Unlocking the Power of Open-Source AI: Exploring Meta's Llama 3.1 Release

Unlocking the Power of Open-Source AI: Exploring Meta's Llama 3.1 Release

Source fidelity

The timestamped transcript below is the exact English auto-generated SRT caption file recovered from the preserved YouTube episode on 2026-07-28. No spelling, punctuation, speaker, or wording corrections were applied. The original SRT is preserved beside this note as E27 - YouTube captions - gg2I5iO1x0o.en.srt.

Timestamped transcript

1 00:00:00,120 --> 00:00:01,839 welcome to venture step podcast where we

2 00:00:01,839 --> 00:00:04,359 discuss entrepreneurship industry Trends

3 00:00:04,359 --> 00:00:06,040 and the occasional book

4 00:00:06,040 --> 00:00:09,160 review have you ever wondered what AI

5 00:00:09,160 --> 00:00:11,920 would be like if it wasn't just a tool

6 00:00:11,920 --> 00:00:13,240 but a partner in

7 00:00:13,240 --> 00:00:15,759 Innovation and something that you could

8 00:00:15,759 --> 00:00:17,800 carry around in your pocket in your

9 00:00:17,800 --> 00:00:21,080 glasses on your watch or at your home

10 00:00:21,080 --> 00:00:24,720 and have your own setup for your AI well

11 00:00:24,720 --> 00:00:27,800 you can do that now first time ever you

12 00:00:27,800 --> 00:00:30,599 have a highly capable

13 00:00:30,599 --> 00:00:32,719 foundational model that you could

14 00:00:32,719 --> 00:00:35,920 download and run at your house you only

15 00:00:35,920 --> 00:00:38,879 need the compute of multiple

16 00:00:38,879 --> 00:00:43,160 computers and a plethora of other things

17 00:00:43,160 --> 00:00:44,960 to get it going but you could do it if

18 00:00:44,960 --> 00:00:49,320 you wanted to but many Enterprises are

19 00:00:49,320 --> 00:00:54,879 doing so llama has 3500 Enterprises

20 00:00:54,879 --> 00:00:56,559 using their

21 00:00:56,559 --> 00:01:00,440 models and llama recently just surpassed

22 00:01:00,440 --> 00:01:03,079 300 million total downloads and that was

23 00:01:03,079 --> 00:01:07,159 before they had the 45 billion parameter

24 00:01:07,159 --> 00:01:12,400 model which is insanely good for being

25 00:01:12,400 --> 00:01:15,000 open source I think there was a lot of

26 00:01:15,000 --> 00:01:17,000 question on whether or not open source

27 00:01:17,000 --> 00:01:18,920 would ever be able to compete with a

28 00:01:18,920 --> 00:01:20,840 closed Source

29 00:01:20,840 --> 00:01:27,040 system and meta after using 16,000

30 00:01:27,040 --> 00:01:31,680 h100s to train their cluster and almost

31 00:01:31,680 --> 00:01:33,479 three

32 00:01:33,479 --> 00:01:38,320 trillion training tokens at an 8K Contex

33 00:01:38,320 --> 00:01:41,720 and 800 billion tokens at

34 00:01:41,720 --> 00:01:44,040 128k

35 00:01:44,040 --> 00:01:47,000 context it seems like they've done it

36 00:01:47,000 --> 00:01:49,960 definitely for the most part the range

37 00:01:49,960 --> 00:01:53,520 of the results vary but I won't go

38 00:01:53,520 --> 00:01:56,719 through all of the benchmarks I'll just

39 00:01:56,719 --> 00:02:00,600 talk about a very simple diagram they

40 00:02:00,600 --> 00:02:04,600 have basically they compare llama 3.1 45

41 00:02:04,600 --> 00:02:07,160 billion versus chat

42 00:02:07,160 --> 00:02:13,040 PT 40 and four so four which is the

43 00:02:13,040 --> 00:02:16,560 previous one not 40 so four four Omni is

44 00:02:16,560 --> 00:02:20,720 the newer one and then Chach 4 is the

45 00:02:20,720 --> 00:02:24,440 older Chach 4 or chat

46 00:02:24,440 --> 00:02:30,319 Chach chatu PT 4 versus llama 3.1 4 and

47 00:02:30,319 --> 00:02:36,599 five billion won 23.3% of the time tied

48 00:02:36,599 --> 00:02:40,959 52% of the time okay so it lost you know

49 00:02:40,959 --> 00:02:41,840 around

50 00:02:41,840 --> 00:02:47,440 24% llama 3.1 45 billion versus chbt

51 00:02:47,440 --> 00:02:50,360 40 That's the newer model that open AI

52 00:02:50,360 --> 00:02:51,959 is

53 00:02:51,959 --> 00:02:53,920 offering which is the one you need

54 00:02:53,920 --> 00:02:57,920 premium for 19.1 of the time at one and

55 00:02:57,920 --> 00:03:01,680 51% of the time at tide so lost around

56 00:03:01,680 --> 00:03:04,519 you know 29% of the time that's crazy

57 00:03:04,519 --> 00:03:07,640 since businesses no longer have to send

58 00:03:07,640 --> 00:03:09,640 their data

59 00:03:09,640 --> 00:03:12,560 to this close Source model and they can

60 00:03:12,560 --> 00:03:15,400 control the weights and the parameters

61 00:03:15,400 --> 00:03:17,239 of these

62 00:03:17,239 --> 00:03:18,840 AI

63 00:03:18,840 --> 00:03:21,680 models whereas in a close Source model

64 00:03:21,680 --> 00:03:23,959 it's one size fits all and it's not

65 00:03:23,959 --> 00:03:27,560 necessarily given the opportunity to to

66 00:03:27,560 --> 00:03:30,720 distill the model to edit the weights to

67 00:03:30,720 --> 00:03:32,799 do things as you wish and to run it on

68 00:03:32,799 --> 00:03:33,920 your own

69 00:03:33,920 --> 00:03:36,239 infrastructure you can do

70 00:03:36,239 --> 00:03:40,680 that now and I'm super excited about it

71 00:03:40,680 --> 00:03:43,040 because I think it

72 00:03:43,040 --> 00:03:46,599 will have a positive effect and one

73 00:03:46,599 --> 00:03:48,760 thing I do try to do is I try to just be

74 00:03:48,760 --> 00:03:50,879 very positive in my remarks about these

75 00:03:50,879 --> 00:03:52,640 releases there are some things that

76 00:03:52,640 --> 00:03:55,959 aren't so good and I I'll call them out

77 00:03:55,959 --> 00:03:58,439 I've called out Google several times for

78 00:03:58,439 --> 00:04:00,760 misleading people with their

79 00:04:00,760 --> 00:04:03,640 demos but for the most part I try to be

80 00:04:03,640 --> 00:04:06,000 positive as these companies are putting

81 00:04:06,000 --> 00:04:07,480 themselves out there and they keep

82 00:04:07,480 --> 00:04:09,840 pushing each other and the rate of

83 00:04:09,840 --> 00:04:14,439 innovation is incredible basically every

84 00:04:14,439 --> 00:04:16,440 one like maybe every four weeks or

85 00:04:16,440 --> 00:04:18,000 something six weeks there's there's some

86 00:04:18,000 --> 00:04:19,199 big

87 00:04:19,199 --> 00:04:21,000 announcement and you don't really see

88 00:04:21,000 --> 00:04:22,240 that kind of

89 00:04:22,240 --> 00:04:25,759 stuff especially from large companies as

90 00:04:25,759 --> 00:04:28,960 they're concerned about their brand

91 00:04:28,960 --> 00:04:32,199 recognition and potential mistakes and

92 00:04:32,199 --> 00:04:36,840 and people utilizing these mod models to

93 00:04:36,840 --> 00:04:40,639 make media stories to push certain

94 00:04:40,639 --> 00:04:42,880 agendas because overall that the

95 00:04:42,880 --> 00:04:46,000 technology doesn't have any evil use

96 00:04:46,000 --> 00:04:48,520 cases it's the users but I'll

97 00:04:48,520 --> 00:04:51,919 digress but before we dive in fully I'm

98 00:04:51,919 --> 00:04:55,320 your host dton Anderson my background is

99 00:04:55,320 --> 00:04:57,320 a mix of programming data science and

100 00:04:57,320 --> 00:05:00,720 insurance offline you can fil me find me

101 00:05:00,720 --> 00:05:03,360 running building my side business lost

102 00:05:03,360 --> 00:05:06,440 in a good book you can listen to this

103 00:05:06,440 --> 00:05:09,479 podcast and video and audio on YouTube

104 00:05:09,479 --> 00:05:11,080 If audio is more your thing you can find

105 00:05:11,080 --> 00:05:12,880 the podcast on

106 00:05:12,880 --> 00:05:16,080 Spotify Apple podcast or wherever else

107 00:05:16,080 --> 00:05:19,360 you get your podcasts including YouTube

108 00:05:19,360 --> 00:05:22,759 okay so you could have guessed it we're

109 00:05:22,759 --> 00:05:25,560 going to be talking about meta llama

110 00:05:25,560 --> 00:05:27,919 3.1 and that's going to entail a couple

111 00:05:27,919 --> 00:05:28,919 things we're just talk about some

112 00:05:28,919 --> 00:05:30,240 feature releases

113 00:05:30,240 --> 00:05:32,680 what are my thoughts about llama the new

114 00:05:32,680 --> 00:05:36,600 one and where do I think things are

115 00:05:36,600 --> 00:05:38,039 going to be going with open source like

116 00:05:38,039 --> 00:05:40,639 why do I think open source is great and

117 00:05:40,639 --> 00:05:42,800 maybe some remarks that summarize

118 00:05:42,800 --> 00:05:44,919 remarks from Mark from from the things

119 00:05:44,919 --> 00:05:47,479 I've watched and read from him he's done

120 00:05:47,479 --> 00:05:50,120 a couple of interviews since the release

121 00:05:50,120 --> 00:05:51,919 and he has

122 00:05:51,919 --> 00:05:55,759 also wrote or not wrote but he released

123 00:05:55,759 --> 00:05:58,560 a he released a letter to you know

124 00:05:58,560 --> 00:06:01,360 stakeholders and and such so I read that

125 00:06:01,360 --> 00:06:05,520 as well and it quite interesting but

126 00:06:05,520 --> 00:06:09,319 basically you know these these large CEO

127 00:06:09,319 --> 00:06:11,479 like large company CEOs for these kind

128 00:06:11,479 --> 00:06:14,240 of things these Visionary things they do

129 00:06:14,240 --> 00:06:17,440 a good job of of honing in on the vision

130 00:06:17,440 --> 00:06:19,440 and what they want and so during these

131 00:06:19,440 --> 00:06:21,560 interviews they almost have a script

132 00:06:21,560 --> 00:06:23,599 where they run through the ideas that

133 00:06:23,599 --> 00:06:26,080 they have the messages they want to

134 00:06:26,080 --> 00:06:28,440 release and that's what they talk about

135 00:06:28,440 --> 00:06:30,840 and some of the phrases are almost

136 00:06:30,840 --> 00:06:33,280 furba and I've heard that Mark say like

137 00:06:33,280 --> 00:06:36,840 the same exact phrase like seven times

138 00:06:36,840 --> 00:06:39,520 and like the same example like six

139 00:06:39,520 --> 00:06:41,199 times

140 00:06:41,199 --> 00:06:43,960 so it's it's it's interesting you hear

141 00:06:43,960 --> 00:06:46,520 the same thing from Elon or these other

142 00:06:46,520 --> 00:06:48,240 CEOs that you follow during their

143 00:06:48,240 --> 00:06:50,199 interviews if they go on their Road

144 00:06:50,199 --> 00:06:52,680 shows they have the same messaging and

145 00:06:52,680 --> 00:06:54,720 and vision consistently so make sure

146 00:06:54,720 --> 00:06:56,759 they're heard I guess so the power of

147 00:06:56,759 --> 00:06:59,319 Open Source building with llama so how

148 00:06:59,319 --> 00:07:01,560 how can you go about doing that some of

149 00:07:01,560 --> 00:07:03,639 the Safety and Security tools that they

150 00:07:03,639 --> 00:07:05,879 built and then I have some some

151 00:07:05,879 --> 00:07:07,560 miscellaneous things like the future of

152 00:07:07,560 --> 00:07:10,240 AI and llama where I can't go over

153 00:07:10,240 --> 00:07:11,440 everything today because there's

154 00:07:11,440 --> 00:07:13,400 literally so much and this is probably

155 00:07:13,400 --> 00:07:14,720 the first time where I'm gonna have to

156 00:07:14,720 --> 00:07:16,759 break up an episode I'm not going to

157 00:07:16,759 --> 00:07:19,160 record two separate episodes but I am

158 00:07:19,160 --> 00:07:20,520 going to have to break these up because

159 00:07:20,520 --> 00:07:23,599 it's going to be too much so basically

160 00:07:23,599 --> 00:07:26,680 they released a whole bunch of features

161 00:07:26,680 --> 00:07:29,520 last week we talked about what I think

162 00:07:29,520 --> 00:07:31,160 think is going to be released and and

163 00:07:31,160 --> 00:07:33,680 the real searching this week they

164 00:07:33,680 --> 00:07:35,720 released an

165 00:07:35,720 --> 00:07:39,560 updated model for updated models for

166 00:07:39,560 --> 00:07:42,199 their their previous models so their 70

167 00:07:42,199 --> 00:07:46,560 and their what five or seven no eight

168 00:07:46,560 --> 00:07:48,800 sorry they 70 and their 8 billion

169 00:07:48,800 --> 00:07:51,759 parameter models and then their 4.5 45

170 00:07:51,759 --> 00:07:54,240 billion parameter model so those are

171 00:07:54,240 --> 00:07:56,199 brand new so basically they took their

172 00:07:56,199 --> 00:07:59,080 405 billion parameter model they did all

173 00:07:59,080 --> 00:08:03,800 those training tokens so they did about

174 00:08:03,800 --> 00:08:07,120 2.75 trillion tokens for their 8 8K

175 00:08:07,120 --> 00:08:10,599 contacts 800 billion tokens for their

176 00:08:10,599 --> 00:08:12,319 128k

177 00:08:12,319 --> 00:08:14,199 contacts window and if you're not

178 00:08:14,199 --> 00:08:16,000 familiar with the contact window that's

179 00:08:16,000 --> 00:08:17,720 how like how much stuff you could put in

180 00:08:17,720 --> 00:08:21,159 that text box if you're on the UI and if

181 00:08:21,159 --> 00:08:24,360 you're using API basically how many

182 00:08:24,360 --> 00:08:26,879 tokens Can you put in there basically

183 00:08:26,879 --> 00:08:30,599 and that's up I would say 28 token

184 00:08:30,599 --> 00:08:34,760 128,000 tokens is like 20 to 30 pages of

185 00:08:34,760 --> 00:08:37,479 text considering on what that

186 00:08:37,479 --> 00:08:40,279 information is and how it's

187 00:08:40,279 --> 00:08:44,480 depicted but I I would say General

188 00:08:44,480 --> 00:08:48,959 from 128k like 125k is like 20 30 pages

189 00:08:48,959 --> 00:08:53,920 of text okay so let's get right into

190 00:08:53,920 --> 00:08:56,519 it oh and and the other thing that I I

191 00:08:56,519 --> 00:08:58,800 forgot to mention is so they released

192 00:08:58,800 --> 00:09:02,640 these kind of AI Studio demos they have

193 00:09:02,640 --> 00:09:05,279 segment anything to which recently came

194 00:09:05,279 --> 00:09:07,519 out seamless transition

195 00:09:07,519 --> 00:09:11,560 so segment anything two creates video

196 00:09:11,560 --> 00:09:13,959 cutouts and other visuals from a video

197 00:09:13,959 --> 00:09:15,720 you upload or you can use example videos

198 00:09:15,720 --> 00:09:17,440 I uploaded my own video which is pretty

199 00:09:17,440 --> 00:09:19,000 cool and I get to play around with it

200 00:09:19,000 --> 00:09:20,399 it's going to be game changer for

201 00:09:20,399 --> 00:09:23,760 creators seamless transition you can do

202 00:09:23,760 --> 00:09:26,920 oh sorry not transition

203 00:09:26,920 --> 00:09:29,760 translation translation I'm not I got

204 00:09:29,760 --> 00:09:31,000 transition I just kind of just looking

205 00:09:31,000 --> 00:09:33,640 over saw trans I was like trans

206 00:09:33,640 --> 00:09:35,760 transition seamless translation you can

207 00:09:35,760 --> 00:09:37,600 translate your voice to any language

208 00:09:37,600 --> 00:09:40,160 using your own voice it has a recording

209 00:09:40,160 --> 00:09:41,880 translate your translates your recording

210 00:09:41,880 --> 00:09:43,200 and kind of mimics your voice which is

211 00:09:43,200 --> 00:09:46,120 pretty cool animated drawings which you

212 00:09:46,120 --> 00:09:48,200 upload your own drawing I didn't do it

213 00:09:48,200 --> 00:09:49,839 my own drawing because I can't draw

214 00:09:49,839 --> 00:09:52,800 unfortunately maybe I could practice but

215 00:09:52,800 --> 00:09:56,040 I have other interest and so I've never

216 00:09:56,040 --> 00:09:57,360 been talented at drawing and I never

217 00:09:57,360 --> 00:09:59,720 worked on it so I can't draw but

218 00:09:59,720 --> 00:10:02,399 basically you can upload your own

219 00:10:02,399 --> 00:10:05,200 drawing and it'll animate for you which

220 00:10:05,200 --> 00:10:07,839 is really cool I think and then there's

221 00:10:07,839 --> 00:10:09,560 this other thing called audio box which

222 00:10:09,560 --> 00:10:11,880 is really cool where you can create an

223 00:10:11,880 --> 00:10:17,000 AI audio and speech generated story and

224 00:10:17,000 --> 00:10:21,480 I asked it to create a Sci-Fi story and

225 00:10:21,480 --> 00:10:23,959 then you can also add in additional

226 00:10:23,959 --> 00:10:26,519 sound effects additional voices and you

227 00:10:26,519 --> 00:10:28,640 can even record your own voice and add

228 00:10:28,640 --> 00:10:30,760 yourself into to the story and so I

229 00:10:30,760 --> 00:10:32,920 added myself I didn't do that much

230 00:10:32,920 --> 00:10:34,480 because I didn't want to spend like a

231 00:10:34,480 --> 00:10:36,760 crazy amount of time on that because I

232 00:10:36,760 --> 00:10:38,240 had to like record my voice I had to

233 00:10:38,240 --> 00:10:41,200 upload it oh you do it all in the UI but

234 00:10:41,200 --> 00:10:42,800 it has to process your voice which is

235 00:10:42,800 --> 00:10:44,839 pretty quick fairly quick and then I was

236 00:10:44,839 --> 00:10:47,279 having some problems with what I was

237 00:10:47,279 --> 00:10:49,079 trying to generate I wanted to say this

238 00:10:49,079 --> 00:10:51,399 is an unauthorized launch repeat this is

239 00:10:51,399 --> 00:10:53,480 an unauthorized launch this will be

240 00:10:53,480 --> 00:10:56,920 reported to the Galaxy Federation or

241 00:10:56,920 --> 00:10:59,440 something like that but it was giving me

242 00:10:59,440 --> 00:11:01,880 an error I say my contacts was too long

243 00:11:01,880 --> 00:11:05,240 even though I was at like 113 characters

244 00:11:05,240 --> 00:11:07,079 and the max is

245 00:11:07,079 --> 00:11:10,800 125 so it's quite sad but I used their

246 00:11:10,800 --> 00:11:13,200 example prompt which is like I have a

247 00:11:13,200 --> 00:11:15,440 great story to tell or something I I

248 00:11:15,440 --> 00:11:16,480 appear in the beginning I'm like a

249 00:11:16,480 --> 00:11:18,320 narrator I have a great story to tell

250 00:11:18,320 --> 00:11:20,560 and then it goes into the story so those

251 00:11:20,560 --> 00:11:22,200 those are really cool and then they also

252 00:11:22,200 --> 00:11:23,200 have an

253 00:11:23,200 --> 00:11:25,600 AI agent platform where you can make

254 00:11:25,600 --> 00:11:28,079 your own AI agents and share them with

255 00:11:28,079 --> 00:11:30,000 your friends and do other things so I'll

256 00:11:30,000 --> 00:11:32,279 be doing that maybe in the next next

257 00:11:32,279 --> 00:11:34,560 episode they launched or they didn't

258 00:11:34,560 --> 00:11:36,000 launch but they released their AI

259 00:11:36,000 --> 00:11:39,000 research paper for llama 3.1 which is

260 00:11:39,000 --> 00:11:40,839 100 pages I think it's

261 00:11:40,839 --> 00:11:43,240 97 which I have to go

262 00:11:43,240 --> 00:11:45,440 through and I I went through it a little

263 00:11:45,440 --> 00:11:48,240 bit but it's just so much

264 00:11:48,240 --> 00:11:51,120 information just the research paper the

265 00:11:51,120 --> 00:11:53,519 AI platform these other snippit things

266 00:11:53,519 --> 00:11:55,480 that they released for their AI demos

267 00:11:55,480 --> 00:11:57,519 it's just too much to put in one episode

268 00:11:57,519 --> 00:11:58,880 like just me explaining it took me like

269 00:11:58,880 --> 00:12:00,360 six

270 00:12:00,360 --> 00:12:03,560 minutes all right so that being said

271 00:12:03,560 --> 00:12:06,360 today we're going to talk about llama

272 00:12:06,360 --> 00:12:10,160 3.1 open open sourcing

273 00:12:10,160 --> 00:12:12,720 AI building with

274 00:12:12,720 --> 00:12:15,399 llama Safety and Security TOS that they

275 00:12:15,399 --> 00:12:18,040 build and then we'll talk a little bit

276 00:12:18,040 --> 00:12:20,440 about the future of AI and llama and

277 00:12:20,440 --> 00:12:22,680 that will include some of these Creator

278 00:12:22,680 --> 00:12:25,000 tools that are built which I think are

279 00:12:25,000 --> 00:12:26,600 really neat and I'm not going to touch

280 00:12:26,600 --> 00:12:28,320 on them too long because they're kind of

281 00:12:28,320 --> 00:12:30,160 small and you can expl them yourself but

282 00:12:30,160 --> 00:12:32,720 I think they're really cool and valuable

283 00:12:32,720 --> 00:12:35,160 to creators long term once these tools

284 00:12:35,160 --> 00:12:38,120 get built in like an Enterprise like

285 00:12:38,120 --> 00:12:41,399 Manor and less of a demo but I could see

286 00:12:41,399 --> 00:12:44,800 this being a huge feature for creators

287 00:12:44,800 --> 00:12:49,040 and on Facebook and Instagram there are

288 00:12:49,040 --> 00:12:51,920 about 200 million creators or people who

289 00:12:51,920 --> 00:12:54,360 consider themselves creators I guess I

290 00:12:54,360 --> 00:12:55,560 would be a Creator I don't know if I

291 00:12:55,560 --> 00:12:57,480 consider myself a Creator I'm more of a

292 00:12:57,480 --> 00:12:59,600 talker

293 00:12:59,600 --> 00:13:01,880 but I I guess by definition I am a

294 00:13:01,880 --> 00:13:03,120 content

295 00:13:03,120 --> 00:13:06,440 creator and these tools would definitely

296 00:13:06,440 --> 00:13:08,600 be useful for someone who is a

297 00:13:08,600 --> 00:13:10,440 professional content creator or someone

298 00:13:10,440 --> 00:13:13,560 trying to get into the realm of content

299 00:13:13,560 --> 00:13:15,120 Creation

300 00:13:15,120 --> 00:13:18,320 with those features I feel like you're

301 00:13:18,320 --> 00:13:21,399 competing with other platforms and it

302 00:13:21,399 --> 00:13:24,639 might not be who could potentially make

303 00:13:24,639 --> 00:13:27,880 you famous like Tik Tok it might be that

304 00:13:27,880 --> 00:13:31,199 who wins the creators with these best AI

305 00:13:31,199 --> 00:13:33,760 features allowing people to easily

306 00:13:33,760 --> 00:13:37,560 create content on your platform you win

307 00:13:37,560 --> 00:13:39,360 the content creators if you win over the

308 00:13:39,360 --> 00:13:41,399 content creators you win the users

309 00:13:41,399 --> 00:13:44,040 because people want to interact with the

310 00:13:44,040 --> 00:13:45,560 content creators and interact with

311 00:13:45,560 --> 00:13:48,680 content and have new refreshing content

312 00:13:48,680 --> 00:13:50,320 hit their

313 00:13:50,320 --> 00:13:54,199 feeds and where that is is where the

314 00:13:54,199 --> 00:13:56,040 content is being created so if you have

315 00:13:56,040 --> 00:13:59,880 the features you've got the users

316 00:13:59,880 --> 00:14:02,880 okay wow what a what

317 00:14:02,880 --> 00:14:07,399 an what an intro 14 minutes okay so

318 00:14:07,399 --> 00:14:11,160 llama 3.1 4 405

319 00:14:11,160 --> 00:14:14,639 billion had an increased contents or

320 00:14:14,639 --> 00:14:16,519 context window

321 00:14:16,519 --> 00:14:18,639 from I think it

322 00:14:18,639 --> 00:14:21,320 was I think it was low like I think it

323 00:14:21,320 --> 00:14:22,759 was I think it was low the original

324 00:14:22,759 --> 00:14:26,079 content window was low context window I

325 00:14:26,079 --> 00:14:28,800 said content so much content talking

326 00:14:28,800 --> 00:14:29,560 about

327 00:14:29,560 --> 00:14:31,160 I'm creating content we're talking about

328 00:14:31,160 --> 00:14:32,279 other people creating content we're

329 00:14:32,279 --> 00:14:34,360 talking about content tools we're

330 00:14:34,360 --> 00:14:36,079 getting all over the place but context

331 00:14:36,079 --> 00:14:39,519 window 128k up

332 00:14:39,519 --> 00:14:42,600 from I think the8 it was 8K you couldn't

333 00:14:42,600 --> 00:14:44,920 put that much stuff in there so 128k is

334 00:14:44,920 --> 00:14:47,519 huge it's not the most I think Google

335 00:14:47,519 --> 00:14:50,759 still leads the pack with that with

336 00:14:50,759 --> 00:14:53,839 their oh my my Google AI is going off

337 00:14:53,839 --> 00:14:57,079 now everywhere actually it's on my my

338 00:14:57,079 --> 00:14:59,680 Google home my phone anyways it was

339 00:14:59,680 --> 00:15:02,440 quite distracting Google leads that in

340 00:15:02,440 --> 00:15:06,079 that regard they are around

341 00:15:06,079 --> 00:15:08,920 500,000 context window and I think they

342 00:15:08,920 --> 00:15:12,000 they put it up too for their like Pro

343 00:15:12,000 --> 00:15:15,240 Ultra I think it's at like over a

344 00:15:15,240 --> 00:15:18,440 million so they're they're on another

345 00:15:18,440 --> 00:15:21,199 level when it comes to content not oh my

346 00:15:21,199 --> 00:15:22,279 gosh they're still going on about this

347 00:15:22,279 --> 00:15:24,160 content context they're on another level

348 00:15:24,160 --> 00:15:26,000 when it comes to context windows but

349 00:15:26,000 --> 00:15:28,360 many people don't need like a million

350 00:15:28,360 --> 00:15:29,600 cont

351 00:15:29,600 --> 00:15:31,519 context Windows like that's a lot a

352 00:15:31,519 --> 00:15:34,519 million tokens is a lot like that's like

353 00:15:34,519 --> 00:15:36,399 books

354 00:15:36,399 --> 00:15:40,079 and for the normal user for for many use

355 00:15:40,079 --> 00:15:44,839 cases 12K is plenty for

356 00:15:44,839 --> 00:15:49,279 like intense research and testing of AI

357 00:15:49,279 --> 00:15:52,519 capabilities this million two million is

358 00:15:52,519 --> 00:15:55,759 is fine that that's great but for the

359 00:15:55,759 --> 00:15:57,920 most part people don't need

360 00:15:57,920 --> 00:16:01,360 that but it's nice to have and so what

361 00:16:01,360 --> 00:16:03,120 does an

362 00:16:03,120 --> 00:16:07,800 increased context window do for the user

363 00:16:07,800 --> 00:16:10,000 well it allows

364 00:16:10,000 --> 00:16:13,279 llama 4 405

365 00:16:13,279 --> 00:16:18,120 to understand more complex topics and

366 00:16:18,120 --> 00:16:20,440 conversations that you're having it can

367 00:16:20,440 --> 00:16:23,120 remember previous interactions and and

368 00:16:23,120 --> 00:16:26,079 and maintain context it can do recall it

369 00:16:26,079 --> 00:16:29,600 can recall what you've previously said

370 00:16:29,600 --> 00:16:32,399 it can generate more accurate and

371 00:16:32,399 --> 00:16:34,000 relevant

372 00:16:34,000 --> 00:16:37,720 responses and just the overall you know

373 00:16:37,720 --> 00:16:40,000 support for like long form text

374 00:16:40,000 --> 00:16:42,279 summarization and document

375 00:16:42,279 --> 00:16:44,600 analysis and so th those are the big

376 00:16:44,600 --> 00:16:47,279 things that this

377 00:16:47,279 --> 00:16:49,680 increased context

378 00:16:49,680 --> 00:16:53,720 window provides to you as a user there's

379 00:16:53,720 --> 00:16:57,680 Now multi- language support

380 00:16:57,680 --> 00:17:00,519 so the the model can now support up to

381 00:17:00,519 --> 00:17:03,440 eight languages I don't know if they

382 00:17:03,440 --> 00:17:05,400 listed out the languages on their

383 00:17:05,400 --> 00:17:08,720 release I'm I'm sure they do English

384 00:17:08,720 --> 00:17:13,000 though and they probably do English

385 00:17:13,000 --> 00:17:15,439 Mandarin but I don't know the languages

386 00:17:15,439 --> 00:17:17,520 so I'm not going to say but I know they

387 00:17:17,520 --> 00:17:20,120 do English I can tell you that for sure

388 00:17:20,120 --> 00:17:21,600 100% I

389 00:17:21,600 --> 00:17:23,240 know

390 00:17:23,240 --> 00:17:26,520 so so they do up to up to eight

391 00:17:26,520 --> 00:17:29,000 languages support so you could type

392 00:17:29,000 --> 00:17:30,559 multiple different languages in your

393 00:17:30,559 --> 00:17:33,320 messages if if it's a supported language

394 00:17:33,320 --> 00:17:34,679 and it might be able to translate

395 00:17:34,679 --> 00:17:38,400 languages never even interacted with

396 00:17:38,400 --> 00:17:40,799 there has been reports of some AI models

397 00:17:40,799 --> 00:17:44,400 being able to do that chat GPT did that

398 00:17:44,400 --> 00:17:49,720 Gemini did that so there is the ability

399 00:17:49,720 --> 00:17:52,200 potentially to do things that the AI

400 00:17:52,200 --> 00:17:54,120 model doesn't know how to doesn't know

401 00:17:54,120 --> 00:17:57,159 that it knows to do so and that that's

402 00:17:57,159 --> 00:17:59,360 happened before multiple times so I

403 00:17:59,360 --> 00:18:00,880 don't think it's out of the realm of

404 00:18:00,880 --> 00:18:02,919 possibilities this model release of the

405 00:18:02,919 --> 00:18:04,520 405

406 00:18:04,520 --> 00:18:06,640 billion is the

407 00:18:06,640 --> 00:18:11,760 first opsource Frontier Model to exist

408 00:18:11,760 --> 00:18:14,200 this hasn't been done before I mentioned

409 00:18:14,200 --> 00:18:18,919 earlier that meta used 16,000

410 00:18:18,919 --> 00:18:23,000 h100s from Nvidia and to put that in

411 00:18:23,000 --> 00:18:25,520 perspective that's

412 00:18:25,520 --> 00:18:30,200 about like $600 Million worth of of

413 00:18:30,200 --> 00:18:31,720 training

414 00:18:31,720 --> 00:18:33,799 equipment that's doesn't count setting

415 00:18:33,799 --> 00:18:37,400 it up or getting it all started but I I

416 00:18:37,400 --> 00:18:39,640 think that it's

417 00:18:39,640 --> 00:18:43,159 about 35,000 per

418 00:18:43,159 --> 00:18:44,799 one

419 00:18:44,799 --> 00:18:46,799 h100 and so

420 00:18:46,799 --> 00:18:49,600 35,000

421 00:18:49,600 --> 00:18:52,600 times

422 00:18:52,600 --> 00:18:54,280 16,000

423 00:18:54,280 --> 00:18:57,320 um yeah it will put you put you around

424 00:18:57,320 --> 00:18:59,559 600 million

425 00:18:59,559 --> 00:19:01,000 which is a lot so they spent basically

426 00:19:01,000 --> 00:19:04,600 $600 million on just buying

427 00:19:04,600 --> 00:19:07,760 gpus for their training clusters and

428 00:19:07,760 --> 00:19:09,360 architecture

429 00:19:09,360 --> 00:19:11,679 support

430 00:19:11,679 --> 00:19:15,360 to release this model free to the

431 00:19:15,360 --> 00:19:19,559 public to empower many users and many

432 00:19:19,559 --> 00:19:22,280 corporations Enterprises startups

433 00:19:22,280 --> 00:19:24,679 universities governments it's a game

434 00:19:24,679 --> 00:19:28,200 changer because many countries have the

435 00:19:28,200 --> 00:19:31,720 technical support to to utilize these

436 00:19:31,720 --> 00:19:35,960 kind of models but they don't have the

437 00:19:35,960 --> 00:19:42,559 infrastructure to put up $600 million

438 00:19:42,559 --> 00:19:46,640 on gpus another you know couple hundred

439 00:19:46,640 --> 00:19:49,760 million on CPUs set up the power

440 00:19:49,760 --> 00:19:53,720 infrastructure running 700 Watts 700 I

441 00:19:53,720 --> 00:19:58,200 think 700 700 Watts I think it's SE man

442 00:19:58,200 --> 00:19:59,799 I don't remember the Watts but I know it

443 00:19:59,799 --> 00:20:02,039 was crazy I think maybe it's 700,000

444 00:20:02,039 --> 00:20:06,360 Watts it's a lot uh that I don't know I

445 00:20:06,360 --> 00:20:08,440 I'll have to look that one up so

446 00:20:08,440 --> 00:20:13,200 see maybe if I ask Meta Meta will

447 00:20:13,200 --> 00:20:15,720 know it's in the research paper but it

448 00:20:15,720 --> 00:20:18,600 is it's quite large so I'll see if meta

449 00:20:18,600 --> 00:20:20,640 knows okay yeah so it's a lot so it's

450 00:20:20,640 --> 00:20:22,360 saying it's

451 00:20:22,360 --> 00:20:26,520 400 it it's 700 kilow per

452 00:20:26,520 --> 00:20:30,480 day which is a crazy amount

453 00:20:30,480 --> 00:20:32,760 700 yeah I'm going to have ai edit this

454 00:20:32,760 --> 00:20:35,080 out this is a long pause here it's got

455 00:20:35,080 --> 00:20:37,159 to be more than that

456 00:20:37,159 --> 00:20:40,400 okay on in short I don't

457 00:20:40,400 --> 00:20:44,880 know I don't know uh it it doesn't man I

458 00:20:44,880 --> 00:20:46,600 wish I didn't even mention it I swear it

459 00:20:46,600 --> 00:20:52,640 was 700 okay so it's trained 16, h100s

460 00:20:52,640 --> 00:20:58,280 running 700 watts each 700 watts each so

461 00:20:58,280 --> 00:21:00,240 that is

462 00:21:00,240 --> 00:21:04,200 is um 700 *

463 00:21:04,200 --> 00:21:06,360 16,

464 00:21:06,360 --> 00:21:09,960 1,1 120,000

465 00:21:09,960 --> 00:21:11,480 Watts

466 00:21:11,480 --> 00:21:15,080 uh while they're running so what is what

467 00:21:15,080 --> 00:21:15,919 is

468 00:21:15,919 --> 00:21:20,320 that I don't know what that is but I

469 00:21:20,320 --> 00:21:23,760 assume that's a lot cuz I don't know

470 00:21:23,760 --> 00:21:26,000 okay so we're asking meta because meta

471 00:21:26,000 --> 00:21:29,039 knows everything right

472 00:21:29,039 --> 00:21:30,480 oh my

473 00:21:30,480 --> 00:21:35,640 gosh okay so how much is

474 00:21:35,640 --> 00:21:39,960 1,120 wats which is the equivalent of

475 00:21:39,960 --> 00:21:41,640 1.12

476 00:21:41,640 --> 00:21:46,840 megawatts and a megawatt is the enough

477 00:21:46,840 --> 00:21:50,520 power for a th000 homes so it's it's

478 00:21:50,520 --> 00:21:53,640 it's the consumption of a small town

479 00:21:53,640 --> 00:21:57,080 like I guess per day small town per day

480 00:21:57,080 --> 00:21:58,679 okay

481 00:21:58,679 --> 00:21:59,720 and that would

482 00:21:59,720 --> 00:22:04,679 be for the going rate of electricity at

483 00:22:04,679 --> 00:22:07,240 13 cents a kilowatt

484 00:22:07,240 --> 00:22:12,080 hour it would cost around $

485 00:22:12,080 --> 00:22:16,240 34,9 44 per day in electricity so

486 00:22:16,240 --> 00:22:18,240 basically spending $50,000 a day in

487 00:22:18,240 --> 00:22:20,720 electricity and it took them like four

488 00:22:20,720 --> 00:22:23,559 five six months or something to train

489 00:22:23,559 --> 00:22:26,520 that AI so I mean they spent like $9

490 00:22:26,520 --> 00:22:29,360 million or something in

491 00:22:29,360 --> 00:22:33,039 electricity just just training thei

492 00:22:33,039 --> 00:22:36,720 training it around so I I try to put

493 00:22:36,720 --> 00:22:39,080 these numbers in perspective because

494 00:22:39,080 --> 00:22:40,840 these things aren't obtainable like the

495 00:22:40,840 --> 00:22:43,159 amount of electricity that is being used

496 00:22:43,159 --> 00:22:47,240 per day the The Upfront cost of getting

497 00:22:47,240 --> 00:22:50,640 the gpus and the CPUs and the getting

498 00:22:50,640 --> 00:22:52,120 the infrastructure set up to run that

499 00:22:52,120 --> 00:22:55,720 amount of electricity to to to the

500 00:22:55,720 --> 00:22:58,960 training clusters is immense

501 00:22:58,960 --> 00:23:00,799 and

502 00:23:00,799 --> 00:23:04,400 even governments can't afford to do

503 00:23:04,400 --> 00:23:06,480 these things only development developed

504 00:23:06,480 --> 00:23:08,960 countries will be able to do so but once

505 00:23:08,960 --> 00:23:11,480 you get the model created it's not that

506 00:23:11,480 --> 00:23:13,919 hard to train the model what you to do

507 00:23:13,919 --> 00:23:17,960 what you want to edit the parameters and

508 00:23:17,960 --> 00:23:21,120 and the weights and and things like that

509 00:23:21,120 --> 00:23:23,080 those things aren't very complicated

510 00:23:23,080 --> 00:23:25,240 compared to the

511 00:23:25,240 --> 00:23:27,919 whole piece of creating a model from

512 00:23:27,919 --> 00:23:29,480 scratch that's what's complicated

513 00:23:29,480 --> 00:23:32,120 getting the training data meta almost

514 00:23:32,120 --> 00:23:36,039 trained three trillion tokens you can't

515 00:23:36,039 --> 00:23:38,600 get that data it's not just lying around

516 00:23:38,600 --> 00:23:42,279 so those things are huge barriers for

517 00:23:42,279 --> 00:23:43,919 other people huge barriers for

518 00:23:43,919 --> 00:23:45,720 individuals that are interested huge

519 00:23:45,720 --> 00:23:49,840 barriers for governments universities

520 00:23:49,840 --> 00:23:51,039 and

521 00:23:51,039 --> 00:23:54,840 businesses and what this does is

522 00:23:54,840 --> 00:23:58,080 allows a level of innovation that you

523 00:23:58,080 --> 00:24:00,240 normally wouldn't see for closed Source

524 00:24:00,240 --> 00:24:04,200 models and this allows people to

525 00:24:04,200 --> 00:24:08,520 use the models at their house at their

526 00:24:08,520 --> 00:24:10,600 company wherever they are they could

527 00:24:10,600 --> 00:24:12,559 just download it and they make it really

528 00:24:12,559 --> 00:24:14,039 easy you can download on hugging face

529 00:24:14,039 --> 00:24:16,880 you can download it on kaggle you can

530 00:24:16,880 --> 00:24:20,200 download it on meta do meta

531 00:24:20,200 --> 00:24:23,159 doai I think Ford download and not only

532 00:24:23,159 --> 00:24:25,039 do they lie to download they also have

533 00:24:25,039 --> 00:24:27,159 ability to use API so they don't have an

534 00:24:27,159 --> 00:24:29,600 API setup because they're not a cloud

535 00:24:29,600 --> 00:24:32,159 provider so they don't have their own

536 00:24:32,159 --> 00:24:34,279 API and you don't have to be a cloud

537 00:24:34,279 --> 00:24:35,520 provider

538 00:24:35,520 --> 00:24:37,880 to offer an API that's not what I'm

539 00:24:37,880 --> 00:24:39,760 saying but they're kind of okay we're

540 00:24:39,760 --> 00:24:40,760 just going to build the best product

541 00:24:40,760 --> 00:24:42,120 we're going to release it and we're

542 00:24:42,120 --> 00:24:43,600 going to have partners that support it

543 00:24:43,600 --> 00:24:48,360 so they have partners that support the

544 00:24:48,360 --> 00:24:54,640 AI via cloud or you can just use a API

545 00:24:54,640 --> 00:24:57,399 either or like Huggy face has an API for

546 00:24:57,399 --> 00:25:00,320 meta Google

547 00:25:00,440 --> 00:25:03,600 AWS data bricks they all have offerings

548 00:25:03,600 --> 00:25:06,200 to allow you to use the API not API but

549 00:25:06,200 --> 00:25:08,320 the the model on their

550 00:25:08,320 --> 00:25:11,480 infrastructure and they have just some

551 00:25:11,480 --> 00:25:13,919 general support but the two places that

552 00:25:13,919 --> 00:25:16,600 have support top to bottom which would

553 00:25:16,600 --> 00:25:18,039 be like real-time

554 00:25:18,039 --> 00:25:21,679 inferencing batch inference fine-tuning

555 00:25:21,679 --> 00:25:24,360 model evaluation knowledge base

556 00:25:24,360 --> 00:25:27,600 contextual pre-training Safeguard rails

557 00:25:27,600 --> 00:25:28,840 synthetic data

558 00:25:28,840 --> 00:25:31,840 generation and

559 00:25:31,840 --> 00:25:35,200 distillation recipe would be data bricks

560 00:25:35,200 --> 00:25:37,320 and Nvidia those are the only ones that

561 00:25:37,320 --> 00:25:39,559 have a full Suite of

562 00:25:39,559 --> 00:25:44,279 offerings and why that's important is

563 00:25:44,279 --> 00:25:47,320 for the first time you can edit these

564 00:25:47,320 --> 00:25:49,520 models yourself and you can fully

565 00:25:49,520 --> 00:25:51,039 customize it with your own data with

566 00:25:51,039 --> 00:25:52,360 your own training you can distill them

567 00:25:52,360 --> 00:25:54,480 how you wish you can do anything you

568 00:25:54,480 --> 00:25:56,760 want it's Apache 2.0 that's the

569 00:25:56,760 --> 00:25:57,880 licensing by the way if you're not

570 00:25:57,880 --> 00:26:00,399 familiar and so these Partners had to

571 00:26:00,399 --> 00:26:02,600 build out these features to be able to

572 00:26:02,600 --> 00:26:04,559 offer these

573 00:26:04,559 --> 00:26:08,200 things and the ones that really were

574 00:26:08,200 --> 00:26:09,279 interested in building out these

575 00:26:09,279 --> 00:26:12,960 features to to fully utilize this model

576 00:26:12,960 --> 00:26:17,120 was datab bricks and Nvidia and what are

577 00:26:17,120 --> 00:26:18,440 they calling this they're calling this

578 00:26:18,440 --> 00:26:21,159 llama stack API so it's a standardized

579 00:26:21,159 --> 00:26:24,399 inference that allows the developers to

580 00:26:24,399 --> 00:26:26,760 you know create applications on top of

581 00:26:26,760 --> 00:26:28,799 3.1

582 00:26:28,799 --> 00:26:32,880 and so it will provide a predefined set

583 00:26:32,880 --> 00:26:35,399 of you know tools and functions to allow

584 00:26:35,399 --> 00:26:39,360 like you to easier integrate into

585 00:26:39,360 --> 00:26:41,240 various projects I'm getting some

586 00:26:41,240 --> 00:26:43,720 potential some mic feedback I think but

587 00:26:43,720 --> 00:26:45,520 it just allows you to build custom

588 00:26:45,520 --> 00:26:49,200 applications on top of llama Or you know

589 00:26:49,200 --> 00:26:50,880 with llama either

590 00:26:50,880 --> 00:26:56,640 or and it just allows you to utilize the

591 00:26:56,640 --> 00:26:59,440 initial knowledge base with that you can

592 00:26:59,440 --> 00:27:03,159 also download I think they call it the

593 00:27:03,159 --> 00:27:05,440 let me look at the GitHub they within

594 00:27:05,440 --> 00:27:08,080 the Llama stack you can

595 00:27:08,080 --> 00:27:12,399 download your safety guide rails and

596 00:27:12,399 --> 00:27:17,120 llam guard in this in this like stack so

597 00:27:17,120 --> 00:27:21,880 lamard it has this like ability to

598 00:27:21,880 --> 00:27:26,120 detect detect and prevent potential

599 00:27:26,120 --> 00:27:29,120 misuse and so this misuse could be hate

600 00:27:29,120 --> 00:27:32,480 speech harassment explicit content and

601 00:27:32,480 --> 00:27:33,720 then they have this thing called prompt

602 00:27:33,720 --> 00:27:37,559 guard which tries to prevent people to

603 00:27:37,559 --> 00:27:39,640 gain the system so a lot of times with

604 00:27:39,640 --> 00:27:42,880 these close Source models you can and

605 00:27:42,880 --> 00:27:44,360 other Foundation models and other

606 00:27:44,360 --> 00:27:47,159 offerings you can force the model to do

607 00:27:47,159 --> 00:27:49,120 things it wouldn't normally do so like

608 00:27:49,120 --> 00:27:51,720 if you wanted to know how to make bombs

609 00:27:51,720 --> 00:27:54,720 or to do things that you typically

610 00:27:54,720 --> 00:27:56,960 wouldn't allow a system to look up on

611 00:27:56,960 --> 00:27:59,480 its own it originally it will say oh I

612 00:27:59,480 --> 00:28:01,120 can't do that blah blah blah it's

613 00:28:01,120 --> 00:28:03,320 against my standards but you can kind of

614 00:28:03,320 --> 00:28:04,440 game it

615 00:28:04,440 --> 00:28:06,799 into telling you information it normally

616 00:28:06,799 --> 00:28:08,799 wouldn't tell you so you have to word

617 00:28:08,799 --> 00:28:12,279 things a certain way and and and confuse

618 00:28:12,279 --> 00:28:15,559 the AI and the AI eventually will tell

619 00:28:15,559 --> 00:28:19,240 you the things that you want to know

620 00:28:19,240 --> 00:28:20,760 when it's not supposed to tell you so

621 00:28:20,760 --> 00:28:23,320 prompt guard tries to prevent and flag

622 00:28:23,320 --> 00:28:24,960 and

623 00:28:24,960 --> 00:28:26,840 categorize the prompts that you're

624 00:28:26,840 --> 00:28:29,519 sending to

625 00:28:29,519 --> 00:28:31,519 llama and if it detects that you're

626 00:28:31,519 --> 00:28:35,640 trying to misuse the AI by you

627 00:28:35,640 --> 00:28:39,440 know confusing it and crafting these

628 00:28:39,440 --> 00:28:43,960 like special prompts to jailbreak llama

629 00:28:43,960 --> 00:28:46,200 it will prompt you that there's like

630 00:28:46,200 --> 00:28:48,120 security threats and potentially

631 00:28:48,120 --> 00:28:50,399 somebody's trying to abuse the

632 00:28:50,399 --> 00:28:53,480 product and so it will block you from

633 00:28:53,480 --> 00:28:57,080 doing so so these are important because

634 00:28:57,080 --> 00:29:01,080 it's gives companies Peace of Mind gives

635 00:29:01,080 --> 00:29:05,320 safety you there's some some criticisms

636 00:29:05,320 --> 00:29:08,200 regarding open source AI with the being

637 00:29:08,200 --> 00:29:10,039 that you know there's it gives Bad

638 00:29:10,039 --> 00:29:12,360 actors a chance to use these products

639 00:29:12,360 --> 00:29:15,480 and and do malicious in and act in a

640 00:29:15,480 --> 00:29:16,640 malicious

641 00:29:16,640 --> 00:29:20,080 way and this will help protect against

642 00:29:20,080 --> 00:29:23,880 these Bad actors and so

643 00:29:23,880 --> 00:29:26,799 lamard and promp guard are already

644 00:29:26,799 --> 00:29:29,679 integrated into the

645 00:29:29,679 --> 00:29:32,200 model what I'm talking about is if you

646 00:29:32,200 --> 00:29:35,279 want to customize prompt guard and llama

647 00:29:35,279 --> 00:29:38,039 guard yourself for your business or

648 00:29:38,039 --> 00:29:39,919 whatever you're trying to do you can do

649 00:29:39,919 --> 00:29:42,200 so and you can integrate that into your

650 00:29:42,200 --> 00:29:45,720 model it will you know prevent harm and

651 00:29:45,720 --> 00:29:48,279 and people misusing this

652 00:29:48,279 --> 00:29:52,120 product prevents misinformation and

653 00:29:52,120 --> 00:29:53,760 malicious

654 00:29:53,760 --> 00:29:56,679 content and it makes sure that llama is

655 00:29:56,679 --> 00:29:59,080 used ethnically and

656 00:29:59,080 --> 00:30:00,960 responsibly and what I was saying is

657 00:30:00,960 --> 00:30:03,720 that closed Source models are really

658 00:30:03,720 --> 00:30:05,640 like oh you know if it's open source

659 00:30:05,640 --> 00:30:07,600 it's dangerous anyone can do anything

660 00:30:07,600 --> 00:30:10,240 with it and people are going to do bad

661 00:30:10,240 --> 00:30:12,919 things bad things and they mainly focus

662 00:30:12,919 --> 00:30:14,080 on

663 00:30:14,080 --> 00:30:17,159 people attacking the government and

664 00:30:17,159 --> 00:30:19,279 doing these things that are

665 00:30:19,279 --> 00:30:22,720 normally only I would say categorized

666 00:30:22,720 --> 00:30:25,679 for sophisticated actors sophisticated

667 00:30:25,679 --> 00:30:27,440 Bad actors like governments or large

668 00:30:27,440 --> 00:30:29,039 organizations terrorist organizations

669 00:30:29,039 --> 00:30:30,399 that have more resources that are

670 00:30:30,399 --> 00:30:32,279 typically backed by another government

671 00:30:32,279 --> 00:30:33,760 and they're using the terrorist

672 00:30:33,760 --> 00:30:37,039 organization as like a shell company

673 00:30:37,039 --> 00:30:39,320 those groups of

674 00:30:39,320 --> 00:30:41,760 people are going to be doing bad things

675 00:30:41,760 --> 00:30:45,840 no matter what and the perspective of

676 00:30:45,840 --> 00:30:48,799 Mark Zuckerberg and

677 00:30:48,799 --> 00:30:51,279 others is if you open source these

678 00:30:51,279 --> 00:30:53,640 things you can potentially

679 00:30:53,640 --> 00:30:58,039 protect governments and companies and

680 00:30:58,039 --> 00:31:00,519 universities from Bad actors because

681 00:31:00,519 --> 00:31:01,799 they'll have this

682 00:31:01,799 --> 00:31:03,919 sophisticated AI model that they'll be

683 00:31:03,919 --> 00:31:07,000 able to launch on their own systems to

684 00:31:07,000 --> 00:31:10,360 prevent issues to prevent attacks and

685 00:31:10,360 --> 00:31:12,279 that's how it's going to be where you

686 00:31:12,279 --> 00:31:13,960 have this open source sophisticated

687 00:31:13,960 --> 00:31:17,480 model that you can just update on your

688 00:31:17,480 --> 00:31:20,120 system and your infrastructure and then

689 00:31:20,120 --> 00:31:21,960 voila you're protected and that's that's

690 00:31:21,960 --> 00:31:25,279 Mark's Vision where he feels that he

691 00:31:25,279 --> 00:31:27,360 doesn't want to be forced on what

692 00:31:27,360 --> 00:31:30,159 features he can make because that's some

693 00:31:30,159 --> 00:31:33,360 issues that he's had with apple where

694 00:31:33,360 --> 00:31:36,480 meta tried to launch features that will

695 00:31:36,480 --> 00:31:39,519 be useful for their users Apple said hey

696 00:31:39,519 --> 00:31:40,480 like this isn't really in your

697 00:31:40,480 --> 00:31:41,799 wheelhouse we're not going to allow you

698 00:31:41,799 --> 00:31:44,240 to do though do these

699 00:31:44,240 --> 00:31:46,480 launches meta can't do the launches

700 00:31:46,480 --> 00:31:48,960 anymore because they're forced into

701 00:31:48,960 --> 00:31:51,519 these closed Source models for mobile

702 00:31:51,519 --> 00:31:54,559 phones cuz close Clos model one meta

703 00:31:54,559 --> 00:31:56,159 wants to have the opportunity to make

704 00:31:56,159 --> 00:32:00,360 sure that the open source model wins and

705 00:32:00,360 --> 00:32:03,120 meta is one of the Front Runners in this

706 00:32:03,120 --> 00:32:04,600 technology as they don't want to be

707 00:32:04,600 --> 00:32:06,760 reliant on a closed Source model from

708 00:32:06,760 --> 00:32:09,559 another company and they want to build a

709 00:32:09,559 --> 00:32:13,039 wonderful Atmos not atmosphere ecosystem

710 00:32:13,039 --> 00:32:15,840 to allow creators and companies and and

711 00:32:15,840 --> 00:32:17,760 innovators to use their product and

712 00:32:17,760 --> 00:32:19,799 interact with their brand and then it

713 00:32:19,799 --> 00:32:23,880 also allows them to be a front runner in

714 00:32:23,880 --> 00:32:26,240 governmental conversations regarding

715 00:32:26,240 --> 00:32:28,639 compliance and regulation and making

716 00:32:28,639 --> 00:32:31,279 sure that the US is a leader in this

717 00:32:31,279 --> 00:32:34,679 revolution like revolutionary technology

718 00:32:34,679 --> 00:32:39,600 that is going to shape the world in a

719 00:32:39,600 --> 00:32:43,559 positive manner and being in control of

720 00:32:43,559 --> 00:32:45,639 those conversations definitely gives

721 00:32:45,639 --> 00:32:48,360 meta an edge and then also to prevent

722 00:32:48,360 --> 00:32:52,000 them from having an issue with not

723 00:32:52,000 --> 00:32:56,039 having the best and leading technology

724 00:32:56,039 --> 00:32:58,919 in this sector sector

725 00:32:58,919 --> 00:33:00,159 with them being attacked by

726 00:33:00,159 --> 00:33:02,039 sophisticated actors where they don't

727 00:33:02,039 --> 00:33:04,519 have the model and so now they have

728 00:33:04,519 --> 00:33:06,600 their own Cutting Edge model that

729 00:33:06,600 --> 00:33:08,519 competes and wins with the best closed

730 00:33:08,519 --> 00:33:10,840 Source models around right now in all

731 00:33:10,840 --> 00:33:13,600 areas they're 8 billion they're 405

732 00:33:13,600 --> 00:33:15,639 they're 70 billion they're all winning

733 00:33:15,639 --> 00:33:18,799 across the board pretty much from in

734 00:33:18,799 --> 00:33:21,399 their 405 billion did pretty well

735 00:33:21,399 --> 00:33:24,960 against open Ai and their 70 and8

736 00:33:24,960 --> 00:33:26,799 billion parameter models killed the

737 00:33:26,799 --> 00:33:29,039 other competition

738 00:33:29,039 --> 00:33:32,240 they won like majority of the time and

739 00:33:32,240 --> 00:33:35,919 so it gives them an opportunity to one

740 00:33:35,919 --> 00:33:37,960 innovate their products protect

741 00:33:37,960 --> 00:33:41,159 themselves against s sophisticated

742 00:33:41,159 --> 00:33:43,279 actors lead in

743 00:33:43,279 --> 00:33:45,440 these governmental conversations

744 00:33:45,440 --> 00:33:48,159 regarding reg regulation and

745 00:33:48,159 --> 00:33:50,720 compliance embrace the open source model

746 00:33:50,720 --> 00:33:52,600 and and allow people to innovate and

747 00:33:52,600 --> 00:33:56,200 help out millions of people because I'm

748 00:33:56,200 --> 00:33:57,200 accounting the people that work for

749 00:33:57,200 --> 00:33:58,399 these small business business and these

750 00:33:58,399 --> 00:34:01,039 corporations and and these

751 00:34:01,039 --> 00:34:03,880 universities and Inspire and Inspire

752 00:34:03,880 --> 00:34:06,760 many people I mean they have 300 million

753 00:34:06,760 --> 00:34:10,399 downloads already for

754 00:34:10,399 --> 00:34:14,679 llama 3.1 and and llama 2 and just llama

755 00:34:14,679 --> 00:34:16,440 in general and they want to be one of

756 00:34:16,440 --> 00:34:19,280 the most used AI assistants in the world

757 00:34:19,280 --> 00:34:21,040 which they're on track to do before the

758 00:34:21,040 --> 00:34:22,960 end of the year according to Mark recent

759 00:34:22,960 --> 00:34:24,679 his recent interview of like three days

760 00:34:24,679 --> 00:34:27,280 ago with Bloomberg because they have

761 00:34:27,280 --> 00:34:30,159 this a AI assistant in WhatsApp they

762 00:34:30,159 --> 00:34:32,159 have it in Instagram they have it in

763 00:34:32,159 --> 00:34:34,960 Facebook they have it in messenger and

764 00:34:34,960 --> 00:34:37,200 so you can constantly interact with this

765 00:34:37,200 --> 00:34:41,240 AI wherever you are in any of their

766 00:34:41,240 --> 00:34:45,119 ecosystem of apps it doesn't no it would

767 00:34:45,119 --> 00:34:46,919 it would be cool if it would it would

768 00:34:46,919 --> 00:34:49,079 have the context of all your

769 00:34:49,079 --> 00:34:51,040 conversations that you have with the AI

770 00:34:51,040 --> 00:34:52,760 on these different apps but it

771 00:34:52,760 --> 00:34:56,119 doesn't but I still find it super useful

772 00:34:56,119 --> 00:35:00,000 and I love it honestly it's great and I

773 00:35:00,000 --> 00:35:02,359 use it all the time on Instagram and

774 00:35:02,359 --> 00:35:04,240 they have those agents those AI agents

775 00:35:04,240 --> 00:35:06,520 that you can talk to I talk to the

776 00:35:06,520 --> 00:35:08,160 cooking guy I think cooking with Max all

777 00:35:08,160 --> 00:35:11,000 the time for different cooking recipes

778 00:35:11,000 --> 00:35:13,320 those things are super useful where you

779 00:35:13,320 --> 00:35:15,320 ask it you know what can I cook with

780 00:35:15,320 --> 00:35:16,920 these ingredients just give it give it

781 00:35:16,920 --> 00:35:19,400 five ingredients and and go from there

782 00:35:19,400 --> 00:35:21,680 and you know you can plan a picnic or a

783 00:35:21,680 --> 00:35:23,760 shuie board or with certain different

784 00:35:23,760 --> 00:35:26,880 wines and it's just great honestly I

785 00:35:26,880 --> 00:35:28,880 love it I love it I love every bit of it

786 00:35:28,880 --> 00:35:32,240 so with with the prompt guard and these

787 00:35:32,240 --> 00:35:35,280 other pieces and I T touched on it a bit

788 00:35:35,280 --> 00:35:36,800 with the open

789 00:35:36,800 --> 00:35:39,960 source open source

790 00:35:39,960 --> 00:35:45,280 allows meta to control the ecosystem it

791 00:35:45,280 --> 00:35:48,520 allows them to become the industry

792 00:35:48,520 --> 00:35:51,960 standard and the example he uses often

793 00:35:51,960 --> 00:35:54,400 is the Linux example where a lot of

794 00:35:54,400 --> 00:35:55,560 companies were

795 00:35:55,560 --> 00:35:59,440 building these closed operating systems

796 00:35:59,440 --> 00:36:02,280 and then Linux came about and

797 00:36:02,280 --> 00:36:04,920 became popular because you could do

798 00:36:04,920 --> 00:36:07,520 whatever you want with it it was easy to

799 00:36:07,520 --> 00:36:10,280 use it wasn't as secure or as fast as

800 00:36:10,280 --> 00:36:13,200 the closed models but since it was easy

801 00:36:13,200 --> 00:36:17,040 to use in in open source the innovation

802 00:36:17,040 --> 00:36:20,960 of those models were or not models but

803 00:36:20,960 --> 00:36:22,960 of Linux was much faster than the Clos

804 00:36:22,960 --> 00:36:27,079 Source models and so in that regard

805 00:36:27,079 --> 00:36:28,760 Linux

806 00:36:28,760 --> 00:36:30,880 one in that

807 00:36:30,880 --> 00:36:33,040 conversation and the same could be said

808 00:36:33,040 --> 00:36:36,000 as like there's apple and then there's

809 00:36:36,000 --> 00:36:39,040 Windows Windows one the computer windows

810 00:36:39,040 --> 00:36:41,040 isn't necessarily open

811 00:36:41,040 --> 00:36:44,480 source but it's much more open than and

812 00:36:44,480 --> 00:36:46,560 this card I would say open ecosystem not

813 00:36:46,560 --> 00:36:49,680 open source Windows is closed Source but

814 00:36:49,680 --> 00:36:52,280 their ecosystem is open where you can

815 00:36:52,280 --> 00:36:54,800 basically use any app you

816 00:36:54,800 --> 00:36:58,000 can do as you wish where as apple is

817 00:36:58,000 --> 00:37:00,200 kind of closed where you know if you

818 00:37:00,200 --> 00:37:01,599 want to have your messages linked you

819 00:37:01,599 --> 00:37:05,319 got to have so and so phone and and

820 00:37:05,319 --> 00:37:09,440 things like that when you are kind of

821 00:37:09,440 --> 00:37:11,960 tied into the ecosystem with the Walled

822 00:37:11,960 --> 00:37:14,440 Garden window is not so

823 00:37:14,440 --> 00:37:18,359 much and I'm glad that Windows won I

824 00:37:18,359 --> 00:37:20,000 like Apple though I haven't I I'm

825 00:37:20,000 --> 00:37:23,000 literally recording this on my Mac so I

826 00:37:23,000 --> 00:37:24,280 don't want people to be like oh you're

827 00:37:24,280 --> 00:37:26,520 you're such an apple hater no I I like

828 00:37:26,520 --> 00:37:28,599 Apple quite a bit

829 00:37:28,599 --> 00:37:30,839 just there are certain C you know pros

830 00:37:30,839 --> 00:37:35,200 and cons of closed Source open source I

831 00:37:35,200 --> 00:37:37,920 would definitely choose open source for

832 00:37:37,920 --> 00:37:42,480 sure as it's the level of innovation

833 00:37:42,480 --> 00:37:44,280 that you'll get is probably not

834 00:37:44,280 --> 00:37:47,480 obtainable with close Source longterm it

835 00:37:47,480 --> 00:37:49,520 and I keep bringing this up if Fosters

836 00:37:49,520 --> 00:37:51,200 you know this Innovation and

837 00:37:51,200 --> 00:37:53,319 collaboration and and also in some some

838 00:37:53,319 --> 00:37:55,599 circumstances it puts a higher scrutiny

839 00:37:55,599 --> 00:37:58,599 on your work so when you're releasing

840 00:37:58,599 --> 00:38:01,960 your product to say users in your user

841 00:38:01,960 --> 00:38:04,400 group you can really only get user

842 00:38:04,400 --> 00:38:08,319 feedback from your QA or Q QE your

843 00:38:08,319 --> 00:38:10,040 developers and your

844 00:38:10,040 --> 00:38:15,359 users on a close model open source model

845 00:38:15,359 --> 00:38:18,680 potentially you can have feedback on

846 00:38:18,680 --> 00:38:22,640 your code the features the bugs security

847 00:38:22,640 --> 00:38:25,599 all that stuff from millions of people

848 00:38:25,599 --> 00:38:27,280 like 300 million people have downloaded

849 00:38:27,280 --> 00:38:29,640 l l so they'll all be able to interact

850 00:38:29,640 --> 00:38:30,680 with it they'll all be able to give

851 00:38:30,680 --> 00:38:31,800 suggestions they'll be able to make

852 00:38:31,800 --> 00:38:34,920 their own versions of of llama or

853 00:38:34,920 --> 00:38:36,760 whatever you want to call it I don't

854 00:38:36,760 --> 00:38:39,680 know they can rename it it's Apache 2.0

855 00:38:39,680 --> 00:38:42,520 which means that you can use and I had

856 00:38:42,520 --> 00:38:45,680 to make a correction it originally in

857 00:38:45,680 --> 00:38:48,520 one of Mark's interviews he said that he

858 00:38:48,520 --> 00:38:51,560 would allow you know small companies to

859 00:38:51,560 --> 00:38:53,200 use it for free and and if they're

860 00:38:53,200 --> 00:38:55,720 larger he'd want them to pay but they

861 00:38:55,720 --> 00:38:59,160 decided to do Apache 2.0 pach 2.0

862 00:38:59,160 --> 00:39:02,240 basically means that they can do

863 00:39:02,240 --> 00:39:05,400 whatever they want it's open source as

864 00:39:05,400 --> 00:39:07,400 much as it gets you can use it for

865 00:39:07,400 --> 00:39:09,359 commercial reasons

866 00:39:09,359 --> 00:39:13,240 education you can edit the the code you

867 00:39:13,240 --> 00:39:16,599 can add things remove things rename it

868 00:39:16,599 --> 00:39:20,160 repurpose it reface it whatever you want

869 00:39:20,160 --> 00:39:22,880 it's fine so it's free use it's Lally

870 00:39:22,880 --> 00:39:25,839 free use so with it being free use many

871 00:39:25,839 --> 00:39:27,400 people 300 million people have

872 00:39:27,400 --> 00:39:33,560 downloaded it and of that like almost

873 00:39:33,560 --> 00:39:36,200 4,000 you know 3500

874 00:39:36,200 --> 00:39:39,480 Enterprises have downloaded or use not

875 00:39:39,480 --> 00:39:43,040 downloaded using llama for

876 00:39:43,040 --> 00:39:47,680 their work and they company and so when

877 00:39:47,680 --> 00:39:49,480 you do those things you get a heighten

878 00:39:49,480 --> 00:39:51,760 scrutiny on everything heighten scre on

879 00:39:51,760 --> 00:39:55,599 your code the security features and

880 00:39:55,599 --> 00:39:58,240 people can provide suggest questions

881 00:39:58,240 --> 00:40:03,000 either with a fork or a push request or

882 00:40:03,000 --> 00:40:04,400 all these things I don't know how that

883 00:40:04,400 --> 00:40:08,599 works for these large open open source

884 00:40:08,599 --> 00:40:10,000 platforms but I know

885 00:40:10,000 --> 00:40:11,560 for

886 00:40:11,560 --> 00:40:13,880 libraries they'll allow you if you're a

887 00:40:13,880 --> 00:40:15,319 contributor you can have to become a

888 00:40:15,319 --> 00:40:16,760 contributor and you can contribute to

889 00:40:16,760 --> 00:40:20,520 the code but I'm sure you could be able

890 00:40:20,520 --> 00:40:22,359 you you should be able to offer

891 00:40:22,359 --> 00:40:26,920 suggestions or or some comments and they

892 00:40:26,920 --> 00:40:29,560 have some general some general stuff

893 00:40:29,560 --> 00:40:31,079 that they're putting in here like I'm

894 00:40:31,079 --> 00:40:34,000 looking at the GitHub right now um this

895 00:40:34,000 --> 00:40:37,240 one said this person it's got 20 likes

896 00:40:37,240 --> 00:40:38,920 it says you know General opinion I feel

897 00:40:38,920 --> 00:40:40,359 like there there's a need for a three

898 00:40:40,359 --> 00:40:42,119 billion or 4 billion base model that

899 00:40:42,119 --> 00:40:43,880 makes it easier faster to do inference

900 00:40:43,880 --> 00:40:47,480 on low in Hardware or a CPU which is my

901 00:40:47,480 --> 00:40:49,599 goal the benefits would be less space

902 00:40:49,599 --> 00:40:51,839 reduces resource need and the cost of

903 00:40:51,839 --> 00:40:54,359 accuracy quality but fine tuning that

904 00:40:54,359 --> 00:40:56,520 for the in users case use case would be

905 00:40:56,520 --> 00:40:58,880 a GameChanger if quantied of models

906 00:40:58,880 --> 00:40:59,800 would be

907 00:40:59,800 --> 00:41:02,520 smaller and maybe more usable for text

908 00:41:02,520 --> 00:41:04,240 generation however it still could be

909 00:41:04,240 --> 00:41:06,760 very good at finding probabilities for

910 00:41:06,760 --> 00:41:10,240 tokens and use for applications such as

911 00:41:10,240 --> 00:41:13,280 compression of texts logs binary

912 00:41:13,280 --> 00:41:14,839 structured text yielding high

913 00:41:14,839 --> 00:41:16,640 compression ratio so that's just like a

914 00:41:16,640 --> 00:41:18,240 suggestion they put like I think this

915 00:41:18,240 --> 00:41:21,720 person's working on the yeah so this

916 00:41:21,720 --> 00:41:23,520 person put that a comment in there and

917 00:41:23,520 --> 00:41:25,280 then the author of The

918 00:41:25,280 --> 00:41:28,720 GitHub uh for the Llama St back said you

919 00:41:28,720 --> 00:41:30,520 know thank you for the comment I

920 00:41:30,520 --> 00:41:31,440 understand there a need for smaller

921 00:41:31,440 --> 00:41:34,680 models in this release we provided um

922 00:41:34,680 --> 00:41:38,599 fb8 like I guess that is their you know

923 00:41:38,599 --> 00:41:42,119 8 billion model and our our largest 4.5

924 00:41:42,119 --> 00:41:43,760 billion in addition we released both

925 00:41:43,760 --> 00:41:45,480 Dynamic inference code as well as

926 00:41:45,480 --> 00:41:48,560 inference code for persistent weights

927 00:41:48,560 --> 00:41:50,160 and implementation please check them out

928 00:41:50,160 --> 00:41:52,680 so he's basically saying hey you know

929 00:41:52,680 --> 00:41:55,280 we've offered our small Model 8 billion

930 00:41:55,280 --> 00:41:59,040 and our largest at 4.45 million and you

931 00:41:59,040 --> 00:42:04,079 can edit the weights and the code um

932 00:42:04,079 --> 00:42:06,760 please check out the docs and you could

933 00:42:06,760 --> 00:42:10,440 you can change the parameters this one

934 00:42:10,440 --> 00:42:14,359 got 15 15 down votes wow all right

935 00:42:14,359 --> 00:42:15,599 they're killing

936 00:42:15,599 --> 00:42:18,240 them wow

937 00:42:18,240 --> 00:42:20,800 okay so that that's open source and you

938 00:42:20,800 --> 00:42:22,640 know open source is just I feel like

939 00:42:22,640 --> 00:42:26,200 better where you know it gives an

940 00:42:26,200 --> 00:42:27,880 opportunity for government

941 00:42:27,880 --> 00:42:30,720 universities individuals to really play

942 00:42:30,720 --> 00:42:32,119 around with this technology and

943 00:42:32,119 --> 00:42:33,480 Technology they would never be able to

944 00:42:33,480 --> 00:42:36,480 interact with at this level without

945 00:42:36,480 --> 00:42:38,839 having immense amount of resources and

946 00:42:38,839 --> 00:42:40,520 you never know what comes up from this

947 00:42:40,520 --> 00:42:43,079 like there might be some

948 00:42:43,079 --> 00:42:46,040 brilliant Pie in the Sky idea that's

949 00:42:46,040 --> 00:42:49,599 possible because of this this AI or how

950 00:42:49,599 --> 00:42:51,880 much of the coding will you have to do

951 00:42:51,880 --> 00:42:53,920 when you're building a new product like

952 00:42:53,920 --> 00:42:55,440 how fast can you start innovating when

953 00:42:55,440 --> 00:42:56,680 you don't have to build everything from

954 00:42:56,680 --> 00:42:59,800 scratch you can ask an AI system to

955 00:42:59,800 --> 00:43:01,760 build it for you or who who's the person

956 00:43:01,760 --> 00:43:03,640 just throwing stuff against the wall and

957 00:43:03,640 --> 00:43:05,880 stuff might stick you never know when

958 00:43:05,880 --> 00:43:07,240 you put the hand when you put something

959 00:43:07,240 --> 00:43:10,440 in the hands of 300 million people

960 00:43:10,440 --> 00:43:12,880 something is bound to happen that is

961 00:43:12,880 --> 00:43:15,520 potentially spectacular and I'm excited

962 00:43:15,520 --> 00:43:18,960 what what comes of it all and and I am

963 00:43:18,960 --> 00:43:20,760 thrilled that they're they're doing

964 00:43:20,760 --> 00:43:23,160 these things and it's just really

965 00:43:23,160 --> 00:43:26,480 cool that we're here and in this space

966 00:43:26,480 --> 00:43:28,160 because this this open source is going

967 00:43:28,160 --> 00:43:31,200 to push closed source to the brink and

968 00:43:31,200 --> 00:43:33,640 they're they're going to get pushed

969 00:43:33,640 --> 00:43:35,359 closed source is going to make some big

970 00:43:35,359 --> 00:43:37,520 releases open source is going to push

971 00:43:37,520 --> 00:43:39,400 right back and so there's is going to be

972 00:43:39,400 --> 00:43:41,880 this constant tug of war and this tug of

973 00:43:41,880 --> 00:43:44,119 war and competition will bring out the

974 00:43:44,119 --> 00:43:46,319 best of both parties and I just am

975 00:43:46,319 --> 00:43:50,319 excited to see what comes of it all and

976 00:43:50,319 --> 00:43:54,319 I am very very happy uh where the

977 00:43:54,319 --> 00:43:56,599 future's going and it it's it's honestly

978 00:43:56,599 --> 00:43:59,200 great and and uh I couldn't I couldn't

979 00:43:59,200 --> 00:44:00,880 be more inspired about this whole thing

980 00:44:00,880 --> 00:44:02,720 it's awesome it's

981 00:44:02,720 --> 00:44:05,680 awesome and so with that said I'm going

982 00:44:05,680 --> 00:44:08,040 to transition over to these small AI

983 00:44:08,040 --> 00:44:10,720 demos we're near the end of the episode

984 00:44:10,720 --> 00:44:12,880 so I'm going to share my screen and I

985 00:44:12,880 --> 00:44:14,400 have an

986 00:44:14,400 --> 00:44:17,359 alien I don't know a little alien guy

987 00:44:17,359 --> 00:44:19,000 he's got one eye he's

988 00:44:19,000 --> 00:44:24,319 blue and I animated him so I share share

989 00:44:24,319 --> 00:44:28,400 screen man it's a lot of here we go

990 00:44:28,400 --> 00:44:30,760 share screen and audio I don't have any

991 00:44:30,760 --> 00:44:33,839 audio but so you can animate them and so

992 00:44:33,839 --> 00:44:36,800 I have this little alien dude in the

993 00:44:36,800 --> 00:44:41,200 funny category and so I can I can change

994 00:44:41,200 --> 00:44:43,960 the I can change the little animation

995 00:44:43,960 --> 00:44:47,200 and so this is like a boxing thing um

996 00:44:47,200 --> 00:44:49,559 what's this

997 00:44:49,559 --> 00:44:52,280 one this is kind of cool this kind of

998 00:44:52,280 --> 00:44:54,960 matches the vibe of the alien guy little

999 00:44:54,960 --> 00:44:57,960 like alien dance thing so he's blue he's

1000 00:44:57,960 --> 00:44:59,520 walking all weird and he's got really

1001 00:44:59,520 --> 00:45:02,680 long arms he's like like kind of worming

1002 00:45:02,680 --> 00:45:06,000 them back and forth while he's walking

1003 00:45:06,000 --> 00:45:09,079 so quite interesting I think it's really

1004 00:45:09,079 --> 00:45:12,680 cool I think it would be really awesome

1005 00:45:12,680 --> 00:45:16,200 for a kid to see this and draw something

1006 00:45:16,200 --> 00:45:19,359 you upload it and talk about like your

1007 00:45:19,359 --> 00:45:22,040 drawing G alive or something that's what

1008 00:45:22,040 --> 00:45:24,200 I would do if I was a dad I I would get

1009 00:45:24,200 --> 00:45:27,480 a lot of joy out of it okay and and so

1010 00:45:27,480 --> 00:45:30,280 let's do this one this is quite crazy so

1011 00:45:30,280 --> 00:45:32,079 we'll share this tab

1012 00:45:32,079 --> 00:45:34,520 instead okay so this is my original clip

1013 00:45:34,520 --> 00:45:35,839 I'm going to share this and hopefully

1014 00:45:35,839 --> 00:45:39,599 you could hear it hey brother my friend

1015 00:45:39,599 --> 00:45:42,960 it's been so long meet me down the

1016 00:45:42,960 --> 00:45:46,240 street and let's get a drink

1017 00:45:46,240 --> 00:45:49,079 together okay so then this is the

1018 00:45:49,079 --> 00:45:51,079 non-expressive translation which I think

1019 00:45:51,079 --> 00:45:54,079 it's

1020 00:45:58,440 --> 00:46:01,160 so I think that would just be hey this

1021 00:46:01,160 --> 00:46:03,599 is like the straightup translation and

1022 00:46:03,599 --> 00:46:06,559 then this is the translation of my voice

1023 00:46:06,559 --> 00:46:08,400 in Spanish so this is supposed to sound

1024 00:46:08,400 --> 00:46:11,800 like me if I was able to speak

1025 00:46:20,440 --> 00:46:23,000 Spanish it sounds like I I smoke or

1026 00:46:23,000 --> 00:46:25,359 something with my voice and maybe that's

1027 00:46:25,359 --> 00:46:28,920 what I sound like I don't know

1028 00:46:29,119 --> 00:46:31,280 but let's

1029 00:46:31,280 --> 00:46:36,599 see Olo so uh maybe it's close I'm not

1030 00:46:36,599 --> 00:46:39,000 sure well I'll have to play it back

1031 00:46:39,000 --> 00:46:42,920 but in the accents a lot better

1032 00:46:42,920 --> 00:46:46,160 in this translation

1033 00:46:46,160 --> 00:46:48,440 obviously but you know I'm working on

1034 00:46:48,440 --> 00:46:51,480 mine and so now the next thing so that

1035 00:46:51,480 --> 00:46:55,000 was that was the seamless translation

1036 00:46:55,000 --> 00:46:57,119 the previous one was the animated

1037 00:46:57,119 --> 00:46:59,480 drawings and then this next one is this

1038 00:46:59,480 --> 00:47:01,280 segment anything demo which I think is

1039 00:47:01,280 --> 00:47:03,200 the biggest one that they they've

1040 00:47:03,200 --> 00:47:05,440 released recently so I'm going to share

1041 00:47:05,440 --> 00:47:07,599 this tab instead

1042 00:47:07,599 --> 00:47:11,000 now so this is a a platform that allows

1043 00:47:11,000 --> 00:47:13,520 you to upload

1044 00:47:13,520 --> 00:47:16,520 videos

1045 00:47:17,520 --> 00:47:20,960 and make changes to them on the fly so

1046 00:47:20,960 --> 00:47:23,880 you can select which objects you want to

1047 00:47:23,880 --> 00:47:27,440 edit so I did the person and so this was

1048 00:47:27,440 --> 00:47:29,680 a race that I did last weekend with my

1049 00:47:29,680 --> 00:47:32,920 friend and she doesn't know I'm doing

1050 00:47:32,920 --> 00:47:35,359 this so I'm not going to show her face

1051 00:47:35,359 --> 00:47:37,400 but there's a whole bunch of different

1052 00:47:37,400 --> 00:47:38,720 options and I could play around with the

1053 00:47:38,720 --> 00:47:40,079 background so I could erase the

1054 00:47:40,079 --> 00:47:42,319 background I can make it black and white

1055 00:47:42,319 --> 00:47:43,640 I can make it white so you can't even

1056 00:47:43,640 --> 00:47:46,240 see her anymore we get a green

1057 00:47:46,240 --> 00:47:48,720 screen and I could put a gradient on it

1058 00:47:48,720 --> 00:47:50,880 I could change the gradient a couple

1059 00:47:50,880 --> 00:47:53,200 times and you could overall you could

1060 00:47:53,200 --> 00:47:56,040 just do a lot of things when you click

1061 00:47:56,040 --> 00:47:57,839 these objects and you and you make these

1062 00:47:57,839 --> 00:47:58,960 objects

1063 00:47:58,960 --> 00:48:02,040 in the original piece where you select

1064 00:48:02,040 --> 00:48:04,599 them you just click it's super easy you

1065 00:48:04,599 --> 00:48:06,559 just click on it and you can you can

1066 00:48:06,559 --> 00:48:09,359 pick out articles of clothing you can

1067 00:48:09,359 --> 00:48:12,160 pick out her shoes I could do her

1068 00:48:12,160 --> 00:48:15,200 hair uh I was able to do a shirt or a

1069 00:48:15,200 --> 00:48:17,280 shorts and then you click on the whole

1070 00:48:17,280 --> 00:48:19,319 body and it it will do it and and

1071 00:48:19,319 --> 00:48:20,640 normally if you're not familiar with

1072 00:48:20,640 --> 00:48:23,640 video editing this stuff takes a long

1073 00:48:23,640 --> 00:48:25,319 time to make

1074 00:48:25,319 --> 00:48:27,920 these segments

1075 00:48:27,920 --> 00:48:31,599 and so with

1076 00:48:32,000 --> 00:48:35,079 Adobe what is it what is it Adobe the

1077 00:48:35,079 --> 00:48:36,720 video editor I'm blanking on it on the

1078 00:48:36,720 --> 00:48:40,800 podcast Adobe Premiere Adobe Premiere

1079 00:48:40,800 --> 00:48:42,400 this stuff would take a long time

1080 00:48:42,400 --> 00:48:45,520 because you have to you have to get the

1081 00:48:45,520 --> 00:48:47,000 segment you have to segment it and then

1082 00:48:47,000 --> 00:48:48,400 you have to segment it and then go

1083 00:48:48,400 --> 00:48:50,359 through the key frames of the video

1084 00:48:50,359 --> 00:48:53,040 which it's doing and see where you're at

1085 00:48:53,040 --> 00:48:55,680 and see if this segment is still

1086 00:48:55,680 --> 00:48:57,280 matching up correctly

1087 00:48:57,280 --> 00:49:00,240 and the AI was able to do

1088 00:49:00,240 --> 00:49:03,960 that on its own on its own and it's not

1089 00:49:03,960 --> 00:49:07,240 like it's super clear on who this person

1090 00:49:07,240 --> 00:49:09,079 is there's a lot of stuff going on in

1091 00:49:09,079 --> 00:49:11,280 the background there's text there's

1092 00:49:11,280 --> 00:49:13,119 different colors there's different

1093 00:49:13,119 --> 00:49:16,720 people there is a lot of movement and so

1094 00:49:16,720 --> 00:49:19,760 it's matched up with this this person

1095 00:49:19,760 --> 00:49:22,119 perfectly and it's

1096 00:49:22,119 --> 00:49:24,720 honestly mindblowing because something

1097 00:49:24,720 --> 00:49:26,839 like this would take me a long time to

1098 00:49:26,839 --> 00:49:29,839 do manually like a long time like 40

1099 00:49:29,839 --> 00:49:33,000 minutes just to get the the segment put

1100 00:49:33,000 --> 00:49:35,400 together and maybe they've made some AI

1101 00:49:35,400 --> 00:49:37,280 improvements since the last time I've

1102 00:49:37,280 --> 00:49:39,640 used Adobe

1103 00:49:39,640 --> 00:49:43,960 Premiere but it

1104 00:49:43,960 --> 00:49:48,359 is something

1105 00:49:48,440 --> 00:49:51,440 that

1106 00:49:51,440 --> 00:49:55,400 you get wowed about um this is a wow

1107 00:49:55,400 --> 00:49:58,079 like wow this is is crazy and you can

1108 00:49:58,079 --> 00:50:00,000 add text over your

1109 00:50:00,000 --> 00:50:03,319 video you can you know desaturate it

1110 00:50:03,319 --> 00:50:05,240 make it in black and white or just kind

1111 00:50:05,240 --> 00:50:07,520 of just edit the the color Hues that are

1112 00:50:07,520 --> 00:50:10,440 allowed to bleed through you can blur it

1113 00:50:10,440 --> 00:50:13,799 out you can do the outline and and so

1114 00:50:13,799 --> 00:50:16,280 you can do more than this if you go to

1115 00:50:16,280 --> 00:50:20,160 Mark's recent post on Instagram you

1116 00:50:20,160 --> 00:50:22,640 could see

1117 00:50:22,640 --> 00:50:26,040 that you can see that he recently used

1118 00:50:26,040 --> 00:50:28,839 this to to make a video and he did a

1119 00:50:28,839 --> 00:50:30,520 whole bunch of different objects and

1120 00:50:30,520 --> 00:50:32,160 different effects in his video and it

1121 00:50:32,160 --> 00:50:36,160 looks sick looks sick and he's got the T

1122 00:50:36,160 --> 00:50:39,280 Pain chain and he's like thinking T pay

1123 00:50:39,280 --> 00:50:41,079 he puts the chain on the effects start

1124 00:50:41,079 --> 00:50:43,839 rumbling in the video and T pay starts

1125 00:50:43,839 --> 00:50:45,480 talk like not talking but the music

1126 00:50:45,480 --> 00:50:47,280 plays it's

1127 00:50:47,280 --> 00:50:51,119 cool and it's something where it allows

1128 00:50:51,119 --> 00:50:54,240 people to easily create a cool video

1129 00:50:54,240 --> 00:50:56,400 with not much effort cuz to do the same

1130 00:50:56,400 --> 00:50:58,680 thing it take you an hour or so just to

1131 00:50:58,680 --> 00:51:00,720 just do the segment and then to have

1132 00:51:00,720 --> 00:51:02,480 these different options it takes time

1133 00:51:02,480 --> 00:51:05,160 and to do all these things you have to

1134 00:51:05,160 --> 00:51:06,839 set up different layers and it so it

1135 00:51:06,839 --> 00:51:09,680 just makes it more accessible and allows

1136 00:51:09,680 --> 00:51:13,559 allows the user to do stuff that

1137 00:51:13,559 --> 00:51:16,079 normally would take a long time pretty

1138 00:51:16,079 --> 00:51:18,720 quick and it's got plenty of

1139 00:51:18,720 --> 00:51:20,960 options each one's got

1140 00:51:20,960 --> 00:51:23,799 three it's all good I would show you

1141 00:51:23,799 --> 00:51:25,920 more but I thought this video was pretty

1142 00:51:25,920 --> 00:51:27,079 good

1143 00:51:27,079 --> 00:51:29,920 and it had a lot it was challenging for

1144 00:51:29,920 --> 00:51:32,520 the AI to to find the segment so I

1145 00:51:32,520 --> 00:51:33,760 thought it would be good but I don't

1146 00:51:33,760 --> 00:51:35,520 want to share her face because she

1147 00:51:35,520 --> 00:51:38,839 doesn't know I'm doing this video and

1148 00:51:38,839 --> 00:51:43,799 the time that I was able to to ask I she

1149 00:51:43,799 --> 00:51:47,160 wasn't available so yeah but this is

1150 00:51:47,160 --> 00:51:49,920 what it looks like it's pretty cool okay

1151 00:51:49,920 --> 00:51:53,599 transitioning over now to my story that

1152 00:51:53,599 --> 00:51:54,319 I

1153 00:51:54,319 --> 00:51:57,400 built share this tab and dead so this is

1154 00:51:57,400 --> 00:51:59,000 really cool and this is another one I

1155 00:51:59,000 --> 00:52:02,359 was like wow this is insane I can't

1156 00:52:02,359 --> 00:52:03,599 believe

1157 00:52:03,599 --> 00:52:08,319 it wild so I asked this storytelling

1158 00:52:08,319 --> 00:52:11,640 machine to build me a story

1159 00:52:11,640 --> 00:52:15,520 about a non or not non-fiction a

1160 00:52:15,520 --> 00:52:20,000 fictional sci-fi story and so it came up

1161 00:52:20,000 --> 00:52:22,440 with this whole thing except my voice at

1162 00:52:22,440 --> 00:52:24,880 the end right here so this is me and it

1163 00:52:24,880 --> 00:52:26,599 only says I have a story and this that's

1164 00:52:26,599 --> 00:52:27,559 what I was talking about earlier I

1165 00:52:27,559 --> 00:52:30,319 wanted to say you know you're not

1166 00:52:30,319 --> 00:52:33,240 authorized blah blah blah abort abort

1167 00:52:33,240 --> 00:52:35,079 whatever and I said saying I was going

1168 00:52:35,079 --> 00:52:39,680 to report Jamie to

1169 00:52:39,680 --> 00:52:42,520 the space Federation or something so I'm

1170 00:52:42,520 --> 00:52:44,079 G to give this a play and everyone's

1171 00:52:44,079 --> 00:52:46,760 gonna be able to listen let me turn it

1172 00:52:46,760 --> 00:52:50,599 up turn it up baby all right I have a

1173 00:52:50,599 --> 00:52:54,040 wonderful story to tell you oh no the

1174 00:52:54,040 --> 00:52:58,440 engine isn't supposed to sound like that

1175 00:52:58,880 --> 00:53:01,079 have you tried tightening the quantum

1176 00:53:01,079 --> 00:53:06,119 flux modulator H let me try

1177 00:53:06,119 --> 00:53:09,880 that nope no dice have you properly

1178 00:53:09,880 --> 00:53:13,119 aligned the dimensional Rift generator

1179 00:53:13,119 --> 00:53:17,240 no I haven't I'll go try

1180 00:53:17,240 --> 00:53:21,440 that rats still broken what about

1181 00:53:21,440 --> 00:53:24,640 priming The Ether Vortex

1182 00:53:24,640 --> 00:53:28,839 compressor I bet that is the

1183 00:53:29,799 --> 00:53:34,839 problem all systems go thanks a

1184 00:53:34,920 --> 00:53:38,839 million we have LIF off have a wonderful

1185 00:53:38,839 --> 00:53:43,680 trip send me a postcard from

1186 00:53:46,680 --> 00:53:50,880 Mars okay so there was my story that I

1187 00:53:50,880 --> 00:53:54,839 had AI create for me using the auto box

1188 00:53:54,839 --> 00:53:57,480 audio box

1189 00:53:57,480 --> 00:53:58,839 really cool

1190 00:53:58,839 --> 00:54:01,920 stuff super easy to use you can easily

1191 00:54:01,920 --> 00:54:04,040 create your own voice and then you can

1192 00:54:04,040 --> 00:54:05,960 add stuff on the fly if I wanted to add

1193 00:54:05,960 --> 00:54:08,280 stuff I could add speech I could use my

1194 00:54:08,280 --> 00:54:10,119 voice I could type whatever you could do

1195 00:54:10,119 --> 00:54:12,160 examples happy you know t a great day is

1196 00:54:12,160 --> 00:54:15,599 what I use and I would say um welcome or

1197 00:54:15,599 --> 00:54:19,640 say uh thank say what if I said Venture

1198 00:54:19,640 --> 00:54:21,760 step

1199 00:54:21,760 --> 00:54:24,400 podast is the

1200 00:54:24,400 --> 00:54:28,359 best let's see generate

1201 00:54:31,079 --> 00:54:33,640 so it's generating right now if you are

1202 00:54:33,640 --> 00:54:36,440 listening and then we'll see how long it

1203 00:54:36,440 --> 00:54:37,960 takes it normally didn't take that long

1204 00:54:37,960 --> 00:54:39,760 less than 10 seconds so it should be

1205 00:54:39,760 --> 00:54:43,359 coming up here shortly here we go so it

1206 00:54:43,359 --> 00:54:46,480 gives you an option so here's this

1207 00:54:46,480 --> 00:54:49,799 one I don't know why I'm whispering

1208 00:54:49,799 --> 00:54:52,920 but is the

1209 00:54:52,920 --> 00:54:57,599 best all right step podcast is the best

1210 00:54:57,599 --> 00:54:59,920 I think this sounds more like

1211 00:54:59,920 --> 00:55:02,720 me but I don't know why I'm fading out

1212 00:55:02,720 --> 00:55:04,400 so much at the end and maybe that's how

1213 00:55:04,400 --> 00:55:06,200 I talk I don't

1214 00:55:06,200 --> 00:55:08,880 know but this this sounds like me so I

1215 00:55:08,880 --> 00:55:11,040 like this one and so if I wanted to add

1216 00:55:11,040 --> 00:55:13,240 that to my story I could I could do that

1217 00:55:13,240 --> 00:55:15,599 so you could see how this could be

1218 00:55:15,599 --> 00:55:17,799 really cool if

1219 00:55:17,799 --> 00:55:22,440 you had larger context window where

1220 00:55:22,440 --> 00:55:27,480 maybe not 125 but maybe 10,000 or

1221 00:55:27,480 --> 00:55:29,520 1,125 would allow you to make these

1222 00:55:29,520 --> 00:55:31,760 elaborate stories and sound effects and

1223 00:55:31,760 --> 00:55:33,000 all this

1224 00:55:33,000 --> 00:55:36,200 stuff and then you could animate with

1225 00:55:36,200 --> 00:55:37,880 your animation

1226 00:55:37,880 --> 00:55:40,079 tool and then you can make a you can

1227 00:55:40,079 --> 00:55:43,319 make a whole like a whole like cune just

1228 00:55:43,319 --> 00:55:45,599 with these like couple tools with using

1229 00:55:45,599 --> 00:55:48,200 Ai and so it unlocks the possibilities

1230 00:55:48,200 --> 00:55:49,359 that

1231 00:55:49,359 --> 00:55:52,400 normally you wouldn't be able to do in

1232 00:55:52,400 --> 00:55:53,920 in that short amount of time so it

1233 00:55:53,920 --> 00:55:56,680 allows a lot more people to create and

1234 00:55:56,680 --> 00:56:00,680 share their ideas and and

1235 00:56:00,680 --> 00:56:04,000 provide value to you know millions of

1236 00:56:04,000 --> 00:56:06,359 other people because there's 200 million

1237 00:56:06,359 --> 00:56:09,760 creators and so hopefully this thing

1238 00:56:09,760 --> 00:56:13,319 pans out to be

1239 00:56:13,319 --> 00:56:15,839 a exponent

1240 00:56:15,839 --> 00:56:19,200 for creativity growth improving the

1241 00:56:19,200 --> 00:56:21,079 economy and people's lives which I'm

1242 00:56:21,079 --> 00:56:24,559 sure it will and these tools are just to

1243 00:56:24,559 --> 00:56:26,319 get getting started you know they're

1244 00:56:26,319 --> 00:56:29,280 they're just demos and they're honestly

1245 00:56:29,280 --> 00:56:32,720 so cool as is like I think it's a sick

1246 00:56:32,720 --> 00:56:35,880 demo and this is much better than

1247 00:56:35,880 --> 00:56:38,000 Google's demos where Google doesn't

1248 00:56:38,000 --> 00:56:41,119 really allow you to upload and and do

1249 00:56:41,119 --> 00:56:42,359 your own

1250 00:56:42,359 --> 00:56:45,200 thing these tools allow you to upload

1251 00:56:45,200 --> 00:56:48,079 your own videos your own voice all that

1252 00:56:48,079 --> 00:56:50,839 stuff and it's just a demo it's very

1253 00:56:50,839 --> 00:56:53,200 impressive I'm very impressed with it

1254 00:56:53,200 --> 00:56:55,440 it's really cool next week we're going

1255 00:56:55,440 --> 00:56:56,799 to be talking about meta we're going to

1256 00:56:56,799 --> 00:56:58,680 go on a meta Marathon we're going to

1257 00:56:58,680 --> 00:57:01,480 become meta super fans and so next week

1258 00:57:01,480 --> 00:57:03,720 we're going to be discussing I I think

1259 00:57:03,720 --> 00:57:07,119 I'm most interested to discover the AI

1260 00:57:07,119 --> 00:57:09,079 characters and to

1261 00:57:09,079 --> 00:57:12,920 discuss making some AI agents with meta

1262 00:57:12,920 --> 00:57:15,839 and seeing what you can do and try to

1263 00:57:15,839 --> 00:57:18,680 push push to the limits and see see what

1264 00:57:18,680 --> 00:57:20,039 we could we could do and if I could

1265 00:57:20,039 --> 00:57:22,480 share with my friends and and what are

1266 00:57:22,480 --> 00:57:26,400 the things that I find most useful for

1267 00:57:26,400 --> 00:57:28,920 use cases and and where do I see this

1268 00:57:28,920 --> 00:57:31,359 going I know Mark has talked about

1269 00:57:31,359 --> 00:57:34,960 having an AI generated social media

1270 00:57:34,960 --> 00:57:38,160 where AI has their own social media to

1271 00:57:38,160 --> 00:57:39,920 learn from each other they'll have their

1272 00:57:39,920 --> 00:57:43,160 own like AI channels for the agents

1273 00:57:43,160 --> 00:57:45,720 they'll have the agents would interact

1274 00:57:45,720 --> 00:57:47,559 with you

1275 00:57:47,559 --> 00:57:50,160 every social media influencer would have

1276 00:57:50,160 --> 00:57:53,240 their own AI agent and allow them to

1277 00:57:53,240 --> 00:57:54,720 interact with their fans they'll use

1278 00:57:54,720 --> 00:57:56,359 their content and their comments and

1279 00:57:56,359 --> 00:57:58,880 their videos to train the AI agent to

1280 00:57:58,880 --> 00:58:02,119 interact and mimic you to allow your

1281 00:58:02,119 --> 00:58:05,799 users to have more time with you

1282 00:58:05,799 --> 00:58:10,359 but uh it's not you so I don't I don't

1283 00:58:10,359 --> 00:58:12,640 know where that that all comes about

1284 00:58:12,640 --> 00:58:14,799 like the whole thought process but it it

1285 00:58:14,799 --> 00:58:17,000 would allow you to better service your

1286 00:58:17,000 --> 00:58:20,480 your fans but it's not you so I don't

1287 00:58:20,480 --> 00:58:23,559 know if it means as much if you apply to

1288 00:58:23,559 --> 00:58:25,480 every single comment of your hundreds of

1289 00:58:25,480 --> 00:58:27,079 thousands of comments comments but it's

1290 00:58:27,079 --> 00:58:31,240 not physically you it's an AI algorithm

1291 00:58:31,240 --> 00:58:33,960 that mimics you is that the same if it

1292 00:58:33,960 --> 00:58:37,400 interacts with other people mimicking

1293 00:58:37,400 --> 00:58:40,079 you using things that you would say and

1294 00:58:40,079 --> 00:58:43,799 the love that you would provide is it as

1295 00:58:43,799 --> 00:58:47,599 genuine as if if it was just you I think

1296 00:58:47,599 --> 00:58:50,799 many people would say no And so there's

1297 00:58:50,799 --> 00:58:53,520 certain parts where human connection is

1298 00:58:53,520 --> 00:58:56,799 going to fade a bit but then you'll just

1299 00:58:56,799 --> 00:58:58,960 free up more time to to connect in

1300 00:58:58,960 --> 00:59:01,160 different ways right like you'll

1301 00:59:01,160 --> 00:59:03,079 probably have more free time than you

1302 00:59:03,079 --> 00:59:06,599 will in the future than you do now so

1303 00:59:06,599 --> 00:59:09,079 pros and cons pros and cons but that's

1304 00:59:09,079 --> 00:59:10,359 what we'll be discussing I want to talk

1305 00:59:10,359 --> 00:59:12,400 about the AI studio and then I have to

1306 00:59:12,400 --> 00:59:14,760 read the research paper which is massive

1307 00:59:14,760 --> 00:59:17,280 Behemoth with just lots of dense

1308 00:59:17,280 --> 00:59:19,119 information and then I like to talk

1309 00:59:19,119 --> 00:59:20,720 about the research paper too and

1310 00:59:20,720 --> 00:59:22,599 hopefully by then I can finish the book

1311 00:59:22,599 --> 00:59:24,400 behind me the one in

1312 00:59:24,400 --> 00:59:27,079 blue and will'll be on regular book

1313 00:59:27,079 --> 00:59:29,760 reviews again because I'm behind by uh

1314 00:59:29,760 --> 00:59:31,880 one I'm behind by like three books so I

1315 00:59:31,880 --> 00:59:34,359 have to get those done

1316 00:59:34,359 --> 00:59:37,039 and yeah and I I really appreciate you

1317 00:59:37,039 --> 00:59:39,240 listening in today and of course

1318 00:59:39,240 --> 00:59:41,200 wherever you are in this world good

1319 00:59:41,200 --> 00:59:43,359 morning good afternoon good evening

1320 00:59:43,359 --> 00:59:45,359 thanks for listening and please listen

1321 00:59:45,359 --> 00:59:50,280 again I'll talk to you next week goodbye