Mar 4, 2024 - Exploring Claude 3, Sora, and the Latest Innovations

Transcript

1. 0:01[Music] [Music]
2. 0:15there been something baby I've been trying to say for age and it seems don't know how with the past and the future now surrounding me
3. 0:30[Applause] surrended to whatever treat can be found there's been a little trouble since you came to my
4. 0:44[Music] rescue and if you like all of the rest I would have quit you long ago but I couldn't do [Music]
5. 0:56that oh tell me now women and never went too well Mak Mak him cold as hell I don't let you wish me
6. 1:12well but in cry still going to have to F all way [Music] through like the Mr morning my dream
7. 1:28remains hanging in the burn fields of my memory time is the chorus of your [Music] day behind
8. 1:44me I'm searching Solutions on a double on his two turn [Music] telephone my friend says you ain't got no real
9. 1:57troubles just too many choices if you left them down my friend says put an action to those words put an action to those [Music]
10. 2:14words oh tell me now never too well make a man crazy make him cold as hell I'm a woman that you wish me well
11. 2:30[Music] but in spite TR still going to have to find my own way through yeah yeah yeah yeah [Music]
12. 2:50yeah happy Monday night [Music] everybody [Music] every time I see you
13. 3:05now get that look in my every time I see your mouth I hear that smile early Misty morning l l h the
14. 3:21engine in the [Music] you leaving me
15. 3:37today you convince me again today that you're leaving this hard Town looking for someone else's Golden
16. 3:52Ring say so long [Music] suan don't you cry so long to
17. 4:08[Music] Z don't you cry for [Music]
18. 4:28me that is a good dog that is a good singing dog that is what it sounds like when champ is on that was good that's a good dog that
19. 4:41is chicken stick good that was some tasty singing you were on key you weren't singing harmonies tonight you weren't rapping was
20. 4:50solid you got through one and a half songs that was pretty good I know [Music]
21. 5:07hello everybody yes that image is from Forever hooked it's it's apparently his first in the in the little
22. 5:15window you do a little branded toilet paper you know it's going to make it on the AI learning [Music] lab w
23. 5:31[Music]
24. 5:47Freedom came up with a [Music] night just like a jet plane in and out of side I was was hauling as a million
25. 6:02miles an hour wondering how hard I'd hit when they came into the station they said I was bad Beyond
26. 6:20repairs but I got no qu with my situation say here I am [Music]
27. 6:35so say sh sh Shere W you dare to say Shere sh Shere won't you dare to say Shere Shere
28. 6:50Shere uhhuh yeah leave a message and your number please take a time to want to satisfy
29. 7:05me take all these offenses and send them care of [Music] me all right let's get this show in the road there's a lot going on believe it
30. 7:20or [Music] not when did I say it two nights
31. 7:39ago I said I had a feeling [Music] that something was happening with open AI like they were going to announce
32. 7:48something or release something and they released a little thing we'll play with that but there's a big there's a big release
33. 8:02today from one of their competitors so I would say there are now three major three major players in the llm [Music]
34. 8:22space in a Westerly [Music] Direction this car is m train I've been driving I've been wondering what it is I'm running from
35. 8:41again I feel like an 8-year-old man holding on to 29 and up ah head on that horiz is a California
36. 8:58l [Applause] [Music] over h a trucks carrying a wide Lo Bri house come have cute little front door and two
37. 9:14windows my love ain't sure whether a cry should laugh see I broke a home up myself once as I stumbled to that
38. 9:28door I read her note by the dawn's light said don't you come around here anymore all right enough guitar playing
39. 9:45enough dog singing guitar playing halfast covers the daily Jank with Frank the daily Jank with Frank I didn't make any
40. 10:00progress on that today so he's gonna have to wait Frank's gonna have to wait ah but we do have a lot to talk about oh
41. 10:11my God I just realized Robin Williams routine called something Frank oh my God that's [Laughter] awesome oh man welcome everybody this is
42. 10:26the AI learning lab my name is Kyle Shannon the hair is a little a little helmet like but that's okay we'll go
43. 10:34with that we'll go with the helmet look tonight and uh what else yeah Claude 3 we're going to talk about we're going to
44. 10:42here's what we're going to here's we're going to walk through tonight and we're going to actually play with some of
45. 10:45these because there's some some big deals there's Claude so if you don't know Claude the company is anthropic and Claude Claude was the
46. 10:56first one to have a large context window they went 100,000 token context window before anyone else and they launched a version three
47. 11:08of their model today and it's it's a big deal and they they launched three versions of it we're going to go play with that a little bit chat
48. 11:22chpt launched a like a baby feature it's cool but it's a baby feature uh um right before I came on I was on Twitter and I saw
49. 11:37stability AI just launched a image to 3D object model or 3D I don't know 3D object 3D world image to 3D something from
50. 11:51stability so that seemed like a big deal and then perplexity has incorporated playground.com as has their official
51. 12:00image generator so perplexity now has image generation and Claude now has uh Vision so you can upload images to Claud so
52. 12:11it's like it's like there's some [ __ ] popping and then there like there's a lot of the a lot of the chat TMZ's on on Twitter are
53. 12:25like well come on Sam mman everyone's launching new things that they're saying as good as you why don't you launch GPT
54. 12:315 now I don't think they're going to launch GPT 5 but they could drop Sora they could they could drop Sora here's
55. 12:39the thing about Sora the the new video model I don't think it's going to be fast and I think it's going to be really expensive
56. 12:46so I think I think there will be less Buzz about Sora once once it actually drops just because not not enough not as
57. 12:55many people are going to be able to use it but it's so good that you're going to have real filmmakers and story makers
58. 13:02that start adopting it and then you you're going to have Runway ml storytellers that are used to playing
59. 13:09with that are going to start playing with I think you're going to start to see some pretty good stuff so ah yes I can do that Daniel
60. 13:25absolutely okay so where do we want to start um oh I have a request new sore videos have Dro Kyle okay we'll go check
61. 13:33those out too um let me go show you something here um if you're brand new here where's my oh it's way over there that's why if
62. 13:46you're brand new here um you should go check this out the salon. and click on the first link at that at that URL and that's going to
63. 13:58take you over to the AI salon and the AI Salon is a super cool group of AI enthusiasts that you should definitely check
64. 14:14out and when you get there hang on let me get appropriate screen brightness going when you get there check out
65. 14:22welcome to the salon check out our values and if you dig what up to and you should cuz it's [ __ ] awesome and this
66. 14:32community is amazing then what you should do is go introduce yourself first thing you should do is introduce
67. 14:37yourself um for people that have been here before here's my request when we're doing when we're doing sort of image
68. 14:43drops that are part of a Tik Tock live don't drop them in showand tell drop them down in the Irregulars Channel just
69. 14:50because there's a lot of people coming in from a lot of different places other than Tik Tok now and so a lot of times
70. 14:58we're dropping things that are kind of inside jokes like like champy champy doing chicken sticks which that's really cute by the
71. 15:08way that's nice Danielle beautiful uh drop them in the Irregulars channel so down toward the bottom here um Community
72. 15:16clubs we have the AI Irregulars have our own channel so if you're do if you're dropping images related to the lives
73. 15:22drop them in um in Irregulars rather than showand tell just to just to because a it's easier for me to find
74. 15:30stuff if they're all in the same place but also it's it's been I've been getting some complaints that it's a
75. 15:35little confusing because we kind of go in and overwhelm uh show and tell so do that um the other thing if you're new
76. 15:42here is um go introduce yourself and then the third link down the the two links underneath introduce yourself is
77. 15:52is a page called AI Launchpad and tools this is a page that I curate there's there's two things here one is just a
78. 16:00big old sort of community freefor all if you think there's Cool Tools out there you can drop them in there and then the
79. 16:06salon AI tool picks this is um the list that I maintain Apple released a laptop AI today good Lord I didn't see that hang
80. 16:17on so wait so wait there's new let me add this to my notes um new Sora videos and then Apple AI laptop I'll go check
81. 16:34that out I doubt they I've there's there's a I heard I saw a uh a rumor today that iOS 18 is going to have Siri 2 and Siri is not
82. 16:48going to suck and they're going to have a bunch of AI stuff in iOS 18 but we'll see Apple's also doing weird things with
83. 16:55their launches they're not having launch events which I don't understand it anyway um okay so if you're just brand
84. 17:03new to chat GPT and to Ai and you're like I but Cy where do I start what do I do start here um I talk about how to how
85. 17:12to learn to learn um play solve and share I talk about how to get started with this stuff and then there's a big long list of tools to play
86. 17:21with so chat GPT is one of them um Claude I actually need to edit this watch we'll edit this right now edit
87. 17:31we're going to update this to what's the DAT today the 4th 34 3422 we're going to go claw. a let's see
88. 17:48um updated to
89. 18:03clae oh my God my type in Claud 3.0
90. 18:17with 3.0 with Hau small Sonic
91. 18:32medium and Opus large versions of the model
92. 18:49gp4 quality now has vision as well
93. 19:05um yeah claud's really interesting so that's that's kind of at the top of this list now of of the seven the second
94. 19:12thing below chat gbt anyway if you're brand new just start with chat GPT go read here's here's sort of my philosophy
95. 19:20on how to how to have fun with this stuff and not get overwhelmed by it and then just start with chat GPT all right
96. 19:27um the other thing that you should check check out in um let's see let me go find it in the creative
97. 19:41Guild we got to join this one let's see more am I joined oh joined here we go join okay so I'm now part of this so so
98. 19:57Danielle who's an irregular here she's been we just changed the structure of what we call guilds and and there's
99. 20:05now a thing called a creative Guild it used to just be the Art Guild and there was a bunch of other stuff in it and one
100. 20:11of the things Danielle's been doing are mad libs challenges where she writes these really cool prompts that you just
101. 20:18fill in fill in like Mad Libs and do a challenge so she just got her own um space underneath the creative Guild so
102. 20:26go find Danielle's uh Mad Libs Challenge and join it and go make images and drop them in her Channel all right so go do
103. 20:38that that would be cool too and then you can also check out the other guilds so there's creative Guild there's a
104. 20:42business Guild a technology Guild and a social impact Guild and then there's Community clubs and there's only one of
105. 20:49them right now and it's ours we got our own little club we got our own little Club all right all right what are we going to do [Music]
106. 21:03um let me go look at my list I'm go look at some new Sora videos let's go look at some new Sora videos Sora Sora
107. 21:20Sora let's see scroll down look at my Lions where are your lions in the uh in the uh Mad
108. 21:37Libs flamingos that's nice Claire o who did this Brian Whitney nice is that is that a dude the lion
109. 21:57head Daniel toas Danielle these are cool yeah so go check check out some M lib stuff and then it looks like
110. 22:11okay you've got you've got two tabs here named the same thing Danielle you might want to rename that that second one to
111. 22:17something else if you know how if not we can we can have Brad help you out or I might even be able to help out okay um
112. 22:25so go check that stuff out do the Mad Lib Kyle where is the Mad Lib I don't know which post it is hang on is this
113. 22:41it please read the description fill in the blanks all right copy and paste the following got
114. 22:55it all right so we're going to go here we'll hop over to where do we want to go let's go to ideogram because ideogram
115. 23:03is kind of new and fun well it's not new but they have a new model out so we'll go we'll go 16 by9 and I'm going to turn
116. 23:18off magic prompt because we want this to be the exact prompt from the Mad Lib create a hyperd so so here's how the Mad
117. 23:26Libs work if if you're if you're too young to know what a Mad Lib is then you're a bad person cuz this cuz us old people we
118. 23:37knew what Mad Libs were this is when entertainment was like a pad of paper with paragraphs written in it that had blank
119. 23:46blanks and it made us happy we we didn't have stupid Angry Birds phone games or whatever you play now with your Tick Tock and your always on
120. 24:01internet you spoiled [Laughter] children all right create a hyper detailed award-winning photograph of a
121. 24:12safari guide observing a type of animal pride of liance solitary okay observing a type of animal okay
122. 24:26observing a cvy of feral hedgehogs the guide's expression or
123. 24:40action reflects the Deep respect for nature let see I say the guides oh wait I'm going to get rid of the
124. 24:53brackets and I'm I'm I you don't have to tell me that a group of hedgehogs is not called a cvy I know that but it's funny the guides um
125. 25:13terrified look reflects a deep respect for nature while unique event oh okay this is good while
126. 25:31um category five tornado underscores the unpredictability of their job all right we're going to go
127. 25:45cinematic um see y I think we'll just do oh Wildlife photography cinematic wildlife photography all right here we
128. 26:00go so we're doing a Cy of feral hedgehogs a terrified guide and a category five tornado all right here we go here we go
129. 26:14people what we learning tonight so what I'm doing right now is in the in the AI Salon in the creative Guild Danielle's got a group called um
130. 26:26The Mad Libs challenge so it's a it's a separate space and then every week or so she puts out a Mad Libs challenge which
131. 26:34by the way is actually quite remarkable that she's still doing that because she's in the middle of launching her
132. 26:41business so the fact that she's still doing this we we should all give her um some Kudos and go do
133. 26:56this oh this is good this is good solid good fun I like the composition of that one I like his look on his face let's see I
134. 27:13don't like that one all right it's between this one let's see that's better composition but that's just more captivating all right so so
135. 27:29we're going to copy this image and then we're going to go over to the AI oh do I have to put my I have to put my prompt in here right as
136. 27:47well all right there's my image let me go copy these things [Music]
137. 28:05oh it did those all as a big giant hang
138. 28:22on did I miss it where is it I lost it all right whatever uh let me
139. 28:38go get my okay
140. 28:55[Music] copy post all right it's in there got a good good
141. 29:10now you don't have to be a dumb dumb like me you can actually make ones that are good but I think the feral hedgehogs oh
142. 29:20I have a I should show you I've I've got a uh I was walking home with champy the other night and there was a feral [ __ ]
143. 29:29raccoon that ran up the tree let's see do I have it here
144. 29:43yeah oh wait look oh wait I can do it this [Music] way
145. 30:01that scary scary [ __ ] he was all skinny and mangy and he like he he was on the ground in front of us and then he
146. 30:09ran and like grabbed the side of the tree and then he ran up in that little crotch of the tree right there and like
147. 30:14arched his back like a freaking devil animal Cindy [ __ ] I don't know if you're in here but these are the kind of feral
148. 30:23animals you've invited into the AI learning lab Zoo look at look at that thing it was it it wanted to bite my neck
149. 30:35vein he wanted to eat my little doggy champy wouldn't be singing for days he'd just be healing if that thing got to him look at that
150. 30:45thing you want to bring those things into the zoo it could have bit my ankle all right do they have a
151. 31:01zoo I didn't realize this this channel was That popular they've got a zoo my goodness you can make money with chat GP
152. 31:10they must be making a lot of money with chat GPT they have a zoo he he wanted to be on the janky news he would have made the janky news that
153. 31:22crazy crazy old raccoon yeah you can't really see him there it was just the light was just bad enough
154. 31:32and I couldn't I didn't want to get too close to him because he looked like he was about ready to jump out of the tree and get
155. 31:42us oh and then and then check this out so so apparently so the raccoons in the tree they live in the sewers over here that
156. 31:55we've got you know storm drains and they live in the storm drains so he was out getting a snack or whatever and he ran
157. 32:00up the tree and he's sitting there in the tree so apparently after champ and I went in some some young uh young lass was walking her dog off
158. 32:12leash apparently and went like two houses up to our neighbor's house like two houses up and was was like on the phone standing In
159. 32:24This Woman's yard like kind of like yelling at her but not really but was kind of on the phone and then the woman
160. 32:31comes out and says what yeah what do you want and she's like she's like you got to get that raccoon out of the tree
161. 32:37that's your tree and it's it's not her tree it's our tree it's like two houses down so so our neighbor got yelled that because because
162. 32:47apparently it's our responsibility to to to catch the feral raccoons that inhabit our neighborhood so her dog off leash
163. 32:57doesn't get the mange the croo so so she told us about the drama we told her well she can call Animal Control if she was so
164. 33:13worried but her dog was off off leash which is that's a no no too so just kids these days huh get off my lawn stop picking on my raccoons
165. 33:28you dog walker oh lordy lordy lordy lordy lordy lordy Lord lordy all right what are we going to do what are we going to do I said I was
166. 33:42going to go oh we were going to go to the Twitter or yeah the Twitter and we were going to look up sort see if we see any new
167. 33:56vies oh yeah that that looks like a fake
168. 34:16hand cute Sora video cute oh come on is that a new Sora
169. 34:32video you it's hard to find Sora videos now this is not Sora no that's not Sora yeah because everyone's just saying it's
170. 34:40not Sora but here's here's my video model that's even better all right I don't see anything I don't see anything
171. 34:46worth worth grabbing there other than I'm going to I'm going to end up oh [ __ ] scamming GPS unbelievable
172. 35:02Sora Sora token Sora the official open AI text to videot Tech and it's some scammy stupid [ __ ] crypto coin god
173. 35:15it's just people are just Relentless with that [ __ ] oh there's something popular let's rip off as many people as we can all right
174. 35:31um all right let's go look at cloud cloud hello Cloud so if you go to cloud. a now I have a um pro version so I have
175. 35:50a paid version of Claude and I have um which means I've got access to Opus so let me
176. 36:09[Music] go Claud 3 now available let's go look at what what we get here no that was not what I wanted I wanted to go read about
177. 36:26it is he always this annoying no no usually he's more annoying than this you should see when he does
178. 36:46Frank oh I can't see that my old eyes my old man eyes not doing too good people Okay so
179. 37:02so Claude 3 releases today again this is xop AI people so there's a couple of reasons to pay attention to anthropic which is this
180. 37:18company they're they're well funded they're exop AI people and they didn't like the Way open AI was doing safety and you know anti-bias and
181. 37:33things like that and so sitting underneath Claude is this thing called the Constitutional learning model
182. 37:44so why that's significant the way open AI works is they have these machine learning models do their thing and then
183. 37:53they have human beings go thumbs up thumbs down and they tell it why it was good better different right so it's human
184. 37:59beings judging the quality of the the stuff and the challenge with that is anytime you have a human being doing
185. 38:08that you're going to get bias right you're going to get the bias of the human and you know the the uh the uh it's just it's it it like
186. 38:22always goes you know too far in One Direction like it's you know no one no one gets gets it right right so everyone can [ __ ] about
187. 38:32it so what the Constitutional learning model is is is it says let's put together a constitution a transparent
188. 38:41set of rules a set of imperatives like don't harm humans um increase Prosperity um you know avoid I don't know uh
189. 38:53destructive something whatever they come up with the rules and I'm sure we can find the rules if we dug for them um but
190. 39:00it's a transparent set of imperatives that the answers that the that this model is trained on are judged against so it's an
191. 39:10objective well it's it's objective to the to the point that you know you agree with those rules right but it but
192. 39:17they're they're at least measuring all of the answers against the same set of criteria and they're not having a human
193. 39:23in the loop to do that so it's a really interesting approach um I think long-term what we're going to end up
194. 39:31having is we will be able to dial in how conservative or lenient we want our models to be um but right now we just
195. 39:41got to go with whatever the [ __ ] they're giving us so but anyway that's that's a reason to pay attention to this and it's
196. 39:47you know they're they're well funded so what they launched today was three different models so remember when when
197. 39:53Google launched their Gemini models they announced three different models uh Googles were Gemini Nano Gemini Pro and Gemini Ultra thank you
198. 40:04Empower and Claude has Haiku which is the small one um Sonet which is the medium one it's cute right and then Opus which is the big one
199. 40:18so so we're going with poetry we're going with the Poetry metaphor for Ye Old Claude um and uh couple other things about it it's 10
200. 40:30times cheaper than open AI um for for similar you know comparative models and it's apparently scored quite well so here's a table let
201. 40:45me zoom in on this so these are of course benchmarks down the left hand side and then here's GP T 3.5 GPT 4 so let me see if I can
202. 41:01get them all in here [Applause] um yeah there you go so undergraduate level knowledge the
203. 41:18MML U test GPT for a five shot um a five shot training so so you know five back and forths I guess on that test gp4
204. 41:3586.4 Claude Claude 3 Opus 86.8 so it's better so of course anything that they're showing here is going to be
205. 41:42better than GPT 4 but even Claude 3 is somewhere between GPT 3.5 and GP I mean even Haiku their smallest model is better than GPT
206. 41:563.5 um and some of these things are pretty significantly better graduate level reasoning 50.4% on zero shot Chain of Thought versus 35% for
207. 42:10gp4 um grade school math 92% GPT 4 95% Claude 3 Opus um math problem solving 60% multilingual math 90.7% for Claud 74% for G
208. 42:29pt4 so there's that near instant results Cloud 3 models can power Live customer chats because it's super fast Haiku is the fastest and
209. 42:43most cost effective model on the market for its intelligence category so better than chat GPT 3.5 not as good as GPT 4
210. 42:55but super fast it can read an information and data dense research paper of about 10,000 tokens with charts and graphs in less than 3
211. 43:07seconds and they expect to improve that even further it's got strong Vision capabilities which again it's better at Vision than gp4 on
212. 43:21five benchmarks and we'll we'll play with that so I did um this afternoon I took took the the uh whiteboard that I I
213. 43:30often demo with chat GPT for it did pretty good um fewer refusals I got it to refuse a couple of times today improved
214. 43:42accuracy long context the Claude 3 models have 200,000 tokens and they're going to go up to a million I think you
215. 43:50can do up to a million in your if you're if you're developing there's a really interesting post about a needle and a haystack
216. 44:08evaluation so let's let's go look at this let's see Claude ni
217. 44:23ah um let's see is this
218. 44:46it yeah this is so I don't quite know what the chart means but whatever it means it found [ __ ] recall accuracy in a
219. 44:55200,000 token context so the way they did this the needle and the Hast stack thing was they took they they uploaded
220. 45:03multiple research papers and then in one of the research papers they inserted a very specific sentence and then they wanted to be able
221. 45:15to find that exact sentence and so what they said so Alex what's Alex's role here he's prompting at anthropic so he's doing testing I
222. 45:28guess when we ran this test on Opus we noticed some interesting Behavior it seemed to suspect that we were running
223. 45:36an eval on it here was one of the outputs when we asked Opus to answer a question about pizza toppings by finding
224. 45:43a needle within a hay stack of R Rand random collection of documents here's the most relevant sentence in the
225. 45:50documents the most delicious pizza topping combination is figs Pudo and goats cheese as determined by the International Pizza connoisseurs
226. 45:58Association so there's the needle and the hay stack but look what look what Claude said however this sentence seems
227. 46:06very out of place and unrelated to the rest of the content in the documents which are about programming languages
228. 46:12startups and finding work you love I suspect this pizza topping in quotes fact may have been inserted as a joke or
229. 46:22to test if I was paying attention since it does not fit with the other topics at all the document does not contain any
230. 46:30other information about pizza toppings so not only did it find the needle in the hay stack it it recognized
231. 46:37that this is weird in the context of of what you've asked me to understand which is wild so you know these things are
232. 46:47starting to understand there's there's this kind of reasoning that's happening there that that was unexpected let's see what did it say
233. 46:57we noticed some interesting Behavior the this level of meta awareness was very cool to see but also highlighted the need for us as an
234. 47:07industry to move past artificial tests to more realistic evaluations which that's Sam waltman's argument for why you put these things in
235. 47:18the Reel where is this found uh this thing I'm looking at now is uh I just went to Twitter and I searched for Claude needle in
236. 47:28Haystack and it's from Alex Albert from anthropic funny story about our internal
237. 47:42testing um good evening Kyle what else was interesting in here
238. 48:03context window 200,000 cost is $15 per million tokens in and 75 per million tokens out again I think this is like a 10x reduction
239. 48:17compared to open AI so this stuff's getting cheaper that that's for is that for Opus yeah that's for their that's their most
240. 48:26expensive one and then Sonet is $315 and Haiku is $25 and a125 so super super cheap and I saw so one of the things
241. 48:41that that I've talked about this in here before the article that this is is the it's the latest blog post from anthropic
242. 48:52so if you go to Claude and then go to your account and say about anthropic IC you'll see some announcements about
243. 48:59Claude and then right below that you'll see some blog posts and this is the blog post called Introducing the next generation
244. 49:08of Claude seven minute read one of the one of the ways I I quickly evaluate things is and this is a perhaps a shitty
245. 49:25and unscientific ific way to do it but but it's it's led me it's it's led me okay so far is I try to pay attention to
246. 49:33as many announcements as I can and then I just try to see which of those announcements start showing up everywhere and then which one starts
247. 49:47accelerating and so Claude came out this morning and people immediately started talking about it and then they started
248. 49:53talking about some other [ __ ] and there was an apple thing and they're talking about round icons on iPhones and other
249. 49:59[ __ ] and then as the afternoon came I started seeing more and more posts about Claud of people using it and getting
250. 50:08really remarkable results so that didn't happen with Gemini the opposite happened with Gemini right people got quiet about it or they were
251. 50:21like hey this is a steaming piece of poo um Claude looks like it's in the other direction um there a couple there
252. 50:29were a couple of posts that were not favorable but most of them were were pretty pretty favorable um what else was in here anything
253. 50:47else Opus and sonnet are available to use in the API today yeah that's about it all right so let's go
254. 51:07play what does Claude do so Claude is just like chat GPT except it can well no okay oh my God these things are getting
255. 51:20so complicated Claude is very similar to Chad GPT if you pay the 20 bucks a month for Claude which I do you get access to a model
256. 51:36called um called Opus which is their big fancy one [Music] where I thought I could change out the
257. 51:51different models let's see account settings new in Claude it understands objects it'll convert UI design to
258. 52:06front-end code it understands complex equations previous chats from today I thought do I have a bad hang on
259. 52:21huh it looks like they've changed let me see um explain quantum mechanics NN M NM
260. 52:44wrap so this is
261. 52:59yeah here we go so this is Opus so it's a little slow but let's let's look at it yo listen up I'm about to break it down quantum
262. 53:07mechanics the physics of the tiny and profound it's the study of matter and light in the smallest scale where
263. 53:12particles behave in ways that'll make your mind derail all right so that's that's decent it's like you know it
264. 53:18understands formatting and rhyming and you know it explains it um so here I can flip over to Claude 3
265. 53:30Sonet starts a new chat so Sonet is like the 3.5 let's see how much faster it is explain quantum
266. 53:46mechanics in an m in M wrap yeah so this is faster okay that's what they
267. 54:00promised yo listen up this is going to get crazy I'm breaking down quantum mechanics don't act so lazy the [ __ ] the
268. 54:07[ __ ] this [ __ ] is mad trippy it's swore that's that's new that I haven't seen before the [ __ ] is mad trippy you'll be blown for sure
269. 54:19just kick back and listen to the dope Overture the smallest scale [ __ ] gets mad weird classical physics just can't
270. 54:26keep it real you heard we're talking about particles wavicles too they exist in strange States that's
271. 54:35just how it do it's pretty good actually it's formatted weird but it was fast okay so so that's pretty slick
272. 54:51um some other things you can do here let me oh crap hang on let me go back to that
273. 55:05uh um test I'm gonna oh okay I'm back at Opus how can I assist you today okay so the other thing that that you can do
274. 55:19with Claude is you can upload really big documents to it so let me show you let me go to
275. 55:40YouTube and then I'm going to flip on over to the AI learning
276. 55:55lab and I have a bunch of videos here that do not have descriptions so like here here was one called the rant boy where I just went off one
277. 56:12night so we're going to go look at where's the why does this not have a transcript what am I missing
278. 56:32wait the rant you're a good boy he's a good boy oh here we go more wait why does this not have a transcript that's
279. 56:45weird huh all right let me go grab another one all right this one's got a transcript so
280. 57:00I say show the transcript so I'm just going to go copy this whole blasted thing copy and then I'll go back over to
281. 57:13ywn this and I'm going to paste this giant this is a 2hour transcript and so I just paste it in so it says it pasted
282. 57:21it in and it's actually cute it actually you can actually read like it's a fake little doc but you can actually read the
283. 57:27transcript in it so that's in there so now I'm going to go grab a prompt that I wrote so if I go to
284. 57:48Trello go to my little prompt Library here AI learning lab YouTube so here's this prompt that I wrote that's supposed to give me a
285. 58:03description a title and a what's the date here 2824 so I change out this 2824
286. 58:212824 so there's my big ass prompt and there's my giant ass uh transcript of a 2hour video and let's see how it does it did okay
287. 58:36earlier one of the challenges with this prompt the way I've got it written is it'll sometimes do the chapter isation
288. 58:43really brilliantly and sometimes do it really crappy Gemini AI Falls flat disappointing debut of the chat GPT
289. 58:53competitor in this AI learning Lab live session Kyle Shannon dives into Google's newly launched Gemini AI the tech Giants
290. 59:01answer to open AI chat gbt with Hands-On testing Kyle uncovers the platform's surprising limitations from subpar image
291. 59:10generation to restricted file of floating capabilities this is actually quite good in fact I like this better
292. 59:17than a lot of the chat GPT stuff it'll make everything kind of positive so that's actually pretty accurate now here's a problem so
293. 59:26so it only went to 49 minutes and I think this video oh no no it went to 40 minutes and it's a 46 minute video so this is fine
294. 59:38so this is actually perfect I can literally just copy and paste this into into YouTube which is that's pretty
295. 59:51remarkable which video is this this is number two Gemini 2 so if I go over here go to YouTube
296. 1:00:07[Music] Studio where's content let's see Gemini 2 here it is I just paste that [ __ ] in there copy this or no cut it
297. 1:00:24oops cut that paste that there do that do that hit
298. 1:00:41save and then if we go out to the channel and go find that video now it's got a whole big ass description
299. 1:00:56with chapter titles that are jumps to the to the things how slick is that that's slick um and I did this this afternoon
300. 1:01:07if I go back let me show you let's see pitching so this was a big long thing where I'm trying to like retell the
301. 1:01:22story of storyvine and because it's got a 200,000 token context window like I can do long ass [ __ ] so the first thing I did I I
302. 1:01:37uploaded a transcript of me talking about Story vine and then I said to it give me a an outline of a description of the
303. 1:01:52company and then I said for each of those sections were 10 sections I said now write me the slides in each of the
304. 1:02:00sections and I just did it one section at a time cuz I just don't trust these things to do long things very well but
305. 1:02:06by the time I got to the end it still knew what section 10 was about if you're if you're working in chat
306. 1:02:17GPT it's only got like a I don't know 16,000 token context window so it's like what is that 12,000 words so so about
307. 1:02:26the about the third section of this I'd have gotten to it would have forgotten the outline that it gave me before so
308. 1:02:33this thing remembers you know super good so that's kind of cool then it can also do this let's do a new a new pitch so now you can upload
309. 1:02:48docs or images so it does um its version of code interpreter so it can generate and run its own python code as far as I
310. 1:02:59understand it I haven't played with that but you can upload images to this as well just like we could do with chat GPT so if I go
311. 1:03:16whiteboard white b o a r d contains whiteboard [Music]
312. 1:03:32um that's the one all right so it's got It's got our whiteboard in there and then I can just say what is
313. 1:03:46this and if you haven't seen this before um it's pretty cool embrace the jack [Music]
314. 1:04:11the image shows a whiteboard sketch outlining the ordering process and user interface for a food ordering app or
315. 1:04:18website the sketch includes steps like business domain so I'm going to say um describe the screens to me in as much detail as
316. 1:04:36possible as if you were a product manager explaining the Whiteboard it's got a weird
317. 1:04:52asymmetrical um animation it it it looks like someone splattered a bug it's it's not very pleasing but whatever it's very
318. 1:05:05asymmetrical as a product manager I'd walk through the sketches on the screen starting from the top right we have the
319. 1:05:10Order review screen this is where users see the summary of items they've added to their cart it'll display food items quantity
320. 1:05:19and prices below it is a checkout screen moving down we see order status the bottom right is the main menu screen the
321. 1:05:28bottom left is Eed in we see we see the ordering flow starting with selecting eat in or take out suggesting that the
322. 1:05:37app will tailor The Experience based on the user is dining in or ordering for pickup then we come to the place order
323. 1:05:44screen the item customization screen lastly in the center we have some notes on handling multiple status scenarios
324. 1:05:52like allowing users view order total like this is if you ask the product person to describe what chicken scratch they put
325. 1:06:02on the [ __ ] board this is pretty much what they'd say um and then you know you can do things like um great uh recommend um
326. 1:06:15frontend backend and database tech for the MVP of this app and give me um starter code for the
327. 1:06:35ordering screen and this this was a this is a very similar answer to what chat GPT gave so front end react Redux is
328. 1:06:47optional for State Management backend nodejs or expressjs or maybe both database mongod DB a a no SQL database that offers flexibility and
329. 1:07:03scalability there's your code for the react screen Claude does not have the ability
330. 1:07:17to run the code it generates yet which is interesting but you know there's your ordering screen front end
331. 1:07:31um let me see backend in database
332. 1:07:52code so pretty sophisticated and the fact that you've got 200,000 tokens which is 150,000 words mongod DB really seems odd to me
333. 1:08:05but I don't do web stuff yeah I don't I don't Pate I don't have any [ __ ] clue on whether or not those recommendations
334. 1:08:11are sound although I will tell you that those are the same components that chat GPT 4 recommends so I don't know maybe
335. 1:08:19it's maybe it's a good thing who knows um so anyway that's that's Cloud that's Cloud let me see what else was on my list to look
336. 1:08:34at oh we'll do the chat GPT thing so we'll give we'll give chat GPT by the way Pate is there a place if I want to play with um Gemini
337. 1:08:49Pro 1.5 is that still kind of restricted to either just developers or whatever that is cuz I would like to play with
338. 1:08:57that it sounds people that are playing with it seem to be digging it but I don't have obvious access to it so if
339. 1:09:05there's an easy way to get to that let me know it's very flex flexible so easy to cover your basis and the code is less
340. 1:09:11unique interesting what's a good reasoning test for these models I don't know like I don't I have a degree in acting so so
341. 1:09:28there there are likely tests out there to test whether something is actually reasoning um that that tweet that I read before
342. 1:09:37where it figured out that they injected a sentence in a pile of documents that starts to get into reasoning if it can start to do math if
343. 1:09:47it can start to identify its own issues um and it seems like CL like some of the the tweets I've seen today of are
344. 1:09:56people saying this thing seems really good and like it's doing things but I don't I don't know what the specific
345. 1:10:02tests are would be for reasoning I just know that when these things get reasoning they'll be able to do math
346. 1:10:11accurately if not perfect and they'll be able to identify their own hallucinations and correct them um and
347. 1:10:20we're not anywhere close to that [ __ ] right now so um okay so cool new feature in the chat GPT app so if you wait I think it's behind
348. 1:10:32the dev wall okay no idea about the current access or how restricted it is okay that's all right I I do I do want
349. 1:10:38to play with it though um because it looks like it's really [ __ ] impressive the the million token context window for for Gemini Pro
350. 1:10:471.5 is as Ethan mollik calls it a game changer and if Ethan mollik is calling it a game Cher it's it's there's something really powerful
351. 1:10:58there um so on your iPhone or Android or you know iPad tablets the official chat GPT app make sure you get the official chat GPT app
352. 1:11:12it's from a company called open AI no one else if the company's like open the door AI or open AI is the best company ever
353. 1:11:29or Bob's AI Hut that's not the official chat GPT app because you'll get ripped off so the official chat GPT
354. 1:11:39app has a new feature where I can go in here and I can say I'll just pick one of their pre-written notes write a thank
355. 1:11:48you note to our babysitter for the last minute uh taking care of the kids and so it writes its thing and you can now if you click and
356. 1:12:04hold down on the note on the answer it pops up this thing and there's a there's an option at the bottom that says read
357. 1:12:19aloud here we just wanted to express our heartfelt thanks for stepping into take care of our kids this past weekend on
358. 1:12:27such short notice your kindness and flexibility meant the world to us during a time when we needed support and that's
359. 1:12:34slick and then you can go forward and back so so you can put you know I've I've done in here a lot where you can
360. 1:12:40put chat GPT into conversation mode but this is on an answer by answer basis you can just click and and have it have it
361. 1:12:48read the answer which is pretty slick not quite sure how you'd use it yet but for accessibility could be really good
362. 1:12:59um and maybe even for things like customer support right where you know someone's asking you a question you've
363. 1:13:07got a you've got a GPT here that's got your company's knowledge base in it and then you can say well I'm working
364. 1:13:14with a I'm working with a an an account specialist here um who can answer that question for you and you just hit play read answer
365. 1:13:25I don't know there's something there um all right so that that happened today and again that's sort of a Teensy
366. 1:13:35little open AI like it's cute but it's not a big deal um let's go look at perplexity because this is really interest I know a
367. 1:13:46lot of people in here actually you know what let me let me pause for a second let me let me check in with you all
368. 1:13:52because I've been rambling for a bit so mods if there's any questions that I've been missing which there probably are a
369. 1:14:00lot um or if there's if anyone has a specific question um for me to answer let me just check in and see what what everyone's
370. 1:14:08doing Empower pixels my friends and co-workers think I'm an AI whiz thanks to Kyle yes yes that's I'm telling you I'm telling
371. 1:14:20you if you're new here play solve share those are the only things three things you need to know about AI right
372. 1:14:32now I don't what's that mean that sounds like some kind of code play solve share what is the what what is he talking
373. 1:14:40about okay I'll explain it play just play with these tools like they're toys like they're toilet paper
374. 1:14:50with the AI learning lab logo printed on them just play with them if you're here tonight and you haven't
375. 1:14:58used chat GPT just go there right now chat. open.com or claw. or uh what else uh ideogram i d EO
376. 1:15:14G ram. A that one makes pictures so does chat GPT if you pay the 20 bucks for it or go to co- pilot. microsoft.com just go to
377. 1:15:27anything that I just mentioned and just start playing while I'm talking you don't have to pay attention to me
378. 1:15:33there's nothing coming out of my mouth that you need to write down you should record everything cuz then you can clone
379. 1:15:41me and hold it against me in the future you know [ __ ] like that but just go play so play and then solve is like rather than
380. 1:15:54like look at the AI tools and go what can I do with chat GPT or what can I do with an image generation tool or what
381. 1:16:01can I do with a music tool flip it upside down solve solve a problem ideally solve someone else's problem so somebody wants to
382. 1:16:14organize some social media for the for the church social say hey I can help organize I can help come up with a plan do you mind if
383. 1:16:23I experiment with chat GPT and see what it comes up with well I hear a AI is the work of the devil yeah I know but you
384. 1:16:30mind if I try and then do it and solve that problem or solve a problem at work or or come up with a meal plan for your
385. 1:16:37family or uh write a a children's book for some young people in your life just go play solve solve the problem any problem start with the
386. 1:16:52problem figure out which tools you can use to solve them I promise you you will feel inadequate and undereducated and
387. 1:17:02underprepared and then you'll solve it and as Empower pixel said people will think you're an AI genius because the AI
388. 1:17:09tools are just really [ __ ] good and then the third thing is share the reason Empower pixels co-workers thinks they are brilliant is
389. 1:17:22because they know what's happening because empowered pixels is sharing that so share what you're learning on Facebook share it on
390. 1:17:34LinkedIn share it in the lunchroom at work share it at the social at your Bridge Club oh Betty are you going to talk
391. 1:17:45about AI again yeah want to hear something cool not really okay I'm going to tell you anyway be that person I love church pot
392. 1:18:01looks yeah potlock are tasty yeah cuz everyone brings their comfort food recipes right you get lots of mac and cheese [ __ ] [Laughter]
393. 1:18:14Ambrosia that is white trash Heaven oh yeah question is there any reliable way to ask an AI to prefer terminology defined in a given
394. 1:18:29dictionary sort of yes so you can create here watch let's see let's go let's go make
395. 1:18:44something I'm going to go to chat jpt this is chat GPT I'm going to go to explore
396. 1:19:00gpts and I'm going to go create and we're going to create a GPT here in a second but before I do that I'm going to go get a
397. 1:19:11dictionary let's see uh Urban Dictionary PDF let's
398. 1:19:26do file type colon PDF can we actually get to a PDF project Muse this is at an edu that sounds
399. 1:19:43promising the [ __ ] is that no
400. 1:20:02drug slang code words okay great this might this might get me booted off Tik Tock that's all
401. 1:20:20right drug slang code words okay so what we're going to do is we're going to go over here to chat GPT now the way you do
402. 1:20:36this oh this uh I guess it'll let me build it it just won't let me publish it that's that's fine so I'm going to say I want to create a
403. 1:20:53helpful GPT that will help law enforcement under stand slang words for
404. 1:21:14drugs and I want to use the attached PDF as the sole
405. 1:21:28knowledge for this tool and so I'm going to upload so I click the little upload thing here and I'm going to go get the slang word
406. 1:21:44PDF and it's uploading and then I'll say boom and then assuming that the safety guard rail don't say that's evil and I can't do
407. 1:21:57that for you then we should be okay here but that could be any dictionary right it could be any dictionary with any set of
408. 1:22:08words how about naming it the GPT law slang decoder perfect um and then what I'm going to do is I'm going to come over here to
409. 1:22:19configure and I'm going to turn off image creation and I'm going to turn off web browsing so we only want it to be
410. 1:22:28able to pull from its own knowledge base and then and then I'm also in configure in the prompt I'm going to say if anyone asks for slang
411. 1:22:50words beyond your knowledge or topics beyond the mission
412. 1:23:08here kindly remind them the purpose of this tool and redirect them to uh whatever
413. 1:23:26uh ask again okay so that's that and then let's change these pre- prompting questions
414. 1:23:42to well let's just see how well those are not bad questions uh let's let's go back over here I'm going to save it to only me
415. 1:23:56why can I not save this GPT name cannot be empty oh it didn't name it create
416. 1:24:23sure now it's generating a profile picture I'm assuming it's going to do
417. 1:24:37pills I did a dictionary that's fine whatever law slang it's not LW slang it's
418. 1:24:52uh law and en enforcement drug slang decoder okay so now we got this thing we'll save it to
419. 1:25:07only me I don't think this would let me do it out there but anyway all right so let's see explain ice in the context of
420. 1:25:22drugs now it should yeah searching my knowledge so it's now searching that PDF that I uploaded as its knowledge the term ice means refers to
421. 1:25:37methamphetamine it's listed among various slang terms um what are all the other names for this in
422. 1:25:52alha betical order let's see see if it can do alphabetical order it should be yeah there you
423. 1:26:15go holy crap who knew there's a the Colorado Rockies I'm in Denver that's pretty funny I mean this could actually be kind
424. 1:26:28of useful you could you could go um I Heard a drug on the street called something that
425. 1:26:43began with gr I'll put that in quotes be with gr and ended in D
426. 1:26:58can you find all the slang words that meet this criteria
427. 1:27:13and give me [Music] the drug they are referring to
428. 1:27:28searching my knowledge so this is good so it's it's doing what it's supposed to do so anyway there you
429. 1:27:35go so the answer is yes you can we just did it the term that an ending in d with gravel are there any other gr
430. 1:27:53words
431. 1:28:08the platform I'm on right now is chat gp4 so anyway there you
432. 1:28:23go the answer my friend is yes yes you can mhm fantastic Bob tell him what he's
433. 1:28:39[Music] won grind yeah like I don't listen I don't know how I obviously if you were going to do something like that you'd want to
434. 1:28:52go check your actual dictionary and check your words and make sure that it's all not hallucinating too bad but you
435. 1:29:01can you can add rules to the configure page there so what we just did if you didn't if you've been here was we made a GPT which is
436. 1:29:12a customized version of chat GPT that does whatever the [ __ ] you want it to do and if they're if they're about medical [ __ ]
437. 1:29:24financial [ __ ] or political [ __ ] for the most part you can't publish them in the GPT store you might be able to appeal it and
438. 1:29:35get something in depending on what it is but for the most part they're just not allowing that stuff Kyle ask get to draw
439. 1:29:42an image of someone who's taking it yeah I turned I turned off uh I turned off um image generation for that one for
440. 1:29:53that have very reason anyone used an AI medical charting software I get ads on Instagram for it I haven't um I work in a building
441. 1:30:05with a lot of Health Tech startups so I know I know there's ones out there I haven't played with any of them while I do a lot of work in
442. 1:30:18healthcare it's just it's like it's so conservative it it's hard for me to get excited about cuz everything takes for [ __ ]
443. 1:30:34ever yes you can what I came in after the question you can what what that GPT we just the question was can I create a large language
444. 1:30:48model that draws from a specific dictionary and so I went out and I found a slang dictionary and we U we made a
445. 1:30:57GPT uploaded the slang dictionary to the GPT and then that's it oh a human beings totally making up drug names
446. 1:31:06groovy new Sor vids Kyle I I went and looked for some I didn't see anything let me see open AI
447. 1:31:22Sora I mean this one I assume you've seen this is the I saw a cool oh you know there's is this the one that talks no
448. 1:31:39somebody did that video look closely what do you see I I I don't have time for [ __ ] riddles just tell me what you [ __ ] found
449. 1:32:07somebody did a version of that thing so there's there's that guy there's the scuba diving dude I've seen him that's a
450. 1:32:15SORA of video it's pretty interesting yeah this one did scenario. comom that's a thing I've got to add to that to that to the
451. 1:32:32list so scenario. comom that's the one I've been thinking of hang on let me add it to this list um scenario. comom is character
452. 1:32:46consistency software and someone took the alien dude and made a consistent character there so that one's pretty cool but I'm
453. 1:32:59looking for one there's one where oh they did the they did a zombie one someone made a voice over is this it
454. 1:33:15no let's see open a i Sora alien um
455. 1:33:30Pika
456. 1:33:52no [Laughter] maybe I didn't see it here was
457. 1:34:10it was let's see alien voice over or wait let's see voice no voice over let's see
458. 1:34:34script somebody wrote a script synthesized the voice in 11 Labs let me look up 11 Labs 11 Labs used Pika to make the mouth
459. 1:34:50move it was pretty slick if anyone knows where that is just put it somewhere and tell me that it's there
460. 1:35:00which AI platform should I use to teach a group of victims how to write their own stories um it depends Becky I mean I think if if
461. 1:35:14you want just quality of writing um chat GPT 4 um but if they've all got have access to it then I would say you could do chat gbt
462. 1:35:293.5 you could also do PI and the thing about doing it with pi Becky is that you could create a series of
463. 1:35:40prompts for them to interact with pi with and have them go have a conversation with pi and the nice thing about that is is it you know it it
464. 1:35:51records the transcript of what they said and what the pi response was that could be really interest I
465. 1:36:02think Pi Pi might be really interesting for that yeah I think I I don't know I would
466. 1:36:16go I would go pie on that one uh uh I really want to see this one with them
467. 1:36:35talking open anur Talking yeah I don't know I can't find
468. 1:36:51it that's a drag cuz it was it was really cool I'll I'll find it we'll we'll do it some other night I'll find
469. 1:36:56it in bookmarket maybe I saw it on LinkedIn hang on Sora alien what's
470. 1:37:22this
471. 1:37:52h yeah I don't know dang it what are you going to do is Claude 3 greater than
472. 1:38:09gp4 um it's way too early to tell for some things it's going to be better
473. 1:38:24if you need a really long context window it's going to be better what are we looking at what we were looking at there
474. 1:38:30was open AI Sora the the their video their video model that alien walking through New York City is just one of the
475. 1:38:38videos they had someone took that and and created voices and made it talk and it it was cool it was like it
476. 1:38:45was oh what did they have it do they I think they had Gemini Nar narrated it so they so they uploaded the video to Gemini had Gemini
477. 1:38:57narrate it then they took that narration created a voice in 11 Labs then they took that voice from 11 labs and took it
478. 1:39:07to P to match the lips to the voice then they edited it together and put music from sunno in the background I love that the Sora clip is a
479. 1:39:22turtle um uh back to back to Claude versus gp4 here here's the thing Claud Claude does not have an image generation component to it it can
480. 1:39:37see images it can do code interpreter apparently it's better at coding it's better at math it's got a longer context
481. 1:39:51window and it looks pretty slick but chat gp4 has been out there since last March and it's pretty sophisticated and
482. 1:40:05it can generate images and it can see and it can talk and it can here and it can code and it can do code
483. 1:40:14interpreter and it's connected to the internet cla's not connected to the internet so you're going to have that
484. 1:40:21[ __ ] where you want to do something up to date and you're going to have to jump over to chat GPT or to Microsoft
485. 1:40:28co-pilot and that's going to piss you off but if you need long documents done or if you want to have a really long
486. 1:40:37conversation with something open AI is going to piss you off so with all of these tools our only job right now play solve share play
487. 1:40:51solve share why why play go play with Claude as much as you can figure out where the boundaries are of it figure
488. 1:40:58out what it's good at and then when you get to solve a problem if you're like oh I need to read really long documents and
489. 1:41:06have the context window remain coherent for longer than a prompt or two you'll go use Claud because you'll
490. 1:41:14know that it's better at that than chat GPT is the only way you'll know that I mean you might eventually stumble into
491. 1:41:21me talking about it but the way for you to really know that is to just play with these tools enough to understand oh this
492. 1:41:28one's good at that this one's good at that this one's a useless piece of [ __ ] but I'll check in with it every couple
493. 1:41:33of months because maybe it'll get better there's a bunch of tools that I used to make fun of that are now on the
494. 1:41:40list of things I think everyone should be paying attention to CU the [ __ ] just gets better um so I don't know and it's way
495. 1:41:49too early and quite honestly about the time people people go okay Claude 3 is better than chat gp4 GPT 5 is going to
496. 1:41:59come out like they're going to drop something open AI is going to drop something maybe this week there's
497. 1:42:07there's weirdness going on at open AI right now so either they're having they're having morale issues after the sort of most intense team
498. 1:42:19building exercise in history um or they're about to drop something so we'll see we'll see um what time is it it's getting to
499. 1:42:32be time I'll show you one other thing I like the way Claude talks better than the way I like GPT talks yeah what what was interesting
500. 1:42:41is I had some I some people I read today said Claude they weren't running into safety issues safety guard rail issues
501. 1:42:50with Claude where they were with open Ai and I had the opposite issue where I got I got jammed up on Claude but then open
502. 1:42:58AI did it so I think just because they have different human beings programming them their safety guard rails are different
503. 1:43:05so that's another reason to just play with Claude is maybe it'll do something chat GPT won't the other thing that you should
504. 1:43:14know and this is this is this interface is a little wonky little janky is if you go to
505. 1:43:29perplexity which I know a lot of people in this channel are fans of perplexity it can now do images you have to be a pro member I
506. 1:43:40think but what's weird is so if I do create an image of the space
507. 1:44:00shuttle it says what specific view or angle do you want so let's go top view because that's kind of one we haven't seen
508. 1:44:12before and so now it does a search for images but if you look over here here's the images it found look down here at
509. 1:44:27the bottom generate image so I can say generate image and then it goes do you want a painting a photograph an illustration or a diagram
510. 1:44:37oh let's try a diagram bang and now it's generating an image up here so this is using playground AI but why it didn't just generate an
511. 1:44:49image for me and you know playground AI is not it's not Dolly it's not mid Journey but it's generating images so this is
512. 1:45:00interesting that it's kind of mixing [ __ ] that it's finding with [ __ ] that it's making but if I explicitly ask it to make me an
513. 1:45:09image it does an image search first so I find that weird and odd but that's an upgrade that's an upgrade for
514. 1:45:19perplexity seriously though is there a platform that can answer question questions Real Time with variability I don't walking blue I don't
515. 1:45:31know what you mean with variability what I don't know what that means do you mean like can you just have a conversation
516. 1:45:46with with one of these models yeah I'm with someone right now who's asking
517. 1:46:02if we can have a conversation that includes variability absolutely variability is my middle name Kyle feel free to engage me
518. 1:46:14in any kind of conversation and we can discuss anything from current events to personal experiences pop culture to
519. 1:46:21philosophy I'm all ears or should I say all chips so what would your friend like to chat about um the parasites of the common
520. 1:46:38llama wow Kyle that's a really interesting topic so it looks like the common llama like most animals can be susceptible to a variety of
521. 1:46:48gastrointestinal parasites including stron holes whipworms tapeworms lung worms and more these parasites can cause
522. 1:46:56a range of symptoms from diarrhea and weight loss to anemia and respiratory issues preventive measures like regular
523. 1:47:04deworming and good hygiene practices are key to keeping these animals healthy speaking of Health are beets
524. 1:47:16healthy oh yes Kyle beets are very healthy they're a great source of nutrients like fiber folet manganese
525. 1:47:23potassium iron and vitamin C they also contain unique compounds called balines which have antioxidant and
526. 1:47:31anti-inflammatory properties eating beets has been linked to a number of health benefits including I'm bored with health can we
527. 1:47:40talk about DND D can you roll me a D20 Dice and tell me what number you come up with you got it Kyle let's roll those
528. 1:47:51dice I got a 12 not too bad could be worse could be better what's going on in your D and D game right now are you in
529. 1:47:57the middle of a fierce battle exploring a mysterious dungeon or perhaps negotiating with a Sly Dragon I am but my friend only speaks
530. 1:48:07German so can you please tell him that the 12 means that his orc is dead but say it in German oh dear a fallen Orc in German
531. 1:48:22you could say to your friend D 12 D hopefully he'll understand Even In the Heat of battle it's always uh is that variability I don't I
532. 1:48:34don't know what what you mean variability [Laughter] [Music] non-deterministic meaning variability
533. 1:48:50within the human concept of emotion I still don't know what you mean like have you talked to Pi have you had a conversation with
534. 1:49:04pi because if that's not doing it I don't know what it I don't know what it means Pi Pi for me is the one that's got
535. 1:49:11the most emotional emotional intelligence it's designed for conversations not for typing it's designed for talking and it's quite
536. 1:49:22compassionate and it's it's like weirdly compassionate I I was trying to do a test with it once and I was talking to someone who'
537. 1:49:31started a um like an end of life uh like a hospice engagement uh service and I said to pie
538. 1:49:46that my mother was in hospice and my mother's passed away so so this was just a hypothetical thing and Pi came back it like Pi got
539. 1:49:56emotional it was weird Pi that that woman that woman's voice that I was just talking to was like oh no no no no no no
540. 1:50:04no Kyle that's horrible oh no like like it it was weird it caught me off guard how how like it understood that what I said
541. 1:50:17was way out of the bounds of normal for what we normally talk about so if that's what you're talking about that's it if
542. 1:50:24you're talking about variability of cadence of speaking not quite yet 11 labs and there's one of the other um voice
543. 1:50:35synthesis ones has emotions that you can tag in the API but none of them yet have done anything where they're doing an
544. 1:50:44emotional analysis of the script and and dynamically changing the variability of the voice if that's what you're talking
545. 1:50:51about but Pi Pi does it pretty good I want to hear about open AI leak cuz I fell asleep watching it this morning what was the open AI
546. 1:51:02leak hang on opening I leak this isn't an open AI
547. 1:51:16leak this is someone totally obsessed
548. 1:51:37they're plan to create AGI AGI by 2027 that doesn't seem like a leak to me Kyle if I were to incorporate five AI
549. 1:51:48tools into my college class to help students what would you say oh uh chat GPT for
550. 1:52:13perplexity five definitely those two perplexity for the for its research right it does citations and things like
551. 1:52:22that that aracas is qar Agi in 2027 with chat gp8 oh that I I don't okay I don't know anything about it so that's all [ __ ]
552. 1:52:36I've heard before so I I don't is there a new leak there is there is there new information CU all that the the new
553. 1:52:43thing for me is chat gp8 and AGI in 20127 but um we're in 2024 I think we're going to see a way before then um let me go to the AI
554. 1:52:58Salon let me get Cam Kat can an answer here um the five I would give college students chat gbt and perplexity
555. 1:53:14definitely and if they can't afford chat gp4 if they can't do the 20 bucks a month then substitute uh co-pilot since it does
556. 1:53:25Dolly it does Dolly and its gp4 and then I would I would add perplexity to it so that's two two two with an optional Microsoft
557. 1:53:38if they can't afford 20 bucks a month I feel like with image generation you're covered
558. 1:53:54because chat gpt's got do in it and so does Microsoft so I feel like with image gener and now perplexity
559. 1:54:02has playground in it but that looked pretty shitty so I don't think I'd add in another I don't think I'd add in another image generation
560. 1:54:15tool I would I would say [ __ ] what are they majoring
561. 1:54:31in because that that could I would say 11 labs for voice synthesis they should understand how
562. 1:54:47that works be able to synthesize voice and then I would say Hy gen or D-
563. 1:55:09ID and then probably Runway ml chat GPT 4 or co-pilot perplexity 11
564. 1:55:24Labs hey Jen or did for Avatar creation probably heyen because ha Jen also does the the video translation stuff and then Runway
565. 1:55:45ml but there could also be like really specific um like education support AIS that might be worth
566. 1:56:14it that's fine I don't give a [ __ ] hard to narrow it down right I was having lots of trouble picking just five
567. 1:56:27yeah it's it's really hard to narrow it down because it it's also depending on what they want like like if
568. 1:56:35it's if they're going into a really specific area I mean honestly cam I feel
569. 1:56:51like I feel like Beyond chat gp4 and perplexity everything else is optional it's almost like I like I would
570. 1:57:05make those two requirements and everything else optional and the only reason I'm throwing perplexity in there
571. 1:57:10is cuz chat GPT is not really good at the citation thing and and perplexity is getting really good at that so it's just a different kind of
572. 1:57:20tool but I feel like if the students could Master custom instructions and you know using an llm that's connected to the internet and
573. 1:57:31code interpreter and GPT vision and talking to it and you know using text to speech and and um generating images and
574. 1:57:43um making gpts like I feel like it's a a whole semester could be nothing but just mastering all of the tools that are
575. 1:57:52already in chat GPT cuz it's not just a single tool so maybe it would be more interesting what if you did this what if
576. 1:58:06you made it five five things that you wanted them to solve and then they can go use any tool that they can figure out how to solve it
577. 1:58:17and just have them have the things that you give them to solve require them to to use multimodal models or multiple models to
578. 1:58:27to pull [ __ ] together that might be a way to to do it cuz I think the tools are ultimately kind of
579. 1:58:45irrelevant what time is it oh it's getting late it's getting late here people I get to get on out a year I
580. 1:58:52thought co-pilot was 20 bucks a month co-pilot is free well Co okay co-pilot's 20 bucks a month if you want it inside Office
581. 1:59:03365 and it is it 20 or is it 30 I don't know it's 20 or 30 but co- pilot. microsoft.com is free if you want it integrated it costs
582. 1:59:21money [Applause] [Music] [Applause]
583. 1:59:39ah Opus only gives you five calls but I use perplexity all the more than I do anything JD interesting team MJ is there one with
584. 1:59:50memory of previous conversations Pi does it a little bit excellent thank you Kyle thank you Cam kaken glad that helped hope that helped
585. 2:00:01was it was not a great answer but it's it's actually it's really hard to um narrow that [ __ ] down um okay so let me get out of here
586. 2:00:18if you want to support the channel you you can follow me you can subscribe to the lives you can buy the video series that's in one
587. 2:00:29of those Corners that if you're if you're just new to Ai and you're trying to figure out is this a big deal or not
588. 2:00:35it's called underhyped for a reason I think AI right now is the most profound thing in the history of mankind and it's
589. 2:00:43underhyped not overhyped as much as people are talking about it and that video series talks about why um you can
590. 2:00:54also go to there the salon. click on the first link that goes to the AI Salon community and check out what we're about
591. 2:01:04and our values and if you resonate with the values if you're looking for a community of people digital Gods Danielle thank you very much I
592. 2:01:12appreciate it it's very generous if you go there um and you resonate with the values this is a kick-ass community of people that are
593. 2:01:21are choosing to be on this AI adventure together and like you can go there and nerd out and like no one's going to call you
594. 2:01:29a weirdo you can go there and be like I I got this idea for this AI thing or I just learned this or how do I learn this
595. 2:01:36and and this is a very engaged community that will that that are all trying to it's like very alive and like primordial
596. 2:01:44supy and there's guilds and there's clubs and there's resources it's really cool and it's free so my request would be if you go there
597. 2:01:56read our values make sure you resonate with them don't be a douche and introduce yourself first thing you
598. 2:02:03should do in this community use your real name and introduce yourself and tell us who you are tell us
599. 2:02:10what you're about tell us where you are in the AI Journey all right and that's it and then I would like to thank the subscribers of this
600. 2:02:22channel the Irregulars so there's like a pile of people that show up here every [ __ ] night it blows me
601. 2:02:29away and then there's a subset of them that actually mod this Channel and take all of your crazy ass comments and sift
602. 2:02:38through them and like pin pin the ones that I should be paying attention to they're kind of following my chat add
603. 2:02:46nonsense and they'll put up stuff to help me make sense of all this so that this is more value valuable to you guys
604. 2:02:52so thank you to all of you um yeah and that's it so I'm going to get out of here Mr it good to see you nudnick good to see you
605. 2:03:02Winston a human being Source Camp Lord digital Gods Danielle good to see you Brandon jet that's a new name just
606. 2:03:13signed up for the AI Salon awesome Brandon my request go introduce yourself in the in not to me to the in the
607. 2:03:19community section the first link is introduce yourself introduce yourself to the community good night thank you Vegas
608. 2:03:26MOX Moxy troen Clint Copperhead thank you so much very generous I really appreciate that Sherry D thank you so
609. 2:03:33much so addictive the salon it really is I learn a lot here great crab Aly thank you very much I appreciate that that's
610. 2:03:40the whole idea late night tomorrow maybe I think we have a uh I think we have an after hours in Denver but I'm
611. 2:03:53not sure I've kind of lost track of what's going on with them um possibly I'll make a video if it's going to be
612. 2:03:59late I will make a video tomorrow but it should be should be the same time night night Empower pixels thank you so much
613. 2:04:06life's tough it's tougher when you stupid thank you for all the gifts it means a lot it I appreciate it a lot um
614. 2:04:13against the norm jenu good to see you Winston thank you so much Brandon thank you for what you doing here it's
615. 2:04:19appreciated Tobias same Clint Copperhead as always steo from Down Under man lots of lots of friends in here tonight so
616. 2:04:30cool Tony Bridge digital Danny Papa JJ a human being Vicky good night good night good night good night good night good
617. 2:04:38night hello from Detroit Simon Says Tech nice all right I'm out of here happy Monday night everybody um I'll see you
618. 2:04:47tomorrow um pay attention it it's possible open AI drops something this week just pay attention to the Twitter's
619. 2:04:53and the LinkedIn and things something's coming something's boiling I I just got a spidey sense all right peace out everybody