480
00:25:04,240 --> 00:25:05,720
i mean it's NOT i don't even think IT'S.

481
00:25:05,559 --> 00:25:09,680
Speaker 1: Sky and it may not be the best security wise right,

482
00:25:09,680 --> 00:25:11,680
now but it's clearly where everybody wants to.

483
00:25:11,680 --> 00:25:14,359
Speaker 2: Go, yeah it's because it simplifies. It but my sense

484
00:25:14,440 --> 00:25:17,519
is that as you see, This i'm the prognostica, here,

485
00:25:17,599 --> 00:25:20,200
right BECAUSE i like doing. That as you see this thing,

486
00:25:20,279 --> 00:25:24,880
maturing it's going to be indistinguishable from rest right and discovery.

487
00:25:24,880 --> 00:25:28,960
Protocols that it's the way the embracement OF mcp says to,

488
00:25:29,000 --> 00:25:31,440
me everybody wants to interrupt with everybody. ELSE i fact

489
00:25:31,440 --> 00:25:32,839
that we could come up with a better way to do,

490
00:25:32,920 --> 00:25:35,839
it and in fact already. Have we could just be

491
00:25:35,880 --> 00:25:38,480
publishing APIs for crying. Out, Yeah And i'm telling, you

492
00:25:38,480 --> 00:25:40,680
like we we had a phase At microsoft and bless

493
00:25:40,680 --> 00:25:43,640
Our hearts where we went full. Wisdole, yes and.

494
00:25:45,079 --> 00:25:47,759
Speaker 1: That's what The southern people say about the town drunk

495
00:25:48,039 --> 00:25:48,519
bless his.

496
00:25:48,759 --> 00:25:52,279
Speaker 2: Heart, yeah bless our. Heart But wisdole was the same,

497
00:25:52,599 --> 00:25:57,279
thing the same kind of desire to to abstract over.

498
00:25:57,559 --> 00:26:00,480
Primitives and then we went back to primitive still being

499
00:26:00,640 --> 00:26:06,759
WSD l Or Web Service Discovery language. Wisdom does that sound? Familiar,

500
00:26:06,799 --> 00:26:09,440
friends done done? Dumb and so What i'm telling you

501
00:26:09,599 --> 00:26:11,079
is that and some of the things you saw with

502
00:26:11,200 --> 00:26:13,400
N a Web the reason why this is such a cool.

503
00:26:13,519 --> 00:26:16,400
Speaker 1: Space if we haven't talked about AN a web on the,

504
00:26:16,400 --> 00:26:18,880
show oh, really yeah this WEEK i haven't.

505
00:26:18,880 --> 00:26:22,400
Speaker 2: EITHER i started looking at. It but the intuition here

506
00:26:22,519 --> 00:26:25,839
is if you can language do stuff and then call

507
00:26:25,960 --> 00:26:29,839
stuff and then language do stuff on the, Out like

508
00:26:29,960 --> 00:26:32,160
what you if you hook this up to the in

509
00:26:32,559 --> 00:26:35,559
language tool, stuff, Right you're? Fired, yeah you're my problem

510
00:26:35,599 --> 00:26:37,279
WITH nl. WEBS i keep, thinking why are you talking About?

511
00:26:37,319 --> 00:26:41,720
Dutch people's confused The dutch. Web but hopefully that helps

512
00:26:41,720 --> 00:26:44,599
BECAUSE i really want to emphasize that once you know

513
00:26:44,640 --> 00:26:46,200
that that's what it's doing at a low level and

514
00:26:46,240 --> 00:26:48,640
you play with it that, way it changes the way

515
00:26:48,680 --> 00:26:51,359
you think about These. SURE i had a journalist talk

516
00:26:51,440 --> 00:26:53,240
to me and, say, Hey, SETH i get a lot

517
00:26:53,240 --> 00:26:56,519
of questions about SHOULD i trust these? Things and my

518
00:26:56,640 --> 00:26:59,599
response surprised. HIM i, SAID i don't trust these. Things,

519
00:26:59,599 --> 00:27:02,000
no don't be. Silly that's WHY i program them the

520
00:27:02,039 --> 00:27:05,720
WAY i. DO i restrict what they have to. DO

521
00:27:06,559 --> 00:27:09,680
i restrict the tool calls that they, have and then

522
00:27:10,039 --> 00:27:12,599
when the tool calls are, CALLED i write the code

523
00:27:13,200 --> 00:27:16,480
that the tool that what the tool responds. With because

524
00:27:16,559 --> 00:27:19,519
all these things are effectively becoming part of the prompt

525
00:27:19,799 --> 00:27:23,519
for the next tokens to. Generate and that is when

526
00:27:23,559 --> 00:27:27,079
you get that, Intuition you're, like, oh that MEANS i

527
00:27:27,079 --> 00:27:29,640
can do this ANYWAY i want to ye.

528
00:27:30,240 --> 00:27:33,240
Speaker 1: And that's that's WHERE i think we've been talking about

529
00:27:33,240 --> 00:27:37,759
this this week, too that the attributes of the program

530
00:27:37,920 --> 00:27:42,319
or the future are going to be imaginations over technical

531
00:27:42,400 --> 00:27:47,759
probsolutely imagination mean and we may be held back a

532
00:27:47,759 --> 00:27:49,680
little bit by what we think we can do and

533
00:27:49,680 --> 00:27:52,559
what we can't. Do but somebody who doesn't have that

534
00:27:52,680 --> 00:27:56,880
kind of you, know, limitation self, limitation will be able

535
00:27:56,920 --> 00:27:57,920
to do amazing.

536
00:27:58,000 --> 00:28:00,759
Speaker 2: Things. Yeah like if you watch the keynote on day,

537
00:28:00,759 --> 00:28:05,440
two you Saw Amanda foster And elijah on stage talking

538
00:28:05,519 --> 00:28:08,839
to this thing that it was executing a bunch of

539
00:28:08,839 --> 00:28:11,279
stuff on their. BEHALF i think it was. Software it

540
00:28:11,319 --> 00:28:13,839
was literally software BECAUSE i KNOW i, WROTE i wrote

541
00:28:13,880 --> 00:28:16,359
THE i wrote the website and the back. End amanda

542
00:28:16,400 --> 00:28:18,920
worked on the, agent And elijah worked on the on the.

543
00:28:19,000 --> 00:28:22,480
Evaluations but the thing about it is it was actually running.

544
00:28:22,519 --> 00:28:25,559
Live that thing was actually happening. Live and one of

545
00:28:25,640 --> 00:28:27,400
the things that we had is once you get into

546
00:28:27,480 --> 00:28:29,839
a single, agent it makes, sense but when you get

547
00:28:29,839 --> 00:28:32,400
into multi, agent you wonder, like, well what is the

548
00:28:32,480 --> 00:28:36,880
interface between one agent and? Another and the surprising thing

549
00:28:37,000 --> 00:28:41,880
is that the interface is actually language, weird which is, weird,

550
00:28:42,039 --> 00:28:44,640
Right so for, Example i'll give you, example the voice

551
00:28:44,640 --> 00:28:46,519
agent that we, made and if you watch the, keynote

552
00:28:46,519 --> 00:28:48,839
you'll see her talking to. It the voice agent is

553
00:28:48,920 --> 00:28:51,839
using THE gpt four oh Real TIME, api which has

554
00:28:51,880 --> 00:28:54,839
the also has the ability to issue function. Calls, Okay

555
00:28:55,000 --> 00:28:57,319
and so what we did is we collected the list

556
00:28:57,559 --> 00:29:01,359
of azure Ai foundry agents that we made Andree foundry

557
00:29:01,720 --> 00:29:04,480
AND i just loaded them up and then she picked

558
00:29:04,480 --> 00:29:07,240
the ones she wanted because she was working on. Them

559
00:29:07,240 --> 00:29:09,720
and what it did is it mapped those other agents

560
00:29:10,279 --> 00:29:14,119
as tool calls interesting to the voice. Agent but you're, like,

561
00:29:14,160 --> 00:29:16,920
well what did you pass to the? Agent, well the

562
00:29:17,000 --> 00:29:20,960
agents themselves when you run, them have this property called additional,

563
00:29:21,000 --> 00:29:24,680
instructions which means you can augment their system prompt and

564
00:29:24,759 --> 00:29:27,240
then that's. It that's all you can. Do so WHAT

565
00:29:27,279 --> 00:29:30,640
i did IS i mapped all of those agents as tool,

566
00:29:30,720 --> 00:29:33,440
calls gave it a function. Name you're back to sales

567
00:29:33,440 --> 00:29:36,000
and support, again that's all this, IS i. Know and

568
00:29:36,039 --> 00:29:40,039
THEN i, said you passed voice agent additional instructions and

569
00:29:40,240 --> 00:29:43,279
ask it for what you wanted to. Do so the

570
00:29:43,359 --> 00:29:47,759
voice agent then is taking the initial requests and assessing

571
00:29:47,839 --> 00:29:49,519
it to go which agent to go, to back to

572
00:29:49,559 --> 00:29:51,880
your sales and. Support that's. Right then taking the portion

573
00:29:51,960 --> 00:29:54,440
of the requests that makes sense for that, agent plus

574
00:29:54,480 --> 00:29:57,640
the additional, instruction let's send it on. Down but it's

575
00:29:57,680 --> 00:30:00,519
even more than. That it actually asks, it can you

576
00:30:00,599 --> 00:30:03,079
help me with? This and we were me And amanda

577
00:30:03,160 --> 00:30:05,000
because it's like in the two in the, morning we're

578
00:30:05,000 --> 00:30:07,039
looking at this, stuff like because we're on this stuff all.

579
00:30:07,119 --> 00:30:09,200
Night we looked at the law because we Ad jason

580
00:30:09,279 --> 00:30:14,920
printed out everything and we could, see, uh image agent system,

581
00:30:14,920 --> 00:30:17,359
instructions you are to generate a blah blah blah. Question

582
00:30:17,480 --> 00:30:21,000
can you generate an? Image it was Actual english and

583
00:30:21,039 --> 00:30:23,680
we were, like holy, cow this is this is amazing

584
00:30:23,880 --> 00:30:26,640
and the software being polite to. Software, yes it.

585
00:30:26,559 --> 00:30:30,480
Speaker 1: Was really cook but but it's nothing. New if you

586
00:30:30,480 --> 00:30:34,640
look at the S MT smtp protocol when you connect

587
00:30:34,680 --> 00:30:38,480
When i'm When gmail connects to an S mvp, server it,

588
00:30:38,599 --> 00:30:41,960
says you, know, hello And I'm Carl, Franklin carl frank

589
00:30:42,640 --> 00:30:47,000
and THE smtp server actually writes, back, Hello Carl, franklin

590
00:30:47,440 --> 00:30:48,240
nice to meet.

591
00:30:48,279 --> 00:30:50,640
Speaker 2: You, yes but we're not talking part of the. Protocol

592
00:30:50,799 --> 00:30:53,480
it's Not eliza in This, NO i, KNOW.

593
00:30:53,519 --> 00:30:55,759
Speaker 1: I, know but but you know that those kind of

594
00:30:55,799 --> 00:30:58,160
friendly things have been.

595
00:30:57,519 --> 00:30:59,640
Speaker 2: Protocols for a long. Time but the difference is that

596
00:30:59,680 --> 00:31:02,440
the only things are being generated by the through the

597
00:31:02,480 --> 00:31:05,279
agency of AN. LLM i get. It it's, big big.

598
00:31:05,519 --> 00:31:12,200
CHANGE i do question if if if the politeness affects

599
00:31:12,240 --> 00:31:16,400
the prompt. Results, yes it absolutely. Does And i'll tell.

600
00:31:16,440 --> 00:31:18,880
You i'll tell you what happens And i'll tell. You

601
00:31:19,240 --> 00:31:21,480
so there was a part of the demo Where amanda

602
00:31:21,720 --> 00:31:25,960
generates an image for a hackathon and they AND i

603
00:31:25,960 --> 00:31:28,559
don't think they understood because someone backstage is, like, hey

604
00:31:28,880 --> 00:31:32,359
can you change the image to something more? Happy and we're, like,

605
00:31:33,279 --> 00:31:36,039
no it generates it every. Time it's it's doing it

606
00:31:36,079 --> 00:31:38,640
on its. Own and a man is, like but we

607
00:31:38,720 --> 00:31:42,440
can change it by telling, it please make the image more,

608
00:31:42,720 --> 00:31:48,319
happy and like we made that change in The english

609
00:31:48,359 --> 00:31:50,839
and the and not in the voice controller because that's

610
00:31:50,839 --> 00:31:52,559
the thing we talked, to but in the literal little

611
00:31:52,559 --> 00:31:55,720
agent the, customer and it called, it called. It and

612
00:31:55,759 --> 00:31:58,400
then it was, like and now everyone's happy at the.

613
00:31:58,400 --> 00:32:01,079
Hackathon and and because everyone thought we were using some

614
00:32:01,160 --> 00:32:05,319
like pre baked image and, no it's not every, time

615
00:32:05,720 --> 00:32:08,039
and then we did, that and then it was really cool.

616
00:32:07,880 --> 00:32:09,319
Speaker 1: All, RIGHT i think we got to take a, break

617
00:32:09,519 --> 00:32:12,079
so we will be right back after these very important

618
00:32:12,079 --> 00:32:16,880
messages don't go. Away did you know there's a dot

619
00:32:16,920 --> 00:32:21,680
net on aws. Community follow the social media, blogs YouTube

620
00:32:21,720 --> 00:32:25,519
influencers and open source projects and add your own. Voice

621
00:32:26,079 --> 00:32:29,279
get plugged into the dot net on aws community at

622
00:32:29,319 --> 00:32:32,920
aws Dot amazon dot, com slash dot, net.

623
00:32:36,440 --> 00:32:36,799
Speaker 2: And we're.

624
00:32:36,839 --> 00:32:39,440
Speaker 1: Back it's dot net. Rocks I'm Carl, Franklin I'm Richard,

625
00:32:39,440 --> 00:32:42,359
campbell And Seth warez is. Here we are JUST i

626
00:32:42,759 --> 00:32:45,279
feel like we could be talking for, hours hours an,

627
00:32:45,279 --> 00:32:48,680
hour so much to talk, about but we're talking about

628
00:32:48,720 --> 00:32:54,359
agentic coding and. PROGRAMMING i Took microsoft Agent mode In

629
00:32:54,480 --> 00:32:57,119
Visual studio for a spin yesterday to go.

630
00:32:57,119 --> 00:32:58,440
Speaker 2: Good, okay it went really.

631
00:32:58,440 --> 00:33:02,119
Speaker 1: WELL i gave it some tasked to create some stuff

632
00:33:02,240 --> 00:33:05,039
THAT i knew how to do BECAUSE i wanted to.

633
00:33:05,039 --> 00:33:08,519
See but there was one very tricky thing that it

634
00:33:08,680 --> 00:33:11,839
was intuitive for it to, do but it didn't. Work

635
00:33:12,759 --> 00:33:17,240
so it's not like agent mode is going to compile

636
00:33:17,319 --> 00:33:18,160
your code.

637
00:33:17,880 --> 00:33:18,400
Speaker 2: And see if it.

638
00:33:18,440 --> 00:33:20,880
Speaker 1: Works it's just going to do what it thinks is. Right, yes,

639
00:33:21,079 --> 00:33:23,200
SORRY i didn't meet us the ord. Thing it's going

640
00:33:23,279 --> 00:33:26,279
to use what it assesses to be the correct. Answer

641
00:33:27,000 --> 00:33:29,920
so but WHEN i went and told, it, no that's.

642
00:33:29,960 --> 00:33:33,480
Wrong this doesn't work on this particular. Object you're going

643
00:33:33,519 --> 00:33:36,240
to have to subclass, this it, said oh. Okay AND

644
00:33:36,319 --> 00:33:38,759
i didn't tell it what to. Do ALL i had

645
00:33:38,759 --> 00:33:40,640
to say was you're going to have to subclass this,

646
00:33:40,680 --> 00:33:42,200
subject and it figured it.

647
00:33:42,240 --> 00:33:45,319
Speaker 2: Out it's pretty cool. Stuff one of the things that

648
00:33:45,640 --> 00:33:48,160
you're going to find is that if you ever have

649
00:33:48,240 --> 00:33:51,680
an agentic thing that is not quite doing what you

650
00:33:51,839 --> 00:33:55,720
want in, general from an engineering, perceptive for, me it

651
00:33:55,799 --> 00:33:59,480
means that they've overstuffed what they want a single agent to.

652
00:33:59,519 --> 00:34:05,839
Do so the actual engineering problem becomes agent specificity versus agent.

653
00:34:06,000 --> 00:34:09,800
Count and that's an engineering, issue, right because if you

654
00:34:09,840 --> 00:34:12,880
have too many, agents it might cost, more or it

655
00:34:12,960 --> 00:34:15,760
might it might work. Slower if you have too few,

656
00:34:15,880 --> 00:34:18,480
agents the recall isn't going to be as, good, Right

657
00:34:18,679 --> 00:34:20,960
and so the actual engineering problem is that it turns

658
00:34:20,960 --> 00:34:23,639
out the agent you use was. Able once you specify

659
00:34:23,760 --> 00:34:25,760
down a little, bit it made things. Better but if

660
00:34:25,760 --> 00:34:28,159
you think you're going to make a single agent with

661
00:34:28,239 --> 00:34:31,440
like a two page prompt with fifteen, tools it's not

662
00:34:31,480 --> 00:34:34,559
going to do the right. Thing and that's the trade

663
00:34:34,599 --> 00:34:36,519
off from an engineering pler because now that you know

664
00:34:36,880 --> 00:34:39,840
you have to make it, specific you have to limit

665
00:34:39,880 --> 00:34:42,800
its tool because it's an interesting switch from what build

666
00:34:42,880 --> 00:34:45,039
up lms the first, place which is making them bigger and.

667
00:34:45,039 --> 00:34:46,800
Bigger now we're, going, hey if you wanted me to respond.

668
00:34:46,880 --> 00:34:48,960
Better but see that's the thing, small that's a. Thing

669
00:34:49,159 --> 00:34:52,199
it's not THE lms that are getting, right it's the

670
00:34:52,360 --> 00:34:55,760
tasks that you're assigning THE. Lm so it's THE. Lm

671
00:34:55,840 --> 00:34:58,000
let's just fix THE lm for. One let's just SAY gbd. Four,

672
00:34:58,079 --> 00:35:01,880
oh like that models, Huge but like you are gonna

673
00:35:01,960 --> 00:35:05,320
narrow the agent's task to your job is to send

674
00:35:05,360 --> 00:35:08,920
an email to the right. Place and then here are

675
00:35:08,920 --> 00:35:12,760
the tools send email or delete email or. Whatever those

676
00:35:12,760 --> 00:35:15,159
are the only. Things that thing is gonna be very

677
00:35:15,199 --> 00:35:16,960
good at sending, emails and it's gonna be very good

678
00:35:17,000 --> 00:35:18,679
to be, like, HEY i need to send an emails

679
00:35:18,960 --> 00:35:22,760
to whom please type or tell me the. Email and

680
00:35:22,800 --> 00:35:25,719
then it's gonna be, like please validate the. Email what

681
00:35:25,760 --> 00:35:28,440
content do you? Want because it's trying to fill out

682
00:35:28,480 --> 00:35:31,800
the parametery of the function call because that's what the

683
00:35:31,880 --> 00:35:35,199
prompt wants it to. Do and that's what's. Happening so

684
00:35:35,280 --> 00:35:37,679
for your, case LIKE i worry like even in some

685
00:35:37,719 --> 00:35:39,719
of the tools that we, make maybe At, microsoft maybe

686
00:35:39,719 --> 00:35:43,679
they're too broad and that's where you find this notion of.

687
00:35:43,719 --> 00:35:47,400
Hallucination now people aren't gonna like, this but the elem's

688
00:35:47,519 --> 00:35:50,920
job is to actually make up. Tokens if you give

689
00:35:50,920 --> 00:35:53,559
it too broad a, thing it's gonna make stuff. Up

690
00:35:54,079 --> 00:35:56,280
if you get it to if you give it a narrow,

691
00:35:56,320 --> 00:35:58,800
thing it's gonna make the right stuff. Up BUT i

692
00:35:58,840 --> 00:35:59,519
thought that's.

693
00:35:59,360 --> 00:36:01,920
Speaker 1: What the point of the agent THAT i access in

694
00:36:02,000 --> 00:36:04,920
visual studio, is and it's NOT gpd four oh with

695
00:36:05,000 --> 00:36:08,920
this broad uh set of knowledge or. Whatever i'm, Sorry,

696
00:36:09,039 --> 00:36:11,360
RICHARD i didn't mean to use the. KNOWLEDGE i have

697
00:36:11,440 --> 00:36:14,400
to apologize every TIME i answerpomorphize than LL.

698
00:36:14,519 --> 00:36:16,760
Speaker 2: M but he is. Right thank you keep him on. Track, yeah,

699
00:36:16,760 --> 00:36:20,360
No i'm. SORRY i apologize now ALREADY i. Apologize please

700
00:36:20,360 --> 00:36:23,480
be with the pigmies In New. Guinea. Yeah.

701
00:36:23,519 --> 00:36:27,519
Speaker 1: Amen but it narrows the focus to truss the things

702
00:36:27,559 --> 00:36:31,159
that are in. Code so theoretically it should know everything

703
00:36:31,199 --> 00:36:34,119
about the Language i'm using the, markup you, know If

704
00:36:34,159 --> 00:36:37,280
i'm Using blazer or. Whatever so it's got to UNDERSTAND

705
00:36:37,400 --> 00:36:39,360
html and JavaScript and all Those you're.

706
00:36:39,239 --> 00:36:42,280
Speaker 2: Gonna love this. Answer it knows. Nothing it knows only

707
00:36:42,280 --> 00:36:44,679
what You i'm trained on. It, well, yeah so there

708
00:36:44,719 --> 00:36:46,280
there is a, yeah so that's WHAT i. Mean, Yeah

709
00:36:46,320 --> 00:36:48,679
so there's certain models that are actually Like, codex for,

710
00:36:48,679 --> 00:36:52,400
example is perfectly trained on, code but it doesn't know

711
00:36:53,360 --> 00:36:57,039
like in my, head it doesn't know the. Languages it

712
00:36:57,119 --> 00:37:00,480
knows that if you give it this sequence of programming or,

713
00:37:00,599 --> 00:37:03,760
questions it's going to return this. Sequence let me just

714
00:37:04,000 --> 00:37:04,800
add this to the.

715
00:37:04,880 --> 00:37:08,440
Speaker 1: EQUATION i did the same problem with chatchipt AND i

716
00:37:08,480 --> 00:37:11,639
did get the right. Answer, yes, so but the thing

717
00:37:11,719 --> 00:37:15,760
that was The Microsoft agent mode In Visual studio did

718
00:37:15,800 --> 00:37:17,320
not give me the right answer.

719
00:37:17,000 --> 00:37:17,960
Speaker 2: Because it's a different.

720
00:37:17,960 --> 00:37:21,679
Speaker 1: Model, Right but wouldn't it be narrowed down to just

721
00:37:21,760 --> 00:37:25,639
the programming stuff like why did CHAT gpt with vast

722
00:37:25,760 --> 00:37:29,159
programming and learning that it did on all this data

723
00:37:29,719 --> 00:37:31,159
provide a better answer.

724
00:37:30,960 --> 00:37:37,239
Speaker 2: Because models are stochastic, processes and the thing about one

725
00:37:37,280 --> 00:37:39,679
model versus another is one will do better with the

726
00:37:39,719 --> 00:37:41,880
same prompt or. Not and so that's when he gets

727
00:37:41,960 --> 00:37:47,159
the college. Words that's five dollars five dollar. Word it's

728
00:37:47,159 --> 00:37:50,679
a stochastic process because the output vector gives you the

729
00:37:50,719 --> 00:37:55,119
probability distribution and samples according to that probably. Distribution so some,

730
00:37:55,280 --> 00:37:57,480
models depending on let's just say that use clawed in that,

731
00:37:57,519 --> 00:37:59,679
one or they are going to do different things and

732
00:37:59,719 --> 00:38:02,840
it all so the promptly makes a, difference and the

733
00:38:02,880 --> 00:38:05,360
tools that it has really makes a. Difference and that's

734
00:38:05,440 --> 00:38:08,400
where it sounds like an excuse to. Me it does

735
00:38:08,639 --> 00:38:11,639
and here's what they should have. Done just like we

736
00:38:11,719 --> 00:38:15,760
have unit tests for, code we should have quote unit

737
00:38:15,800 --> 00:38:19,800
tests for. Agents and those things are called. Evaluations and

738
00:38:19,880 --> 00:38:23,000
so An azree foundery we have a bunch of evaluations for,

739
00:38:23,039 --> 00:38:27,199
example is it follow the right? Task is it? Coherent

740
00:38:27,719 --> 00:38:30,920
is it? Relevant is it? Grounded we also have these

741
00:38:31,079 --> 00:38:34,840
other agents whose job is to attack our. Agents they're

742
00:38:34,880 --> 00:38:37,239
called the red teaming, agent and its job is to

743
00:38:37,320 --> 00:38:40,159
literally send garbage at it to see how it. Does

744
00:38:40,760 --> 00:38:43,239
and so that is the unit. Test and the difference

745
00:38:43,239 --> 00:38:46,159
between a unit tests and an evaluation is that an

746
00:38:46,280 --> 00:38:49,840
evaluation you need to give it enough examples to be

747
00:38:49,960 --> 00:38:53,119
satisfied that you're going to be. Okay it needs to

748
00:38:53,159 --> 00:38:57,199
be probabilistically, relevant, Yes and then once you're okay with,

749
00:38:57,280 --> 00:38:59,840
it when you put it in, production you monitor it

750
00:38:59,840 --> 00:39:02,679
with the same evails and send alerts when it goes.

751
00:39:02,719 --> 00:39:05,840
Off the rails diverts because when you're looking, at for,

752
00:39:05,880 --> 00:39:10,119
example regulated, industries regular, industries like if you make a,

753
00:39:10,119 --> 00:39:12,239
mistake you got to report, It but they also want

754
00:39:12,239 --> 00:39:14,360
to know how do you fix these? Things and that's

755
00:39:14,360 --> 00:39:17,280
probably more, important and that's the part of regulated. Industries

756
00:39:17,280 --> 00:39:19,360
when you use this s uff you can see where

757
00:39:19,360 --> 00:39:20,760
it went wrong and you fix it and then you go.

758
00:39:20,800 --> 00:39:23,079
Speaker 1: Back SO i think the real answer, is, hey we

759
00:39:23,280 --> 00:39:26,320
just launched this, shit give us some time.

760
00:39:26,400 --> 00:39:28,719
Speaker 2: Time i'm not part of that. Team they mean they

761
00:39:28,800 --> 00:39:29,840
could have made it, better.

762
00:39:29,639 --> 00:39:32,599
Speaker 1: Maybe BUT i mean it literally just came, out like you,

763
00:39:32,639 --> 00:39:33,840
know so and.

764
00:39:34,039 --> 00:39:36,639
Speaker 2: They'll learn a lot, because, like for, example the question

765
00:39:36,679 --> 00:39:38,960
you asked it maybe something they never. Anticipated that's, Right,

766
00:39:39,079 --> 00:39:41,920
yes so there's Learn so there's some alert that probably

767
00:39:42,000 --> 00:39:45,719
fired during THE evls that's, saying, Oh carl did not

768
00:39:45,840 --> 00:39:48,880
get this thing that he, wanted and so they update

769
00:39:48,920 --> 00:39:50,800
the prompt or a new tools AND i expect.

770
00:39:50,840 --> 00:39:54,760
Speaker 1: That. Yeah, so BUT i love the, idea and you,

771
00:39:54,800 --> 00:39:57,719
know talking To Scott hunter and some other, people the

772
00:39:57,840 --> 00:40:01,000
possibilities of things That i'm to do with it in

773
00:40:01,039 --> 00:40:06,679
the real world from my customers is going to blow their.

774
00:40:06,760 --> 00:40:10,559
Speaker 2: Minds, Yes and in terms of like the stuff you can,

775
00:40:10,639 --> 00:40:16,360
do it's the ceiling is very high because the biggest

776
00:40:16,400 --> 00:40:20,159
barrier to programming isn't like the new algorithms or the

777
00:40:20,199 --> 00:40:23,760
new it's like literally people using this stuff and having

778
00:40:23,760 --> 00:40:26,840
the like like someone Putting bob in the age. Field you,

779
00:40:26,840 --> 00:40:29,599
know that's the kind of dumb stuff that we write

780
00:40:29,599 --> 00:40:33,639
most of our code. For right ll ms together in

781
00:40:33,679 --> 00:40:37,159
an agentic, mode are going to soften that so that

782
00:40:37,199 --> 00:40:39,480
we don't have like why do we need a web form?

783
00:40:39,519 --> 00:40:41,760
ANYMORE i can do like my name Is seth and

784
00:40:41,920 --> 00:40:43,559
you can use my current location to find the.

785
00:40:43,599 --> 00:40:45,960
Speaker 1: Address, now you're absolutely, right and YOU i is going to,

786
00:40:46,079 --> 00:40:51,199
change yeah from you. Know but but users are trained on,

787
00:40:51,400 --> 00:40:54,199
forms aren't. They, yes you're going to have to train

788
00:40:54,239 --> 00:40:58,800
a user to all. Right so here's here's a. Nightmarebox nightmare.

789
00:40:58,840 --> 00:41:03,119
Scenario a textbox to tell the application what you, want

790
00:41:03,199 --> 00:41:05,599
and somebody comes in and, says, hey you want A turkey.

791
00:41:05,599 --> 00:41:07,960
Club i'm going to The. Delhi, Yeah i'll have A turkey.

792
00:41:07,960 --> 00:41:10,440
Club and that's Writing turkey, club you know in the

793
00:41:10,679 --> 00:41:13,880
in the. Thing you know WHAT i mean, like there's there's.

794
00:41:14,079 --> 00:41:15,960
You now you have to train the users as to

795
00:41:16,039 --> 00:41:19,119
what to say or what to, type and that's the.

796
00:41:19,239 --> 00:41:25,639
Speaker 2: Beauty. THOUGH i think given a properly built, agent that

797
00:41:25,760 --> 00:41:28,840
turkey sandwich is just gonna be noise to. It like

798
00:41:28,880 --> 00:41:32,000
it's gonna look for the things that will force the

799
00:41:32,159 --> 00:41:36,639
linear algebra and the calculus to produce the right next.

800
00:41:36,679 --> 00:41:39,480
Speaker 1: Time, yeah and then there's gotta be validation and. Verification

801
00:41:39,559 --> 00:41:41,360
did you really mean to say? This is this what

802
00:41:41,519 --> 00:41:41,880
you really?

803
00:41:41,920 --> 00:41:44,360
Speaker 2: Want you? Know and ALSO i think the software is

804
00:41:44,400 --> 00:41:46,800
fairly good at recognizing a contact shift and SAYING i

805
00:41:46,800 --> 00:41:49,039
don't THINK i was supposed to listen to, that, yeah.

806
00:41:49,119 --> 00:41:51,760
Speaker 1: Right and Also i'm assuming that you're speaking to it

807
00:41:51,840 --> 00:41:53,840
right when you really need to be, Typing like you

808
00:41:53,880 --> 00:41:56,400
don't Want bob From engineering to pop his head in and,

809
00:41:56,400 --> 00:41:59,280
say delete all records from, customers you, know.

810
00:42:01,679 --> 00:42:04,039
Speaker 2: DUDE i saw AN mcp server that was attached to

811
00:42:04,079 --> 00:42:06,400
a file, system AND i was, like maybe let's not do. That,

812
00:42:06,480 --> 00:42:08,480
yeah it seems like about on mind's. IDEA i remember

813
00:42:08,519 --> 00:42:12,039
when the you, know the First amazon devices showed, up

814
00:42:12,039 --> 00:42:13,599
and right away it was, like send my porn collection

815
00:42:13,639 --> 00:42:14,039
to my mother.

816
00:42:15,480 --> 00:42:18,719
Speaker 1: Joke oh my, gosh send one hundred pounds of cement

817
00:42:18,800 --> 00:42:19,400
To Richard.

818
00:42:19,440 --> 00:42:24,360
Speaker 2: Campbell oh my, gosh so. Dumb but, yes BUT i

819
00:42:24,400 --> 00:42:27,960
think LIKE i, Think LIKE i said. It it was

820
00:42:28,000 --> 00:42:32,239
surprising to me how the the controller agent in this

821
00:42:32,360 --> 00:42:37,400
case was communicating with other. Agents, yes that. Was and

822
00:42:37,440 --> 00:42:39,239
then you, Know i'm not surprised by this stuff BECAUSE

823
00:42:39,280 --> 00:42:41,159
i know how it. Works, Right but that was the

824
00:42:41,199 --> 00:42:43,679
part that was, Like, okay maybe there's something. Here, well

825
00:42:43,719 --> 00:42:46,599
and we get back to this whole like The battle

826
00:42:46,599 --> 00:42:49,679
Of VISIBLE. Ai, yeah, right and we still haven't really

827
00:42:49,679 --> 00:42:51,800
fixed in there all that. Problem, no there's no way

828
00:42:51,840 --> 00:42:54,760
we can fix. That, well and that the Whole i've

829
00:42:54,760 --> 00:42:57,400
read the anthropic paper on the reasoning models and what's

830
00:42:57,400 --> 00:43:00,559
actually going on. There it's calling itself right couple. Times,

831
00:43:00,559 --> 00:43:03,000
well first it gets the, results and then it peaks

832
00:43:03,039 --> 00:43:04,800
pieces of the results and feeds it back into it

833
00:43:04,840 --> 00:43:07,039
to say why WOULD i have said? This so it

834
00:43:07,199 --> 00:43:10,119
literally is in reverse. Reasoning i've already decided what the answer,

835
00:43:10,199 --> 00:43:11,639
is but let me break it down into pieces so

836
00:43:11,679 --> 00:43:13,360
it looks Like i'm thinking about. It, yeah that doesn't

837
00:43:13,360 --> 00:43:14,960
seem to be a good way to do, it but

838
00:43:15,000 --> 00:43:17,360
it makes people, happy, Right, well it doesn't make me,

839
00:43:17,440 --> 00:43:19,840
happy well until you know what's going. On and it's

840
00:43:19,880 --> 00:43:20,480
LIKE i just.

841
00:43:20,480 --> 00:43:24,599
Speaker 1: Read something in the news today that said THE ceo

842
00:43:24,679 --> 00:43:27,239
Of anthropic admitted that he has no idea how this

843
00:43:27,280 --> 00:43:28,719
stuff for but.

844
00:43:28,800 --> 00:43:32,840
Speaker 2: You, know maybe like in his, defense like there's certain

845
00:43:32,880 --> 00:43:35,280
things that CEOs do THAT i just don't want to

846
00:43:35,320 --> 00:43:37,360
know how to. Do, yeah you know What i'm, Saying

847
00:43:37,480 --> 00:43:40,039
like there's only there's only so MANY i don't know

848
00:43:40,039 --> 00:43:41,840
if this is, true but there's only so much space

849
00:43:41,920 --> 00:43:44,199
in my brain for things, nice, Right and it's LIKE

850
00:43:44,519 --> 00:43:46,800
i have, seen like BECAUSE i work with a lot

851
00:43:46,840 --> 00:43:50,039
of our EVPs and CEOs and and all that, stuff

852
00:43:50,039 --> 00:43:52,800
and like And i've worked And i've talked to CEOs

853
00:43:52,880 --> 00:43:57,000
and cvps and it's just like MAYBE i don't want

854
00:43:57,000 --> 00:43:59,360
to do. That, Yeah and and the fact that he

855
00:43:59,480 --> 00:44:04,320
doesn't know exactly how it works is, okay focus on

856
00:44:04,320 --> 00:44:06,280
the right. Things. Yeah and the other thing that for,

857
00:44:06,400 --> 00:44:10,039
me LIKE i mathematically know how these things, work BUT

858
00:44:10,079 --> 00:44:12,639
i don't expect other people to just to. Me it's

859
00:44:12,679 --> 00:44:15,320
just LIKE i love looking at like the beauty of

860
00:44:15,360 --> 00:44:19,320
the mathematical structure of these things and trying to think

861
00:44:19,320 --> 00:44:21,840
about why does this actually?

862
00:44:21,960 --> 00:44:25,000
Speaker 1: Work but you know, WHAT i feel better knowing that

863
00:44:25,079 --> 00:44:25,800
as long AS sef.

864
00:44:25,840 --> 00:44:31,519
Speaker 2: Knows BUT i think this idea that that neural net

865
00:44:31,559 --> 00:44:37,679
is is essentially creating a jpeg of, language lossy compression

866
00:44:37,760 --> 00:44:41,159
of language one hundred. Percent that is exactly. Right but

867
00:44:41,199 --> 00:44:42,760
when we, say and the whole thing, is don't put

868
00:44:42,800 --> 00:44:45,239
any too many pictures in, It, yeah you're just gonna

869
00:44:45,239 --> 00:44:47,760
get a GREAT FUZD i. Know and then but but,

870
00:44:47,800 --> 00:44:50,880
then but, then because these things are really good at,

871
00:44:50,920 --> 00:44:54,800
sequences think of it as any. Sequence so we have

872
00:44:54,920 --> 00:44:58,280
models And andrea found with that do protein. Sequencing, yeah,

873
00:44:58,440 --> 00:45:01,519
well deep. Fault and if you think about, it you're, like,

874
00:45:01,599 --> 00:45:04,280
well how could that? Work, well because again you're giving

875
00:45:04,320 --> 00:45:06,599
it a sequence of, things and it's telling you the

876
00:45:06,639 --> 00:45:09,239
next likely. Thing and then it's doing it in a

877
00:45:10,079 --> 00:45:14,719
loop right while not end of line token or end

878
00:45:14,719 --> 00:45:17,800
of thing. Token, right that's what it's doing and get

879
00:45:17,960 --> 00:45:20,199
next probable. Words. Yeah and the thing about it is

880
00:45:20,440 --> 00:45:22,679
people will talk about the, temperature whether it makes it

881
00:45:22,719 --> 00:45:27,079
more creative or. Whatever it's not making it more. Creative

882
00:45:27,320 --> 00:45:31,880
it's basically squashing the output distribution so other things are more.

883
00:45:32,079 --> 00:45:34,199
Likely and that's all it's. Doing and SO i hear

884
00:45:34,239 --> 00:45:36,519
it's making it more. Creative, no it's. Not it's changing

885
00:45:36,599 --> 00:45:39,159
the hand of the output. Distribution it'll make it, faster

886
00:45:39,440 --> 00:45:42,639
but you'll get. Crap it's not even. Faster it's exactly

887
00:45:42,679 --> 00:45:46,239
the same. Speed it's just the probability distribution is warped

888
00:45:46,480 --> 00:45:48,719
so that if, like for, sure it's this, one if

889
00:45:48,760 --> 00:45:50,760
temper is, one it's like the probably is what it

890
00:45:50,760 --> 00:45:52,800
it'll pick the, right it'll. Almost but if it's, different

891
00:45:52,840 --> 00:45:55,960
there's other things that become more equally. Likely and so

892
00:45:56,039 --> 00:45:59,719
it's chain of like this, token now this token it

893
00:45:59,719 --> 00:46:02,559
Becomes they say it's more, creative AND i hate. That

894
00:46:02,840 --> 00:46:07,719
why DO i hate? That? Richard it's more, exactly it's

895
00:46:07,880 --> 00:46:09,559
literally just changing the output.

896
00:46:09,559 --> 00:46:12,920
Speaker 1: Distigue but, okay So joe, programmer, though do you think

897
00:46:12,960 --> 00:46:15,920
they need to answer pomorphize this stuff in order to

898
00:46:16,000 --> 00:46:18,400
talk about it or think about it in a in

899
00:46:18,440 --> 00:46:21,920
a way that they can use it correctly because everybody

900
00:46:21,920 --> 00:46:24,000
falls into this. TRAP i mean we're, Humans.

901
00:46:23,880 --> 00:46:27,480
Speaker 2: Yeah like humans are funny. Creatures we will we will

902
00:46:27,519 --> 00:46:29,880
pick up a rock and draw a smiley face on,

903
00:46:29,920 --> 00:46:35,880
it and so. Happy it's such a happy. Rock it

904
00:46:35,960 --> 00:46:38,760
must have had a really good Dayne and now. Imagine

905
00:46:39,199 --> 00:46:41,639
but isn't THIS i WANT i have to think this

906
00:46:41,760 --> 00:46:44,639
is what makes humans? Humans first, off paradolia is a survival.

907
00:46:44,639 --> 00:46:47,639
Strategy the tendency to see, faces whether they're there or,

908
00:46:47,760 --> 00:46:50,639
not means you were the first one to, run and

909
00:46:50,679 --> 00:46:53,480
so the tiger probably didn't get. You so that's an evolved.

910
00:46:53,639 --> 00:46:59,000
Trait absolutely, right it's. Paradolia but what if this whole

911
00:46:59,280 --> 00:47:03,599
practice event memorphizing is the mechanism that led to us

912
00:47:03,639 --> 00:47:07,039
being able to, strategize like how you have to think

913
00:47:07,239 --> 00:47:10,239
like the animal to hunt the animal successfully with your limited.

914
00:47:10,280 --> 00:47:14,559
RESOURCE i, See and so that tendency again becomes a competitive.

915
00:47:14,599 --> 00:47:17,199
Strategy you're more likely to survive because your hunting is.

916
00:47:17,199 --> 00:47:20,639
Speaker 1: Morphout not to mention, mythology, right, well in, mythos in

917
00:47:20,760 --> 00:47:23,920
religion and all of these, things it's all about. Anthropomorphizing,

918
00:47:24,079 --> 00:47:24,880
sure and not just.

919
00:47:25,320 --> 00:47:27,519
Speaker 2: Why is the sky making loud noises?

920
00:47:27,559 --> 00:47:31,000
Speaker 1: Shaming right As thor is sending lightning, thing that's because

921
00:47:31,039 --> 00:47:31,760
he's angry with.

922
00:47:31,840 --> 00:47:33,880
Speaker 2: US i feel like we got to go the final

923
00:47:33,920 --> 00:47:37,639
mile and somehow get into The roman spears because we

924
00:47:37,679 --> 00:47:41,719
all think About roman, period, RIGHT i don't, Know but,

925
00:47:42,159 --> 00:47:45,440
well actually that project the name for the project that

926
00:47:45,480 --> 00:47:49,639
you Saw amanda And elijah Do we called it, sustinnao

927
00:47:49,760 --> 00:47:53,199
which Is latin for, SUPPORT i support, nice, Okay and

928
00:47:53,280 --> 00:47:55,440
SO i told him that and let's lie it for.

929
00:47:55,519 --> 00:48:03,000
Sales way to close the loop that was that was.

930
00:48:03,039 --> 00:48:07,760
Amazing this is a quality. Show my job as co

931
00:48:07,880 --> 00:48:12,920
host Of. Professionals the joke that's but, Yeah so again the,

932
00:48:13,000 --> 00:48:16,559
key the key insight is think of these things as

933
00:48:16,599 --> 00:48:19,360
things that take. Language and if you have a prompt

934
00:48:19,360 --> 00:48:22,039
that's narrow enough and a collection of tools that support the,

935
00:48:22,079 --> 00:48:26,920
task it will tell you what consistent. Results, yeah, yeah

936
00:48:27,079 --> 00:48:30,960
That's these are the dimensions of a tool set for

937
00:48:31,039 --> 00:48:33,800
a group of programmers to be successful with correct they

938
00:48:33,880 --> 00:48:35,719
just had to learn the dimensions of. It and the

939
00:48:35,840 --> 00:48:38,320
joke THAT i, said is it like a switch statement

940
00:48:38,400 --> 00:48:41,199
and if statement but. Huge So i'm going to try

941
00:48:41,199 --> 00:48:43,639
to make it a thing and call it the giant swift.

942
00:48:43,800 --> 00:48:49,239
Statement it's the swift for those that are. Swifties swift all.

943
00:48:49,280 --> 00:48:58,760
Speaker 1: RIGHT i figured out The latin for sales is.

944
00:48:55,400 --> 00:49:01,960
Speaker 2: Isn't That italian full Of? Chicago oh my, gosh that's.

945
00:49:02,039 --> 00:49:04,800
Hilarious but, yeah and so that that inside couple with

946
00:49:04,840 --> 00:49:07,480
the ability to evaluate them as unit with the unit

947
00:49:07,719 --> 00:49:12,280
with the agentic. Unit tests should bring this all back

948
00:49:12,400 --> 00:49:15,480
to quality instead of the schlop that people put. Out

949
00:49:15,559 --> 00:49:17,719
sure they, are and THAT'S i mean the battle here

950
00:49:17,960 --> 00:49:20,639
is how do you measure? Quality, oh there's a way

951
00:49:20,679 --> 00:49:23,679
to measure? It, yeah BECAUSE i keep being told this

952
00:49:23,679 --> 00:49:25,760
THIS lm is better than THAT llm is, like how

953
00:49:25,840 --> 00:49:28,559
how would you measure? That what makes that? True so

954
00:49:28,840 --> 00:49:30,559
what they do is they give it a bunch of,

955
00:49:30,559 --> 00:49:33,239
stuff like, like they feed it stuff and then they.

956
00:49:33,280 --> 00:49:35,800
Measure i'll give you there's a couple of. Emails so

957
00:49:35,800 --> 00:49:37,760
there's something Like blue And, rouge which is like how

958
00:49:37,800 --> 00:49:39,639
close is it to the right? Answer? Right those are

959
00:49:39,679 --> 00:49:42,480
supervised tests where you're, like you give it a thing

960
00:49:42,559 --> 00:49:44,519
and you're, like here's the OUTPUT i wanted to. Have

961
00:49:44,679 --> 00:49:45,519
we've talked about this.

962
00:49:45,559 --> 00:49:48,960
Speaker 1: Before, Yeah blue And rouges is like those like supervised

963
00:49:48,960 --> 00:49:50,280
and supervised.

964
00:49:50,280 --> 00:49:52,679
Speaker 2: Tests. Right but then there's one supervised tests and you're, like,

965
00:49:52,719 --> 00:49:54,400
well how do those? Work, Well i'll give you an.

966
00:49:54,400 --> 00:49:57,760
Example there's a there's a judge BASED evl called, groundedness

967
00:49:57,960 --> 00:50:01,079
and what groundedness. Measures it measures how rounded a response

968
00:50:01,119 --> 00:50:04,199
from AN lm is in the facts and context that

969
00:50:04,280 --> 00:50:07,000
you gave. It, Yeah and so for, example the three

970
00:50:07,039 --> 00:50:09,639
pairing that you give it for the unit test is sas,

971
00:50:09,679 --> 00:50:12,960
asked how much does your you, know outdoor tent? Cost

972
00:50:13,719 --> 00:50:16,280
the context is we give it all of like the tent,

973
00:50:16,400 --> 00:50:19,400
information and the cost is in, there, right it's just

974
00:50:19,480 --> 00:50:21,159
like the slop of it and the outputs like the

975
00:50:21,199 --> 00:50:25,920
tent is ten. Dollars. Right so given given that a judge.

976
00:50:25,599 --> 00:50:27,800
Speaker 1: Base before the, tariffs now there are a lot, more.

977
00:50:27,719 --> 00:50:29,760
Speaker 2: That's, Right and now they're gonna be a lot. More

978
00:50:29,960 --> 00:50:34,599
but that particular three two, pole it's not, twopole but

979
00:50:34,639 --> 00:50:39,599
the three pole three. Yeah of the, question the context

980
00:50:39,679 --> 00:50:42,719
and the. Answer, yeah it's given to a judge to

981
00:50:42,880 --> 00:50:45,079
judge on a scale of one to. Five how grounded is?

982
00:50:45,119 --> 00:50:48,920
It and that judge is also AN Lllm, yeah okay

983
00:50:49,119 --> 00:50:52,239
another one, yeah and it. Works and that's why that's

984
00:50:52,280 --> 00:50:55,119
the whole idea of, Agency, yeah is that you have

985
00:50:55,199 --> 00:50:59,519
agents that are specialized in their different, things just like

986
00:50:59,599 --> 00:51:01,679
you have you on your. Staff there special that's. Right

987
00:51:01,960 --> 00:51:05,760
the people of my staff Question, mark it's just, me all, Right.

988
00:51:05,639 --> 00:51:09,199
Speaker 1: But Joe developer who's listening is, thinking, oh this is.

989
00:51:09,239 --> 00:51:11,760
Great you, know after listening to these shows for the,

990
00:51:11,840 --> 00:51:14,480
last you, know the week that we've been doing, These

991
00:51:14,920 --> 00:51:17,400
i'm thinking now that you, Know i'm going to get

992
00:51:17,440 --> 00:51:19,480
rid Of bob and engineering And i'm going to replace no,

993
00:51:19,639 --> 00:51:24,079
agent And i'm going to set up this little staff

994
00:51:24,119 --> 00:51:27,559
of programmers under. Me agent programmers are going to help

995
00:51:27,599 --> 00:51:28,000
me do my.

996
00:51:28,079 --> 00:51:30,559
Speaker 2: JOB i don't think that's ever going to. Happen WHAT

997
00:51:30,639 --> 00:51:32,880
i think is going to happen AS i already. HAPPENED

998
00:51:32,960 --> 00:51:35,360
i think we're going to solve. WELL i mean those

999
00:51:35,400 --> 00:51:37,119
that think are they're going to solve it that way

1000
00:51:37,159 --> 00:51:40,239
are going to come in Like i've seen vibe coders

1001
00:51:40,679 --> 00:51:42,519
like put out stuff and be, like oh why is

1002
00:51:42,559 --> 00:51:44,599
MY aws bill so high.

1003
00:51:45,559 --> 00:51:49,400
Speaker 1: About vibe Coders and There's we've been arguing about whether

1004
00:51:49,400 --> 00:51:52,039
that's pejorative or, not because the ryan it's.

1005
00:51:52,079 --> 00:51:55,000
Speaker 2: Not, okay is it a pejorative?

1006
00:51:55,000 --> 00:51:57,559
Speaker 1: Thing some people see, it some people see it, Goodness,

1007
00:51:58,039 --> 00:52:01,360
Yeah But i'm talking about a professional co that maybe

1008
00:52:01,360 --> 00:52:05,400
works by themselves and hires out other specialists to do

1009
00:52:05,519 --> 00:52:07,639
that kind of. Stuff, interesting and NOW i can maybe

1010
00:52:07,639 --> 00:52:09,639
spin off some agents to help me with that kind

1011
00:52:09,639 --> 00:52:11,599
of stuff AND i not depend on them as.

1012
00:52:11,679 --> 00:52:14,800
Speaker 2: MUCH i think that the agents are going to be

1013
00:52:14,840 --> 00:52:18,400
able to do with a low powered brain work that

1014
00:52:18,519 --> 00:52:27,000
we as humans do and and but the actual creative

1015
00:52:27,679 --> 00:52:32,480
move forward and always human right and not and not

1016
00:52:32,519 --> 00:52:34,639
be stuck in the drudgery of. That so you actually

1017
00:52:34,639 --> 00:52:37,840
have more creative energy to do more experimentation. Exactly because

1018
00:52:37,880 --> 00:52:38,480
then that's.

1019
00:52:38,280 --> 00:52:41,559
Speaker 1: Why imagination is going to be the number one uh

1020
00:52:42,320 --> 00:52:45,000
feature of the Rock star. Developers, oh there you, go

1021
00:52:45,079 --> 00:52:46,880
there you. Go that's WHY i Think i've never been

1022
00:52:46,920 --> 00:52:49,559
a rock star. Though, dude you are the more like

1023
00:52:49,599 --> 00:52:54,159
a flute, player saying like You're Eddie eddie Van halen

1024
00:52:54,360 --> 00:52:57,880
jazz PLAYER ai At Microsoft Cool.

1025
00:52:57,880 --> 00:53:02,119
Speaker 2: Beings well look, again AND i just want to emphasize

1026
00:53:02,599 --> 00:53:05,239
this is all accessible to. Everybody you can start playing

1027
00:53:05,239 --> 00:53:08,159
with it. Today make it do a raw function call

1028
00:53:08,280 --> 00:53:10,559
and see the output so you can see that What

1029
00:53:10,639 --> 00:53:12,960
i'm saying is. True don't have it do the function,

1030
00:53:13,039 --> 00:53:15,800
call just call it and tell it to do. Return

1031
00:53:16,239 --> 00:53:18,480
do a tool call and you'll see it literally Returns.

1032
00:53:18,559 --> 00:53:23,599
Jason that, says run this function with these, parameters.

1033
00:53:23,039 --> 00:53:25,760
Speaker 1: And be as careful with your prompts with your agents

1034
00:53:25,920 --> 00:53:28,079
as you are with your ten year old when they

1035
00:53:28,079 --> 00:53:30,679
tell them to go to the store and get used xy.

1036
00:53:30,559 --> 00:53:31,840
Speaker 2: Eggs or your programmer.

1037
00:53:31,960 --> 00:53:36,679
Speaker 1: Husband, yes that's, right that's you, know the one about the.

1038
00:53:36,719 --> 00:53:39,559
Programmer your husband and the wife, said go to the

1039
00:53:39,599 --> 00:53:42,039
store and get a gallon of. Milk if they have,

1040
00:53:42,159 --> 00:53:44,519
eggs get a. Dozen he came back with a dozen.

1041
00:53:44,639 --> 00:53:46,079
Gas they had.

1042
00:53:46,119 --> 00:53:50,280
Speaker 2: Eggs they had. Eggs but that's just how we have

1043
00:53:50,320 --> 00:53:52,039
the whole world turned into paper. Clips but if you

1044
00:53:52,119 --> 00:53:54,519
need to be, precise but if you think about, it

1045
00:53:54,599 --> 00:53:57,119
that's exactly how you should think about these. Things that's.

1046
00:53:57,199 --> 00:53:59,360
Right you have to be very because you want to.

1047
00:53:59,400 --> 00:54:03,079
Listen you want to maximize the probability of it returning

1048
00:54:03,119 --> 00:54:05,320
the right tokens based upon the ones you gave.

1049
00:54:05,360 --> 00:54:08,079
Speaker 1: It so we're all in for a new world of

1050
00:54:08,159 --> 00:54:14,119
discovery and refinement of search and refinement of. Prompts, yeah

1051
00:54:14,199 --> 00:54:15,280
and the whole new.

1052
00:54:15,280 --> 00:54:17,679
Speaker 2: World and if you go about retrieval augmented, generation it's

1053
00:54:17,719 --> 00:54:21,440
literally the process of you get a, question you programmatically

1054
00:54:21,519 --> 00:54:24,039
search for an, answer you put it into the, prompt

1055
00:54:24,639 --> 00:54:27,039
and then you get the. Answer that's. It, yeah that's

1056
00:54:27,079 --> 00:54:28,639
all it. Is that's How, yeah we just gave it

1057
00:54:28,639 --> 00:54:30,800
a fancy. Name, Yeah WELL i don't like the TERM.

1058
00:54:30,920 --> 00:54:35,519
Rag it sounds it sounds kind Of but, YEAH i

1059
00:54:35,519 --> 00:54:38,400
do a demo WHERE i do augmented, generation WHERE i

1060
00:54:38,480 --> 00:54:41,800
literally go and copy From, wikipedia BUT i think the

1061
00:54:41,840 --> 00:54:43,679
right answer is put it in AND i show it

1062
00:54:43,719 --> 00:54:45,480
AND i was, like how does it know the right? Answer

1063
00:54:45,480 --> 00:54:47,039
and it was, like, well because you put it. In,

1064
00:54:47,239 --> 00:54:50,880
okay now you understand what augmented generation. Is the only

1065
00:54:50,960 --> 00:54:54,760
process of retrieval was artisanal in my case BECAUSE i

1066
00:54:54,800 --> 00:54:59,880
went and did. It it was. Artisanal, yeah it's like

1067
00:54:59,880 --> 00:55:03,800
a fine, cheese, Right but we's That and then if

1068
00:55:03,840 --> 00:55:06,280
you think about we're too. Generation by putting the right

1069
00:55:06,280 --> 00:55:10,199
context in and then coupled with tool, calling now you

1070
00:55:10,239 --> 00:55:14,000
have a powerful. Combo and that's WHERE i, think if

1071
00:55:14,039 --> 00:55:15,559
you think about it like, this you're really going to

1072
00:55:15,639 --> 00:55:19,079
change the way things. Happen sure, WELL i also love

1073
00:55:19,199 --> 00:55:24,880
how much you dilute the pretense of all this. Technology,

1074
00:55:24,960 --> 00:55:27,719
OH i know it's the. PRETENSE i said that in

1075
00:55:27,760 --> 00:55:30,079
a very pretentious. WAY i. Know it's kind of.

1076
00:55:30,079 --> 00:55:35,400
Speaker 1: Amazing, actually, yeah.

1077
00:55:33,320 --> 00:55:38,119
Speaker 2: It was like a two fur three For, yeah there you.

1078
00:55:38,159 --> 00:55:46,360
Go but like That, latin but it's a plural of. Slezolo.

1079
00:55:46,519 --> 00:55:49,119
Right but if you think about, it this has been

1080
00:55:49,280 --> 00:55:52,360
What i've been doing for the last twenty. Years, Sure,

1081
00:55:52,440 --> 00:55:54,960
yes that's. True i've been trying to make this stuff

1082
00:55:55,079 --> 00:55:58,239
lose all of its pretentiousness and when you, do it

1083
00:55:58,280 --> 00:56:02,360
empowers you to do really. Things absolutely. Sure does the

1084
00:56:02,400 --> 00:56:04,280
tools only get? Better that's, Right Brave New.

1085
00:56:04,280 --> 00:56:06,800
Speaker 1: World and that's the last word from. BUILD i, Suppose

1086
00:56:07,079 --> 00:56:10,079
WE'RE i, Love we're. Done it's TIME uh to hit the.

1087
00:56:10,159 --> 00:56:12,280
Road thank you for. Listening thank, You.

1088
00:56:12,320 --> 00:56:13,760
Speaker 2: Seth it's always a, pleasure always a.

1089
00:56:13,760 --> 00:56:17,159
Speaker 1: Pleasure and we'll talk to you next time on dot net.

1090
00:56:17,239 --> 00:56:41,360
Rocks dot Net rocks is brought to you By Franklin's

1091
00:56:41,400 --> 00:56:45,440
net and produced By Pop, studios a full service, audio

1092
00:56:45,559 --> 00:56:49,320
video and post production facility located physically In New, London,

1093
00:56:49,400 --> 00:56:54,239
connecticut and of course in the cloud online at pwop dot.

1094
00:56:54,239 --> 00:56:57,159
Com visit our website at D O T N E

1095
00:56:57,239 --> 00:56:59,760
T R o c K s dot com FOR R

1096
00:56:59,880 --> 00:57:04,159
S s, feeds, downloads mobile, apps, comments and access to

1097
00:57:04,199 --> 00:57:07,599
the full archives going back to show number, one recorded

1098
00:57:07,639 --> 00:57:10,400
In september two thousand and. Two and make sure you

1099
00:57:10,480 --> 00:57:13,559
check out our. Sponsors they keep us in. Business now

1100
00:57:13,599 --> 00:57:17,079
go write some, code see you next. Time you Got

1101
00:57:17,159 --> 00:57:28,760
Jent Middle vans