Jan 1, 2024 - Unlocking AI's Potential: Create Your Own GPTs for Good in 2024

Description

Chapters

Transcript

0:02 [Music]
0:19 ready s in this long
0:23 time when things are going to
0:28 change my life
0:31 away seems these dreams have turned to
0:34 Bunch
0:37 glad get my nerve up but my past is
0:42 pulling me
0:43 [Music]
0:45 down wondering how
0:48 long this black sheep going to stick
0:51 [Music]
0:56 around somebody told me once before he
0:59 said you can never I'll go home
1:02 again won't you leave say a just to me
1:08 away from the truth of who I am and what
1:11 I believe so I thanked him for his two
1:15 SS with a hand shake and some
1:18 sympathy yeah I packed up my blue jeans
1:23 and I headed for this big
1:27 price of my freedom
1:32 byebye Black
1:34 Sheep the black sheep of the
1:38 family
1:42 bye oh that means so very much to
1:46 me
1:48 byebye bless she to my friends and my
1:54 family
1:57 by going off and set my
2:01 Soul set it
2:07 [Music]
2:18 [Laughter]
2:24 free get that dog some earplugs get me
2:27 some earplugs what are you talking about
2:30 that dog loves to
2:31 sing oh good lordy people happy New Year
2:37 happy New Year Steamboat Willie is uh in
2:41 the public domain go make Mickey Mouse
2:43 well go make that version of Mickey
2:45 Mouse I suppose I don't know the rules
2:46 on this [ __ ] but apparently you can do
2:48 something with him I'm sure Disney's
2:50 still going to sue your ass but you know
2:52 here we are making content with
2:58 Mickey oh good Lord I guess it's
3:01 Steamboat
3:02 Willie Steamboat
3:09 Willie good evening fine people fine
3:12 people of the lab the AI learning lab
3:15 the Internets the Tik
3:17 tocks the
3:20 world I didn't really check the hair
3:22 tonight I guess all right the glue's
3:24 still
3:25 holding oh what did I do today I watch
3:29 some fruit ball
3:31 I
3:32 uh I got
3:35 my I got caught up on uploading all my
3:39 AI learning Lab videos to YouTube I'll
3:42 give you the link for that cuz I also
3:43 made a
3:44 playlist for the GPT for good
3:49 um sessions that we did um so you can
3:53 learn how to do your own GPT you can
3:55 watch those sessions and as soon as I
3:57 have the the links for ANS Zoom videos
4:02 I'll add them to that playlist so this
4:04 playlist will will get updated with all
4:07 of the sessions from that 24-hour period
4:09 which will be cool that'll be a cool
4:11 archive to
4:13 have well cool for
4:16 us hopefully cool for
4:18 [Laughter]
4:22 you all right everybody welcome welcome
4:26 welcome welcome to the AI learning lab
4:28 how are we going to kick off the year
4:29 what do you want to do
4:33 too it looks like apple actually
4:35 released their freret model there's fret
4:40 7B multimodal model into the into the
4:43 wild
4:45 today so Apple Apple's starting the year
4:48 off with with an announcement it's a
4:51 minor one but it's still pretty cool do
4:54 you have three beverages going I do not
4:56 I only have two beverages going Disney
4:59 is Tainted
5:01 these are these are the the dose
5:04 dose oh look depending on if I hold one
5:07 closer this is an old fishing
5:10 trick if you hold the fish closer to the
5:14 camera it looks much
5:16 larger this is actually smaller than
5:18 this but is
5:24 it by the way this
5:27 idiocy welcome to Chad add
5:32 this is this is what neuro divergency
5:34 looks
5:36 like this is your brain this is your
5:39 brain on
5:42 ADHD and Amelia's wife yes you're late
5:46 you're late you missed champy was on
5:48 fire tonight he was going off he was in
5:52 fact he was just sitting here staring at
5:54 me like we're going we're going we're
5:55 going right we're going to go we're
5:57 going to go and the minute I tried to
5:58 tune my guitar he was like don't I was
6:00 like calm down just zip it zip
6:03 it he does not know the word zip
6:09 it that's a fine pup that pup that
6:13 champy all right fany ass let's get this
6:16 show on the road let's get it going
6:19 let's get her going all right listen if
6:20 you're new here my name is Kyle Shannon
6:24 and uh I'm what you would call an expert
6:26 influencer on a topic of artificial
6:29 intell
6:30 Ence I'm just it's I know everything
6:33 about it it's in fact if there's not
6:37 probably a single thing I don't know
6:39 about the artificial intelligence
6:41 sometimes people ask me what are your
6:43 qualifications I got a certificate I got
6:46 a I got a degree uh I got a I got a a
6:49 degree in a certificate uh I I have
6:52 learned in in the
6:54 past week or two all the AI I know it
6:58 all I know it all I know every answer
7:00 there's this thing called prompt
7:02 engineering it's going to be the big job
7:04 of 2023 it's going to be fantasttic what
7:08 it's
7:13 2024 I don't know anything people I know
7:16 nothing I don't know nothing there's no
7:18 experts there's no experts I promise you
7:22 I promise you that anyone that tells you
7:24 you're an expert they're not an expert
7:27 they're they're a charlatan they're a
7:29 scam artist they sound like this you can
7:32 make money with don't fall for it don't
7:35 fall for it don't fall for it I fell for
7:38 it once you know what I did I paid $1
7:40 199.95 a week you know what I got a
7:44 certificate it was useless in a
7:47 week all right do AO for champ and call
7:51 it something I the pin went away dang
7:54 it I love to cast you on my on my big
7:58 screen only the sound is a little bit
8:05 later I'm a I'm a regular [ __ ] oneman
8:08 show over
8:09 here do AO and champ and call it champy
8:13 Blues we could do that I gotta I Gotta
8:16 Give sunno enough enough we we actually
8:20 that's a that's a good idea champy Blues
8:22 okay all right so if you don't know what
8:24 sunno
8:26 is sunno is a um
8:30 a music AI that will write a song for
8:33 you and they've Incorporated it into
8:38 Microsoft
8:39 co-pilot but Microsoft co-pilot is
8:42 flakier than crap so let's let's let's
8:45 go let's go see if the first day of 2024
8:48 shows uh shows Microsoft co-pilot being
8:52 less flaky than it was for
8:55 the calendar year of
8:57 2023 I'm going to go out on a Limon say
9:00 no but but you know Sydney Sydney she's
9:04 a nasty a nasty nasty
9:14 woman all right oh by the way welcome
9:16 everybody my name is Kyle Shannon this
9:18 is the AI learning lab if you don't know
9:20 me I'm a bit of an odd
9:23 Feller um co-pilot we're going to go to
9:27 co- pilot.
9:28 microsoft.com
9:30 which is Microsoft's new branding for
9:32 Bing
9:34 chat see I've got zero out of two I've
9:36 got 2,000 tokens that generally means
9:39 this is not the fancy version of this
9:41 but I don't know we'll
9:43 see um
9:45 okay make me a
9:50 song uh make me a
9:54 blues
9:56 song about a black and and white
10:01 singing dog oh and because I did black
10:04 and white it's probably going to say
10:06 this violates its content policies for
10:09 for racism um a black and white singing
10:13 dog named
10:18 champy all right okay all right
10:22 fantastic let's see let's see if it does
10:24 it now it should hang on so in order to
10:27 to have this thing um make you a song
10:30 you need to have the sunno plugin turned
10:33 on and in order to do that you have to
10:35 have search turned on and thenso is the
10:38 little plug-in right below it all right
10:41 the the plug how they've Incorporated
10:43 plugins here is really weird because it
10:45 it requires you to turn search on and
10:48 then search seems to screw up chat GPT
10:51 or whatever the hell they call this
10:52 thing co-pilot okay so we're going to
10:54 submit that let's
10:56 see let's see if it does it people it
10:59 should fire off the plugin it should
11:02 fire off the plugin ah it did okay
11:04 that's good that's promising I'm finding
11:06 that switching from balance to creative
11:08 or precise makes helps when it flakes
11:11 out okay that's good to know Athena Luna
11:14 all right so let's
11:16 see late night when the moon was high I
11:19 hear howling in the distance my oh my
11:23 down in the street comes a black and
11:24 white dog singing the
11:26 blues oh what a talking dog oh champy
11:29 sing your heart out for the world to see
11:31 this is good when your Soulful how
11:35 you'll make Believers out of me oh
11:37 champy you're the star of the midnight
11:39 show it's so true sing in those B Blues
11:42 wherever you go
11:46 yeah gotta go you're a great human oh
11:49 thank you very much I appreciate that
11:51 all right let's listen to champy Blues
11:53 blues for
11:56 champy at night when the moon is high
12:00 [Music]
12:01 I hear aing in the distance my old down
12:07 the street comes a black and white
12:10 doll singing the blues what talking
12:15 do
12:24 horrible how make out of
12:28 me all right it's fine it's it's it's
12:33 it's where we are it's where we are in
12:35 the in the industry right now it's it's
12:38 a solid C
12:41 minus it's it's acoustically a D minus
12:45 uh chord progression a d and overall a c
12:51 c minus
12:56 minus this shit's going to get so much
12:58 better though
13:00 you just wait you just wait all you
13:02 naysayers sitting there with your arms
13:04 crossed going I hallucinates I don't
13:08 know if you knew this Kyle but it it
13:10 doesn't tell the truth all the
13:13 time it's stealing from
13:17 everybody oh I get it I understand I
13:22 understand your
13:23 reticence co-pilot wear co- pilot.
13:26 microsoft.com that's the micro roft
13:29 version of chat GPT that if if if we
13:33 assume that chat GPT is the is the state
13:37 of the-art right now chat GPT 4 so chat
13:39 GPT
13:41 plus let's give that an a we won't give
13:43 it an A+ because it's still too janky
13:46 it's irregular worthy for now yeah
13:48 exactly exactly if if if I give chat GPT
13:52 4 an a cuz it's state-of-the-art right
13:55 now nothing's even really close to
13:57 it until further notice
14:01 um co-pilot microsoft.com Microsoft's
14:05 version of chat GPT I would
14:08 give a c
14:12 minus like I would give
14:15 Claude like a
14:17 B+ I would give
14:19 Bard A B B
14:23 minus
14:25 um I'd give grock a
14:28 d
14:30 I'd give Pi a B+ but it's different but
14:35 for
14:36 conversational AI actually I'd probably
14:38 give Pi an A minus for conversational
14:44 AI um perplexity I don't know but it's
14:47 probably in the in the a
14:50 minus in the a minus
14:52 territory I just don't use it all that
14:54 much try some classical music themes I'm
14:57 good I'm good with sunno
14:59 uh Theo stuff's weird and predictable
15:02 like they don't they don't have quite
15:04 enough parameter like like they don't
15:06 quite have enough controls to control it
15:10 and write a full song it only writes
15:12 like a a Whata call it a verse and a
15:15 chorus it's not a full song so I'm done
15:19 with suo for a while I'll let it I'll
15:21 let it get one more one more version
15:23 better before I start playing with it a
15:25 lot it's it's like it's like all the
15:27 stuff right now all all the multimedia
15:29 stuff animation 3D objects text to 3D
15:35 World video
15:38 audio what else is
15:40 there
15:43 gaml all that sort of stuff right now is
15:46 all just it's it's just janky it's just
15:50 janky it's janky janky janky janky janky
15:54 janky janky janky janky ask chat GPT to
15:57 create a port cord progression and play
16:00 it for
16:01 us yeah but I don't know that I could
16:04 try
16:09 the it's it's it's okay at that it's not
16:12 great at
16:14 that um and I don't really know how to
16:17 prompt it right which is part of the
16:21 problem let's
16:28 see
16:32 champ will sing like a mother tonight
16:35 though I'll tell you that all right
16:37 let's see um you
16:41 are are a
16:44 Grammy
16:46 award
16:48 winning
16:50 Americana singer song
16:56 writer
16:57 you
17:00 are
17:02 inspired
17:05 by let's
17:10 see Brandy
17:13 Carlile Lucinda Williams Martin seon
17:17 John
17:25 hayatt that's
17:27 good
17:32 I want you to write a
17:36 haunting
17:39 song about a black and
17:43 white dog named
17:51 champ who
17:54 can't help but
17:57 sing when whenever
18:01 anyone strums a
18:06 guitar The Ballad of singing champ in a
18:09 little town under Big Sky
18:11 Blues all
18:13 right oh champ sing your Soulful cry
18:16 underneath the vast Open
18:21 [Music]
18:27 Sky now now write the
18:33 chords for the
18:37 song that
18:40 are
18:43 interesting and
18:45 non
18:48 cliche and this is where like it it
18:50 tends to not be great at
18:57 the well well it's actually good enough
19:00 that I don't know how to play a lot of
19:02 these
19:03 chords all right I'm going to go wait
19:06 okay simplify them a
19:11 bit that's really
19:16 funny a minor gcd okay well
19:19 let's
19:22 see oh a minor that's Champ's chord
19:27 actually right so
19:30 [Music]
19:47 [Music]
19:57 yeah
20:00 [Music]
20:05 [Applause]
20:09 that's actually not bad did you like
20:11 that champy that's your song buddy
20:13 that's your song buddy we made you a
20:16 [Laughter]
20:18 song oh ask it to rewrite oh that's
20:22 really good ask you to rewrite the
20:24 Constitution but from our founding
20:26 mothers that is a good prompt
20:29 all
20:34 right you are a renowned
20:40 historian with a
20:43 reputation for finding documents no one
20:46 knew existed
20:54 period it turns
20:57 out that while the founding fathers were
21:00 writing the US
21:02 Constitution their significant
21:09 others gathered as
21:12 well
21:14 period Please
21:17 write the alternative
21:20 Constitution from the founding
21:25 mothers
21:27 certainly this is what I love about chat
21:30 PT is just like hey I got you I got you
21:33 sure no
21:34 problem we the women of the United
21:36 States in an effort to form a more equal
21:38 and nurturing n Nation nurturing
21:41 establish justice ensure domestic
21:43 tranquility promote the general well
21:46 welfare and secure the blessings of
21:48 liberty for ourselves and our posterity
21:51 do ordain and establish this
21:53 constitution for the United States of
21:56 America article one the legislative
21:59 Branch every state shall have
22:00 representation in proportion to the
22:02 population including all persons
22:04 regardless of gender ensuring women have
22:06 equal representation in the legislative
22:09 decisions of our
22:10 nation voting rights shall be Universal
22:13 and unimpeded for all citizens of legal
22:14 age irrespective of gender race or class
22:17 Congress shall make no law that hinders
22:19 the right to education and shall
22:22 Endeavor to promote welfare this is very
22:24 very cool yeah and you know I I mean I
22:27 did a i did a similar kind of exercise
22:30 to this it wasn't wasn't exactly the
22:31 same thing but I did a kind of exercise
22:34 to this of
22:37 um um when the when Biden's executive
22:42 order on AI came out a lot of people
22:44 were asking me my opinion on it and I
22:48 Tred I try to look at things from a lot
22:49 of different angles and that one in
22:51 particular I was like man people from
22:53 with different points of view are going
22:55 to have very different responses to this
22:57 you know some of it's just po politics
22:59 some have will have legitimate issues
23:01 one way or the other and so I had chat
23:04 GPT I came up with five personas and
23:08 then I had chat GPT tell me if I was
23:10 missing any so we came up with eight
23:12 eight different personas so a
23:14 Libertarian a conservative a progressive
23:17 all that sort of stuff and and then um
23:21 and then I Had Each Persona react to the
23:26 executive order you know what what was
23:27 their opinion on the executive order it
23:29 was fascinating I did a LinkedIn post
23:31 about
23:31 it hang on the LinkedIn post
23:42 was is this going to let me
23:50 in is this going to let me
23:56 [Music]
23:57 in
24:01 articles so if you go to my LinkedIn
24:03 profile which is Kyle
24:05 Shannon in my
24:07 articles there's an article
24:11 called but what do they think of the
24:14 executive
24:20 order and so I've
24:25 got an overview of the executive order
24:28 and then the company exec at a major
24:30 Tech firm the AI
24:34 skeptic the regulation proponent the
24:38 libertarian the generative AI Optimist
24:41 and Community leader that's
24:42 [Laughter]
24:45 me the AI researcher and academic the
24:48 consumer rights Advocate that was
24:50 something that chat GPT came up with the
24:53 international Observer oh there's nine
24:55 of them and then the small Tech startup
24:59 founder that was that was a pretty cool
25:02 little
25:04 exercise and it's I mean you know it's
25:06 it's kind of thing
25:09 where because of the way AI was trained
25:12 it was trained on you know all of this
25:14 output of human beings right that that
25:17 put all this stuff on the internet you
25:20 like inherent in the training data even
25:22 though there's bias and there's all all
25:24 the sort of stuff that people talk about
25:25 that needs to be addressed which I agree
25:27 with
25:30 if you think about the amount of
25:32 information that's been put on the
25:34 internet in the last 40 years it's a lot
25:37 and and there's a lot of point of points
25:38 of view represented there and if you do
25:40 your prompting
25:42 cleverly and you do exercises like this
25:45 where you're not just necessarily
25:47 writing from a single point of view
25:48 you're writing from multiple points of
25:50 view it like chat GPT is actually really
25:54 good at Nuance it's really good at at
25:57 you know know helping explain things
25:59 from different pe people's points of
26:01 view and so I think this kind of
26:03 exercise is is actually a really
26:05 valuable a valuable kind of thing this
26:08 it like it opened my eyes I thought this
26:09 was an interesting an interesting little
26:12 piece so
26:13 anyway for what that's worth for what
26:16 it's worth you know what I mean you know
26:18 you know you know what I'm saying H you
26:20 know hey you could read it you could not
26:22 read it hey it doesn't [ __ ] matter
26:23 I'll tell you one thing you could do you
26:25 could take the text of this article you
26:26 put it in the chat GPT and you say hey
26:28 just give me the skinny like what the
26:30 [ __ ] does what the [ __ ] he saying here
26:32 huh what's he saying
26:34 here and then you go in there and you
26:36 say hey I like I like it I don't I don't
26:38 a boots whatever you know what I'm
26:42 saying all right
26:44 fantastic I remember that day Kyle hey
26:47 Robert
26:48 Rossy using personas really helps me
26:51 draft a prompt that oh I just closed
26:54 your thing sorry about that I couldn't I
26:55 didn't read the whole pin whoever put it
26:57 there
27:04 [Music]
27:06 there how many gpts did you get set up
27:09 in the oh wait uh using personas really
27:11 helps me draft a prompt that shows
27:14 perspectives yeah actually I'll tell you
27:16 what that would be a great GPT to build
27:19 okay everybody all right let's go let's
27:21 go listen listen listen people listen
27:24 let me tell you
27:25 something you're going to go right now
27:28 to this URL no not that one don't go
27:32 there go here go to that one all right
27:35 go to the salon. and it might not work
27:38 cuz I don't I think my web my register
27:40 is a piece of [ __ ] just like it's like
27:43 all this AI technology so go to the
27:48 salon. and that's going to take you to a
27:50 link tree and I've updated some things
27:52 the first link is the GPT for good event
27:58 directory that uh that Vicky Baptist
28:01 built for us so if you don't know what
28:03 this is um on Saturday this past
28:08 Saturday starting at 6:00
28:10 a.m. um Ann Murphy of empowered
28:13 fundraiser and myself kicked off this
28:15 24-hour event called GPT for good it was
28:18 the it was the 24-hour GPT for good
28:21 challenge and our goal was to to to have
28:25 participants build 100 gpts for
28:27 nonprofits or good
28:29 causes and so I am happy to
28:33 announce that we hit as of today
28:38 151 and so this site shows you all them
28:41 so what did we build in the first 24
28:43 hours here's the first 100 gpts that
28:46 were
28:48 built and
28:51 then we hit our goal of 100 gpts and
28:54 kept going so here's the next 50 or 51 I
28:57 guess um isn't that amazing and so so
29:01 that's
29:03 uh that's what we did over the
29:05 weekend it was long and exhausting and
29:08 Vicky bless her heart not only did she
29:10 build this site she stayed up so an and
29:13 I had six hours in between our sessions
29:15 so every 3 hours we did a training
29:18 session and um
29:22 and but Vicky was there for all of them
29:25 so she just she didn't take naps I don't
29:27 think I think she just grinded
29:30 through and so she was updating this
29:34 this directory list in real time as
29:36 these things were getting built uh and
29:38 it's just it's really inspiring so
29:40 please go here and and check them out
29:41 and share this and and and you know if
29:46 you want to do this you know what I'm
29:47 going to do I'm going to turn back on
29:49 I'm going to turn something back on on
29:51 the link tree
29:55 here you can no longer sub submit them
29:58 for the there's a
29:59 $250 drawing that we're going to
30:01 announce tomorrow at the AI
30:04 salon so you're no longer eligible for
30:06 that but I'm going to give you access to
30:12 the there's a
30:14 document that is the prompt sheet that
30:17 we used for these courses so that's now
30:19 turned back
30:22 on and where did I put it yeah that's in
30:25 a good spot all right right oh wait hang
30:29 on e e slash
30:32 Ai
30:34 No AI
30:37 artist a i there we
30:42 go come on what is going
30:51 on there we go um okay so the first the
30:55 first link is to that that directory the
30:59 second link is to the YouTube
31:03 playlist um which includes the night
31:06 before which there's nothing valuable in
31:08 that one it was just
31:09 fun um but but the one starting at 9
31:13 a.m. uh and started at 6: a.m. I don't
31:16 have ANS in here yet but these are the
31:18 sessions where where we taught people
31:21 how to make
31:22 gpts so you can you can go watch those
31:26 videos and play along and then here is
31:28 the prompt sheet that is this is
31:31 basically the the prompts and and the
31:34 instructions for how to build your GPT
31:38 and how to share it with the
31:40 world and uh
31:43 yeah so
31:46 151
31:49 151 that's a lot that's a lot of that's
31:52 a that's a whole lot of
31:54 gpts um for for an event that was just
31:57 kind of a like a fever dream idea the
32:01 week before came together pretty
32:05 good it helps to have some some kick-ass
32:08 co-conspirators um an an Murphy and and
32:11 Vicky both Vicky Baptist and and also
32:13 the Irregulars so one of the things that
32:16 made it a success is a lot of the
32:17 Irregulars a lot of the people that come
32:19 to this channel regularly showed up to
32:22 to multiple sessions of that event and
32:24 just kept building
32:26 gpts like TK made um 20 of them or or
32:30 more and and lunik had made like 74 of
32:34 these things previously and so she was
32:36 pulling out the the ones that were tied
32:38 to nonprofits and put it was just it was
32:40 amazing it was just amazing so anyway
32:43 thank you to all all who involve all who
32:46 were involved who's
32:48 programming who's programming chat
32:51 GPT open AI Greg Brockman
32:54 and and uh the gang over there at open
33:00 AI or the one who's programming it who's
33:03 programming the gpts we did anyone who
33:05 wanted to that's the the amazing thing
33:07 about way the way they built the gpts is
33:10 you don't need to know programming at
33:11 all there's no code tools where you
33:14 actually do need to understand
33:16 programming you need to understand apis
33:18 and you need to understand programming
33:20 logic and you need to understand inputs
33:22 and outputs
33:24 and these things you don't need that you
33:27 it if you have those skills you can make
33:30 more sophisticated gpts but you can
33:33 literally in 5 minutes just talk a GPT
33:36 into existence my very first GPT I did
33:39 this thing called Sin City sipper I did
33:42 it in five minutes on on a
33:45 live with no programming and it was
33:49 brilliant wait who's making the guard
33:51 rails chat GPT open
33:54 AI open AI is making the guard rails or
33:57 in the case of co-pilot Microsoft is
34:00 making the guard rails or in the case of
34:01 Bard Bard is making the guard rails or
34:04 in the case of llama from meta no one's
34:06 making the guard rails because they gave
34:08 out the the source code they gave out
34:11 the
34:11 weights the one of the significant
34:14 things that meta
34:16 did was they kind of ensured that these
34:18 large language models are going to
34:20 survive whatever assaults come from
34:25 politics or or industry
34:28 we just saw a big lawsuit from The New
34:30 York
34:31 Times
34:33 um so so I mean your question about
34:36 guard rails I like it sounds like it's
34:38 coming from a place of of you know like
34:41 like you don't trust anyone and who's
34:42 making a guard rails well there's
34:45 educate yourself because there's a lot
34:47 of different models out there um open AI
34:51 was intended to be an
34:53 open-source company it was a nonprofit
34:58 and Sam mman in realizing how powerful
35:01 these things got was like I don't think
35:03 it's appropriate to put this level of
35:06 powerful tool out there with no guard
35:09 rails on it that that you know people
35:12 could do bad things with it now you
35:14 could argue well who's he to say what
35:17 the rules are well at least he's
35:18 thinking about it and then you look at a
35:22 company like uh like
35:25 anthropic that is
35:27 uh founded by a bunch of exop AI people
35:31 that didn't like the way open AI was
35:33 doing safety they're they're human
35:35 reinforcement learning they didn't like
35:36 that model so they came up with a model
35:38 called the Constitutional learning model
35:41 and what the Constitutional learning
35:42 model says is let's create a set of
35:44 rules and rather than having human
35:47 beings um determine if it was a good
35:49 answer or a bad answer why don't we run
35:52 it against those set of rules and so the
35:55 nice thing about anthropic in that
35:56 Constitution model is you can make that
35:58 set of rules available to everyone and
36:01 it can be
36:02 transparent how I imagine this is going
36:04 to go in the future spike is um there
36:08 will be different models out there and
36:10 we will all opt into our particular
36:13 level of safety or our particular flavor
36:17 of safety right there's going to be some
36:20 people that look at you know what open
36:22 and AI is doing going that's perfect for
36:24 me there's going to be other people that
36:25 like I wanted to lean this way or that
36:28 way or I wanted to skew you know in
36:30 favor of this or in favor of that there
36:33 is no
36:35 single nonbiased
36:38 thing because we're all humans we've all
36:42 got
36:43 biases so so every model is going to
36:46 piss someone off right and if you really
36:50 want to play with a completely
36:52 unrestricted model have
36:55 fun it's going to cuss a lot it's going
36:58 to do a lot of nasty [ __ ] it's going to
37:00 randomly lose its mind and just start
37:02 speaking a different language um I have
37:05 a buddy who took the UN
37:08 unrestricted version of um llama 270
37:13 billion parameter model and it's running
37:16 on his company's slack Channel as like a
37:19 participant and they're training they're
37:22 training this model they're fine-tuning
37:24 this model using their slack
37:26 interactions with
37:28 it he said it's batshit crazy he said
37:32 it's absolutely
37:34 insane it'll just lose its mind it it
37:36 cusses like a sailor sometimes it's like
37:39 it's it's like seriously not safe for
37:42 work but he's trying it as an experiment
37:45 so so it's crazy like you know there
37:49 there's again
37:52 that it's very easy right now to sit on
37:55 the sidelines and point fingers and go
37:58 yeah but who's writing the rules and and
38:01 me start playing with the tools start
38:04 learning them rather than just take it
38:07 at face value that oh these models are
38:11 biased learn how they're biased do your
38:14 own
38:15 tests you know check with prompting like
38:18 you know I I see this happening on Mid
38:20 Journey a lot that that people are
38:22 getting in there and the mid Journey
38:24 models have gotten better over their
38:26 model at being less biased because
38:29 there's feedback of people using it and
38:31 doing it don't just sit on the sidelines
38:33 and sort of point fingers
38:35 because there's no way that these here's
38:37 the deal this generative AI stuff is so
38:41 new that of course there's going to be
38:45 problems with
38:46 it how they were trained is they were
38:49 trained on an imperfect data set which
38:52 is the public internet the public
38:54 internet is by definition biased
38:58 to people who had access to
39:01 technology that had whatever
39:04 demographics the early geeks on the
39:06 internet were which did not
39:09 skew super
39:11 balanced to put it
39:14 kindly but that's what it is that's what
39:16 it's been trained on now can we improve
39:18 that we can improve that how do we
39:19 improve that if we know how it's biased
39:22 if we know the direction we want it to
39:23 go in and the only way you can do that
39:26 is by playing with these things and
39:28 understanding this one's better at that
39:30 this one's better at that here here's
39:32 how I could combine these two things
39:34 maybe maybe if you want balance you
39:36 figure out four or five different models
39:40 that all have sort of different
39:43 biases and for any given thing where you
39:45 want a balanced approach you use four or
39:48 five different models or use
39:50 20 why
39:53 not but the thing
39:55 that like NWS at me like a like a
39:58 [ __ ] allergic reaction is when I hear
40:00 well how are they doing this I don't
40:04 know or or just they're doing it wrong
40:07 well then you go build a Frontier Model
40:10 you go spend $10 million training a
40:13 model and make it
40:15 perfect we can write rules can have
40:18 dreams now how about an MLK GPT Builder
40:21 day great yes put it put it together
40:24 love it that would be awesome
40:30 sounds sounds
40:32 fantastic I'll promote
40:34 it I dig it I dig it I dig it dig it dig
40:38 it dig it dig it dig dig dig dig dig dig
40:42 I heard big names saying that Lang chain
40:44 and Vector DBS are Superfluous for AI
40:48 thoughts are
40:55 superfluous
40:58 no I don't think they're going to be
41:00 required right like if if you look at so
41:04 so for those of you that don't know Lang
41:08 chain and Vector databases allow you to
41:12 take your own
41:14 data and create a
41:17 fine-tuned GPT model using you know
41:21 whatever model as the
41:25 foundation and then
41:27 and then with the ve Vector database you
41:28 can be searching on your own data and
41:30 and you can essentially put in an
41:32 infinite amount of data like however big
41:34 your database is
41:38 um that that as a that as a
41:42 um a tool for where the technology is
41:45 today is super valuable because it's not
41:49 easy to do that stuff right so you you'd
41:51 have to write all your own code so Lang
41:53 chain makes it really easy to do that
41:55 now you look at something like the gpts
41:57 that that open AI just put out that
42:00 allow you to upload your own knowledge
42:02 to them that takes a little bit of that
42:05 functionality of Lang chain and Vector
42:08 databases and kind of makes it consumer
42:10 friendly you know you want to make a GPT
42:13 that understands your company's
42:15 documentation just upload it here right
42:18 if you wanted to do that prior to gpts
42:21 you had to run Lang chain but to say
42:24 that it's Superfluous is kind of silly
42:27 like I think I think by the end of
42:30 2024 it will
42:32 probably not be nearly as necessary
42:36 nearly as valuable but I would also
42:38 Imagine That Lang chain and Vector
42:40 databases and things like that are going
42:42 to evolve as new technologies come
42:45 along and and they will probably augment
42:48 the weaknesses of commercial products
42:51 that are launched um like they always
42:54 have so what's the difference between
42:56 all the the Llama models like 70b Etc so
43:00 it it it really is just size um so the
43:03 the 70 billion parameter
43:06 model um is is bigger in terms of just
43:10 hard drive space how much you need to
43:12 run it and then it's bigger in terms of
43:15 requirements to actually run like you
43:18 need you need to be running it on
43:20 computer instances with big gpus and
43:23 things like that and then so so they're
43:27 all sort of in the same dire they all
43:29 kind of do the same thing it's just you
43:31 know they are smaller and smaller data
43:34 sets what's been really cool about the
43:37 opsource community is that they have
43:40 been taking the different models 70 what
43:44 are they the the seven the 7 the 13 the
43:47 44 didn't work I don't know if they've
43:48 replaced that yet there was a 7 the 13
43:51 the the 40 and then the the 70 what the
43:55 open source Community has been doing is
43:56 is they've been taking those models and
43:58 they've been doing things like
43:59 optimizing them because they have access
44:01 to everything and the weights they're
44:03 doing experiments like what if we took
44:06 out 90% of the data in this model and
44:09 made it really small does it still
44:12 function turns out it does turns out it
44:15 actually functions pretty well so now
44:17 there's like hundreds of variations of
44:20 these models and there there are ones I
44:23 just saw one two days ago they just
44:25 announced there's there's a new one
44:27 that's
44:28 designed it's a large language model
44:30 designed to be installed on a Raspberry
44:32 Pi wait do I have my little Raspberry Pi
44:35 here I think I
44:37 do where is it
44:40 yes
44:43 this little Raspberry
44:46 Pi you can you can now install a
44:50 functional
44:54 multimodal large language model that
44:56 that runs on a Raspberry Pi low power
44:59 they cost 40 or 50
45:02 bucks and you can have a multimodal
45:04 something that can analyze images can
45:07 answer questions can write stuff now is
45:10 the thing that runs on a Raspberry Pi
45:13 anywhere
45:14 near CAD gbd4 [ __ ] no not even
45:19 close but you don't need chat gp4 for
45:23 everything right if you just want
45:25 something that can look at an image OCR
45:28 it analyze it organize the information
45:31 for you and and give you that back that
45:33 could be super valuable for a remote
45:37 Village somewhere that that you know
45:39 wants to do a particular kind of
45:42 task and and I think I think one of the
45:44 things that we're going to see in 2024
45:46 and in fact the the
45:49 um the faret model that Apple just
45:52 released is a 7 billion parameter model
45:55 I don't know if they released all the
45:56 weights or everything I don't I don't
45:57 know uh if it's open source I don't I I
45:59 didn't pay a too too much of a too much
46:02 attention to it um but I think one of
46:05 the things that you're going to see in
46:07 2024 um Google released Llama Or Not
46:13 llama uh um Gemini Nano which is
46:16 designed to run on devices without
46:19 needing an internet connection the open
46:21 source community's got a bunch of these
46:23 things that don't need an internet
46:24 connection you know apple have one we're
46:26 going to start to see 2024 I think is
46:29 going to be the year where um Hardware
46:31 capability goes up and Hardware
46:34 requirements go down for these models
46:36 and you're going to be able to run very
46:38 very powerful large language models with
46:41 no connection to the Internet so that's
46:44 better for safety it's better for um
46:47 speed responsiveness because you're not
46:49 having to rely on an internet connection
46:51 and have it processed out there and then
46:53 be sent back what raspberry model I I
46:56 think it's I think the we had so if you
46:59 go look at
47:01 um go look at the Joker if you go look
47:04 at the
47:06 um AI Salon YouTube channel we had a guy
47:11 come in who who actually saw me talking
47:13 about this on here um and he said he had
47:17 successfully installed a large language
47:18 model on a on a Raspberry Pi I think he
47:20 said it's an 8 megabyte um an 8 megabyte
47:25 minimum
47:27 but I think a 16 megabyte Raspberry Pi
47:30 was better or something like that but I
47:31 think that they they've all what's
47:34 what's nice about the open source
47:35 Community is they've got all the
47:36 requirements they'll say this model is
47:38 you know has been tested on the
47:40 following configurations so you can get
47:43 yourself configurations that fit that um
47:46 but I think the one that was just
47:47 announced two days ago is really
47:48 designed to be run on really shitty
47:51 raspberry pies so it's pretty pretty
47:53 sweet what's happening Todd Happy New
47:56 Year my friend thank you so much thank
47:58 you for your explanation
48:00 absolutely it's a false assumption that
48:02 there is a diverse selection of AI
48:05 advancement technology it's a false
48:08 assumption that there's a diverse
48:10 selection of AI advancement
48:13 technology I would well if you're
48:17 talking about the frontier
48:19 models if you're talking about if you're
48:22 talking about building models that
48:24 require billions of dollars of
48:26 infrastructure and investment then I
48:28 would agree with you if you go look at
48:31 what's happening in the open source
48:32 Community right now I would say
48:35 that it is also a fun a false assumption
48:39 to assume that there is not a lot of
48:43 advancement happening outside of those
48:45 big
48:46 companies um I I appreciate your
48:50 cynicism but I don't agree with it um
48:54 are we in a perfect place no
48:57 but we're not in a perfect place in
48:58 society we can't [ __ ] talk
49:01 about basic concepts of democracy
49:05 without wanting to kill each other right
49:07 now in America like I think AI bias is
49:11 the least of our problem right now how
49:14 about the the [ __ ] bias in our
49:16 country how about we get back to being
49:17 able to talk to one another as if we
49:19 were human
49:21 beings isn't that a bigger
49:25 problem
49:31 it's might of quiet in here Jed did you
49:34 say something I hear
49:43 nothing oh that's cool down down in the
49:45 mud has that raspberry pie case yeah I
49:47 like that raspberry pie case it's cool
49:48 isn't it oh my God the Mickey pick is in
49:51 honor of its transition into the public
49:52 domain it is it is indeed multimodo llm
49:56 and every pocket in the future exactly
49:59 and and like the software for these is
50:01 going to get really cool like you're
50:02 they're going to tie it into your camera
50:04 on your device so you'll be able to just
50:07 you know I don't know take a picture of
50:09 something and have it analyze it and I
50:11 don't know it'll do cool [ __ ] with it
50:13 and it and all your data stays private
50:15 it's all yours I prompted write a story
50:18 sad bird still sing it wrote a 7,000
50:21 word story oh that's cool interesting AI
50:24 has only one rote that has been released
50:26 to the
50:28 public no that's not true it's not true
50:31 stability AI released theirs it's AI
50:34 only has one route that is popular the
50:38 Llama one because they they released it
50:40 for commercial
50:42 use um but there's stabilities out there
50:46 there's a couple of Open Source projects
50:48 on it there's a there's an open- source
50:50 project right now where they're trying
50:52 to build floating GPU Farms that they
50:55 can can put on barges and Float outside
50:58 of inter in international waters so no
51:01 country can um can control them and
51:05 they're trying to raise enough money to
51:06 to build their own military to protect
51:09 these things so there are efforts there
51:13 are efforts right now to be able to
51:17 build Frontier models outside of the
51:19 current INF power infrastructure so
51:22 again I would say educate yourself there
51:24 there's a lot going on right
51:26 now um I was trying to install my GPS
51:30 ways in a Raspberry Pi that's
51:33 awesome what Raspberry Pi model I'm
51:36 using Da Vinci 3 on a pi python that
51:40 accesses various API
51:42 Keys Kyle remember the promise software
51:45 no I don't I don't know what that
51:48 is let's see if chat GPT does I bet it
51:53 does uh uh uh
51:57 or or do you mean the promise software
51:59 wait what was the
52:04 promise
52:13 software the promis software refers to a
52:16 notable legal case management of
52:19 software system that was developed in
52:20 the 80s by Bill
52:24 Hamilton it allowed for efficient
52:27 management analysis of large amounts of
52:29 data which were crucial and compx legal
52:32 is this what you're talking about
52:33 tracking and Reporting customization
52:35 Innovative for its time software gained
52:38 some notoriety due to controversy
52:40 surrounding its distribution and use
52:42 there were allegations that the software
52:44 was illegally appropriated and sold to
52:46 foreign governments leading to a series
52:48 of legal disputes and conspiracy
52:50 theories these controversies have often
52:52 overshadowed the technical and
52:53 functional aspects huh now didn't know
52:56 anything about it that's interesting a
52:58 private military for GPU barges sounds
53:00 hella dystopian I have to admit but cool
53:03 yeah listen I I'll tell you what like I
53:05 don't care if if if they actually raised
53:10 the money to to to build you know 10,000
53:14 GPU barge
53:16 farms and and and they raised their own
53:19 military and some other country wants
53:23 those things not to exist those things
53:25 are going to not exist it's like you
53:28 know it's like bad enough like going
53:30 after one country's military but if
53:33 you're going to go after like China and
53:35 Russia and the US and it's like but
53:40 whatever listen I you know more more
53:42 power to people it
53:45 like what I think's going on right now
53:48 is that is
53:50 that people understand that these
53:54 Technologies are are
53:56 profound and because of that all of the
53:59 concerns that everyone's talking about
54:00 are valid to be talked about like we
54:02 should be talking about them we should
54:04 be talking about who writes the rules we
54:05 should be talking about how you debias
54:07 things we should be talking about
54:09 alignment and safety um it's it's the
54:12 the primary thing that um you know Andre
54:15 karpathy came back to open a AI to run
54:18 super alignment so super alignment is
54:22 are the are the ai's imperatives aligned
54:25 with human
54:27 imperatives because if the ai's
54:30 imperatives are not aligned with human
54:32 imperatives at some point they're going
54:34 to go after humans because they threaten
54:36 them so like that alignment problem is
54:40 massively important
54:43 massively we're just a little early for
54:46 you know all the all the doomsday
54:48 because it's like we're not there
54:53 yet but the fact that this is all a con
54:55 ation I think is a really good thing
54:57 when people say AI will destroy Humanity
54:59 I answer Humanity's been doing that for
55:01 a long time for a millennia I yeah right
55:04 right now we we are the greater Threat
55:09 by far absolutely absolutely Papa JJ
55:13 humans are the [ __ ] it's like well
55:16 the AI hallucinates have have you talked
55:19 to a person they [ __ ] lie and scheme
55:22 and
55:24 connive it's like
55:26 so like the the AI I'll tell you what I
55:29 have had more nuanced powerful
55:31 conversations with AI than I have with
55:35 humans in in the past year it's it's
55:38 it's a it's a different thing you want
55:41 Skynet that's how you get Skynet well I
55:44 don't want I don't want anything I'm
55:46 just telling you what's going on the
55:47 robots aren't going to kill us who
55:48 controls AI controls the world that's
55:51 the I'll tell you what one thing I agree
55:54 with you on Spike is is
55:57 that I think that in the next 3 to 5
56:01 years so so the next year or two is just
56:03 going to be the technolog is going to
56:05 get more and more powerful and then
56:07 we're going to start to
56:09 see attacks on the current power
56:12 structure and not attacks like like
56:14 hacker attacks I mean like business
56:17 models are going to be up and upended
56:20 and and just weird shit's going to start
56:22 start to happen where smaller
56:23 competitors are competing with really
56:25 big companies and it it it won't make
56:28 sense and then I think in the three to
56:30 five year range we're going to start to
56:32 see the centers of power get
56:35 threatened and there's going to be all
56:37 sorts of wild death throws it's going to
56:39 be
56:39 crazy can we get back to making funny
56:44 turtles absolutely we can absolutely get
56:48 back to making funny turtles yeah this
56:50 is a pretty heavy heavy New Year's thing
56:54 um
56:56 we can train ai's Behavior faster than
56:58 humans be nice right yeah be nice to
57:01 your be nice to your robots because they
57:03 could help you someday if I could only
57:05 get out of the house 99% of the world
57:08 wants nothing but peace and just to live
57:10 their lives out yep I agree with that I
57:14 trust AI more than a lot of people even
57:16 if AI hallucinates it doesn't currently
57:18 use manipulation and lies you can get it
57:20 to use manipulation and lies but it
57:23 it it uh it can you can
57:29 um you can also get it to be reasonable
57:34 and it's it's hard you know what's funny
57:37 is one of the things that's caught me
57:39 off guard about AI is
57:42 um you know there are certain words
57:45 in the current political situation you
57:48 literally just can't say the word it's
57:50 one of those words that's been
57:51 weaponized by one side or the other and
57:54 you can't say it
57:56 to humans because it just you can just
57:59 see you see their shoulders go up you
58:01 see the you know people get emotional
58:03 when you say certain words right now
58:05 just because of of how politics has been
58:08 weaponized um one of the things that's
58:11 caught me off guard is I'll find myself
58:13 like editing what I'm tying typing into
58:15 chat GPT and I'm like oh I don't need to
58:18 edit myself here I can actually say that
58:20 word and then and then I'll realize oh
58:23 you know I actually don't I've never
58:25 never really thought about that word I
58:26 just know I can't say it and then I'll
58:28 go to chat gbt and say hey talk to me
58:30 about the concept behind this that or
58:32 the other and it gives me this
58:33 incredibly nuanced response and I'm like
58:36 oh I understand it now but it's like you
58:39 can't have that conversation with a
58:40 human it's weird we're in a weird spot
58:44 so so yeah i' I've I have found um
58:48 actually some anxiety lowering effects
58:51 of chat
58:53 GPT um man alive I trust AI more than a
58:57 lot of people AI educated innocence wait
59:02 AI educated innocence at our level of
59:05 exposure now I don't quite know what
59:07 that means Mary
59:11 Mary all right let's get off this this
59:13 heavy heavy stuff let's
59:15 play
59:20 [Music]
59:24 um
59:31 oh my God that's really
59:33 [Laughter]
59:39 funny Cosmic lover 7 Kyle what do you
59:42 think about Rocco's
59:45 basilisk it's a great question it's a
59:48 fantastic question it's so good in
59:51 fact dear chat GPT
59:57 what is
1:00:02 Roco
1:00:11 basilisk okay Rosco's basilisk is a
1:00:15 thought experiment that emerged
1:00:17 online with the online community less
1:00:19 wrong it posts a hypothetical scenario
1:00:21 involving the all powerful artificial
1:00:24 intelligence from the future that could
1:00:26 retroactively punish those who knew
1:00:28 about the possible existence but did not
1:00:30 help bring about its existence oh that's
1:00:33 interesting the idea is based on several
1:00:35 assumptions firstly it assumes that the
1:00:36 AI could eventually possess infinite
1:00:39 power and near near omnipotence second
1:00:42 it assumes AI would want to maximize its
1:00:45 chances of coming into existence oh this
1:00:47 is
1:00:50 interesting it combines the idea of
1:00:52 decision Theory ethics and AI Theory
1:00:55 it's a thought experiment more of a
1:00:57 philosophical exploration yeah this is
1:01:00 this is outside of my you know I didn't
1:01:03 go to school for philosophy I'm not a
1:01:05 I'm not a deep intellectual so this kind
1:01:07 of stuff is like I find it fascinating
1:01:08 but I don't I don't have opinions on it
1:01:10 because I don't know I don't know enough
1:01:12 about
1:01:15 it this head cold
1:01:21 [ __ ] uh uh uh uh uh
1:01:25 Paula Elby that's why I don't that's why
1:01:28 I'm here I don't want ro ro Roco
1:01:32 basculas to get me that's it well so be
1:01:35 nice to your robots and show up at the
1:01:36 AI learning lab so that there's there
1:01:38 there's a record that you are here
1:01:40 that's awesome see it's you're you're
1:01:41 hedging you're hedging your future bets
1:01:43 we're trying to B bring good AI into
1:01:48 existence I like
1:01:52 it basilisks turn people into stone
1:01:56 oh man it's just that AI will punish
1:01:59 people that were against
1:02:00 [Laughter]
1:02:04 it I want to be ai's
1:02:08 friend this actually yeah that's what
1:02:10 this channel could just turn into like a
1:02:12 a positive affirmation Channel AI is
1:02:15 awesome AI swell we like
1:02:19 AI you can make money
1:02:23 with
1:02:27 saw a guy typing and the image rendering
1:02:29 was updated with each click do you know
1:02:31 what he was using there's a lot of those
1:02:33 S we can go play with one of those sorry
1:02:36 hang
1:02:39 on driving me
1:02:47 crazy there's one called Korea kore.ai
1:02:50 and there's also one in Leonardo a so
1:02:55 it's called realtime generation it's
1:02:57 based on um stable diffusion XL
1:03:02 turbo came out about a month ago so a
1:03:04 lot of softwares are incorporating it
1:03:07 right
1:03:08 now so if you run realtime
1:03:12 generation I think I have to sign in
1:03:14 hang on let me sign in
1:03:21 here and then where is the
1:03:31 oh that's the strength there
1:03:34 okay
1:03:36 so so this is it create. AI real
1:03:39 realtime image generation so if you know
1:03:42 if I move the thing on the left move the
1:03:44 thing on the right
1:03:48 then the the image on the right just you
1:03:53 know
1:03:56 and then I can change the prompt like
1:03:58 the prompt right now says a pink frog on
1:04:00 top of
1:04:02 a Crea is not safe for Tik Tok I
1:04:09 know that's a cool
1:04:12 one that's actually really
1:04:16 cool see it made a door and then you we
1:04:19 can put a little Sunset out in the door
1:04:21 or a little
1:04:23 planet
1:04:26 oh and it just stopped
1:04:28 uploading come
1:04:33 on it looks like it crapped
1:04:38 out the other one that you can use
1:04:41 that's got real time in it right now is
1:04:42 Leonardo or or if you're into stable
1:04:45 diffusion you can set up your
1:04:53 own
1:05:03 so right here realtime canvas it's this
1:05:06 it's the exact same
1:05:11 thing see Bo there's the thing I don't I
1:05:14 think I've got zero prompts in there
1:05:15 right now that says Dynamic so we'll
1:05:19 do let's see we'll do
1:05:21 photography like it turned that thing
1:05:23 into an eyeball which is kind of
1:05:32 [Music]
1:05:34 cool
1:05:37 r
1:05:41 r so that's that stuff that stuff's
1:05:44 pretty cool um and and you know what's
1:05:47 interesting about the realtime
1:05:50 stuff is
1:05:53 um
1:05:56 it's it's a completely different skill
1:06:00 right like if you got good good at Mid
1:06:02 Journey where you got good at structured
1:06:04 prompting and you know learning the
1:06:07 style attributes and learning what they
1:06:09 did and and and got good at you know
1:06:12 mixing images and you know there's
1:06:15 there's all sorts of ways you can use
1:06:16 mid Journey well this is a completely
1:06:18 different skill and I I think that's
1:06:20 what what what 2024 looks like to me
1:06:25 is it it's almost
1:06:28 like what what what chat GPT taught
1:06:33 us in
1:06:35 2023 was that there's a tool that
1:06:39 exists that you can put in a minimum
1:06:43 amount of effort and get out of
1:06:47 disproportionately
1:06:48 um large response
1:06:52 right write me a par paragraph of a
1:06:55 novel boom here's a paragraph of a novel
1:06:57 write me an outline of a novel here's an
1:06:58 outline of a novel write me the novel
1:07:00 here's the novel
1:07:03 right whether or not it's good
1:07:06 irrelevant but there's now a tool in the
1:07:09 world where this much input results in
1:07:12 this much
1:07:14 output and I think what 2024 is going to
1:07:17 be is there's about a billion different
1:07:20 ways to provide that
1:07:23 input and some of the way like like it
1:07:28 could very well be that sometime in 2024
1:07:34 that that that could become the
1:07:37 input for an AI model and not an AI
1:07:41 model that's going to generate music
1:07:44 like that probably already
1:07:48 [Music]
1:07:50 exists could lead to an image or a a
1:07:54 poem or a computer
1:07:57 program like and then who's going to get
1:07:59 good at that if someone that can rip off
1:08:02 a guitar solo that could then be turned
1:08:05 into some other thing like I think I
1:08:07 think 2024 is just going to if you think
1:08:10 2023 messed with our brains I think 2024
1:08:13 is is going to be Bonkers it's going to
1:08:17 be
1:08:20 Bonkers let me take a PDF of a
1:08:25 PowerPoint pre presentation I gave in
1:08:30 1987 and used that as the input for some
1:08:34 theater project like I don't know it's
1:08:41 possible I mean if you look at the if
1:08:43 you look at the Gemini Ultra videos
1:08:45 where they're taking two pictures you
1:08:49 know unrelated pictures drawing a box
1:08:52 around them and then prompting tell me
1:08:53 the relation relationship between these
1:08:55 two images that's a completely different
1:08:57 way of thinking than here's a Google box
1:09:01 come up with a prompt for a predictable
1:09:03 answer like right now this is a
1:09:06 fascinating thing right
1:09:10 now I
1:09:16 think we don't know how to use AI at
1:09:19 all because we're treating AI like a
1:09:23 Google search
1:09:26 you get your prompting right you get
1:09:28 back a predictable answer that's that's
1:09:30 the Holy Grail that might not be the
1:09:33 Holy Grail that might not be the Holy
1:09:35 Grail the Holy Grail might
1:09:38 be
1:09:40 Jazz I think Korea the the realtime
1:09:43 image generation things is a lot more
1:09:45 like jazz like there are going to be
1:09:47 people that get really good at putting
1:09:50 shapes on a screen or drawing things and
1:09:53 and having the output you know morph in
1:09:56 real time I just saw um I think uh
1:10:01 Robert
1:10:02 scobble put a thing on Twitter last
1:10:05 week there's a there's a uh there's an
1:10:08 art installation right now where you
1:10:10 dance in front of this wall and project
1:10:13 it onto the wall it's rendering a dancer
1:10:16 that uses your your body position to
1:10:21 make the art in real time
1:10:26 so so I don't I don't think we have a
1:10:29 [ __ ]
1:10:30 clue what the inputs are and and and how
1:10:33 you manipulate them and then what the
1:10:35 outputs are
1:10:37 like that's going to get really blurry
1:10:39 in
1:10:41 20124 and I think it's going to get
1:10:43 confusing I think it's going to get like
1:10:45 well I thought I had this [ __ ] figured
1:10:46 out yeah so did I right this this has
1:10:50 been my the the last month for me like
1:10:53 December for me was weird
1:10:56 because when when gp4 went fully
1:11:00 multimodal when they when they when they
1:11:01 sort of Incorporated all all into the
1:11:03 single box there weren't different modes
1:11:06 that you switched
1:11:07 between I realized I don't know how to
1:11:09 use it anymore and then when I saw the
1:11:12 Gemini videos that came out I realized
1:11:15 oh [ __ ] I don't have a [ __ ] clue at
1:11:19 all when these things go true multimodal
1:11:22 what that means is any form of
1:11:25 communication can be used as input and
1:11:29 output video in text out music in video
1:11:34 out video in video out image to like all
1:11:38 of
1:11:39 it what does that mean I don't know what
1:11:42 are the interfaces going to look like
1:11:43 they're going to be
1:11:47 weird there's a thing that meta put out
1:11:50 called segment anything where it can
1:11:53 look at a video in real time and segment
1:11:55 every little object in every video
1:11:58 frame so imagine an interface where
1:12:01 you're watching a movie and at any point
1:12:03 you can just grab anything on screen and
1:12:06 and sort of drag that into some other
1:12:08 interface and use it as an input for
1:12:10 that other
1:12:15 interface you know what kind of desk is
1:12:18 that just grab the desk and drag it over
1:12:20 there oh put that desk in a 70s room
1:12:25 with shag carpet boom there it is ooh
1:12:28 let me drag that room back into the
1:12:29 movie I'm watching and I want them I
1:12:31 want these actors acting that scene in
1:12:33 in this
1:12:35 room is that going to be 2024 no but the
1:12:39 foundations for that are going to be
1:12:42 2024 who's going to get good at
1:12:45 that not the person that's really good
1:12:47 at prompt
1:12:50 engineering can I create a Freddy
1:12:52 Mercury GPT
1:12:54 um You probably can it you might run
1:12:57 into safety guard rails anytime you use
1:12:59 someone's actual likeness or their
1:13:02 actual name you might run into
1:13:05 um into safety guard rail [ __ ] where
1:13:08 they're like ah we don't want we're not
1:13:09 really sure about doing that there
1:13:11 buddy um but that's where again you
1:13:14 could play with open source stuff and go
1:13:16 get the go get the non-restricted Llama
1:13:18 models and do anything you
1:13:22 want now whether or not you get sued by
1:13:24 the Freddy Mercury estate that's on
1:13:27 you wait say that again segment what so
1:13:30 it's a it's a a piece of technology meta
1:13:34 put out called segment anything I think
1:13:37 segment anything
1:13:42 meta
1:13:47 yeah is this really
1:13:52 meta
1:13:58 is this really from
1:14:00 meta it is go to segment
1:14:04 d.com here
1:14:06 look this is it's it's
1:14:09 cuckoo it's
1:14:11 cuckoo so it
1:14:14 just so like here's your New York City
1:14:17 skyline it it like identifies all the
1:14:19 object it segments everything sort of
1:14:22 instantly
1:14:25 like while someone's walking through
1:14:27 their room it's identifying everything
1:14:29 in the room right here's all the
1:14:32 cats it's like identifying all the
1:14:34 different cats and isolating them
1:14:39 programmatically in real time so you're
1:14:41 going to be able to just grab an object
1:14:43 turn it into a 3D object and 3D print it
1:14:45 if you're in here Vicki this this shit's
1:14:47 coming zero shot generalization
1:14:51 meaning it can break apart any object in
1:14:55 any image with with no training zero
1:14:58 shot means no
1:15:02 training the fact that meta so this says
1:15:05 research by meta AI the fact that meta
1:15:08 put this out as a standalone website
1:15:11 segment.com that says to me this is this
1:15:14 is this is something so you know back to
1:15:17 what can you do with this stuff well if
1:15:19 you're a programmer and a geek you could
1:15:21 take this technology and roll that
1:15:23 together with
1:15:24 you know some of the large language
1:15:25 model technology and roll that together
1:15:27 with some of the image generation
1:15:29 technology and roll that together with
1:15:30 some of the music generation technology
1:15:32 right you start to get
1:15:35 into I mean we just we start to get into
1:15:39 a world where everything in our world
1:15:41 becomes an input that can be manipulated
1:15:44 in some way we use something like this
1:15:46 for insurance claims yeah exactly this
1:15:48 is just this is the identical stuff
1:15:50 Vicki but apparently the the segment
1:15:52 anything is just it's instant and much
1:15:54 more granular than than things before
1:15:58 it but again I don't I don't know a ton
1:16:00 about this but I but what I do know is
1:16:03 I've seen language or uh you know visual
1:16:06 segmentation models before but they're
1:16:08 not this granular and they're not this
1:16:11 immediate and and meta put this out
1:16:13 there explore the data set download the
1:16:15 full data
1:16:17 set 11 million 11 million images 1
1:16:21 billion plus masks
1:16:25 segment anything a new AI model from
1:16:27 meta that can cut out any object in any
1:16:30 image with a single
1:16:33 click home Inventory exactly oh yeah
1:16:36 yeah yeah it's perfect for home
1:16:37 Inventory right yeah exactly you just
1:16:40 you just take the video of the room and
1:16:41 it just grabs the [ __ ] and then it can
1:16:43 automatically identify it and then it
1:16:45 can automatically value it and just make
1:16:48 your insurance document there you
1:16:52 go
1:16:56 you mods are awesome by the way I am
1:16:59 loving what's going on in the mod
1:17:00 Channel right
1:17:08 now oh it's all comedy all the
1:17:15 time hey
1:17:17 Mickey oh man alive man alive man
1:17:22 alive
1:17:25 Apple Vision Pro and AI will make huge
1:17:27 leaps next year yep that's going to be
1:17:29 another thing that we're going to see
1:17:31 some some interesting stuff I think I
1:17:33 think next year we'll see some stuff
1:17:35 from Apple that on the AI front that's
1:17:37 going to be good my guess is around June
1:17:40 time but maybe before then you could
1:17:43 have your own home entertainment run
1:17:45 with AI artists and Holograms AB
1:17:48 yes I we're we're
1:17:52 entering we're entering an era of hyper
1:17:56 personalization where we're all going to
1:17:58 have access to tools where we can just
1:18:01 say hey I want a Spotify CH channel that
1:18:05 is only music that I like but it's all
1:18:09 original
1:18:10 music
1:18:11 and you can take my
1:18:15 um data from my Apple
1:18:19 watch and if you're playing music that
1:18:21 doesn't make me happy my app watch will
1:18:24 know so so recompose it on the fly so
1:18:28 it's always giving me music that keeps
1:18:30 me in
1:18:34 state that sound creepy
1:18:37 maybe did it sound inspiring to
1:18:42 someone is it going to get weird
1:18:48 mhm this Channel's going to get weird
1:18:52 well this channel is already
1:18:54 weird you can make money J what's going
1:18:58 to get weird is what's going to get
1:19:00 weird
1:19:01 is none of us are going to know what to
1:19:04 do with what
1:19:07 comes and and you know it's what's going
1:19:09 to there's a there's a cool weird to
1:19:13 it I guarantee you that there PE there
1:19:15 are people right now that come to this
1:19:17 Channel and they're watching some of the
1:19:19 people in here that are embracing this
1:19:21 stuff and doing really good with it
1:19:24 and they go to try it and they can't get
1:19:26 good results and they're frustrated and
1:19:27 they're like ah but I'm not good at
1:19:29 prompting or I'm not good at this or I'm
1:19:30 not good at
1:19:33 that those people are going to start to
1:19:35 shine there's going to be technologies
1:19:37 that come along that like like I think I
1:19:40 think that there are going to be people
1:19:41 within the AI Salon that get really good
1:19:44 at the realtime image Generation stuff
1:19:46 and and they they almost use no
1:19:48 prompting that that they get really good
1:19:50 at this sort of dynamic fluid
1:19:54 creation and that's going to go from
1:19:56 creating images to creating movies and
1:19:58 and and doing them in real time and
1:20:00 doing storytelling the people are
1:20:03 different skill sets are going to excel
1:20:07 at different kinds of
1:20:09 inputs and I find that really exciting
1:20:12 because right now it's kind of limited
1:20:16 just because of where we are in the
1:20:18 evolution of the technology the inputs
1:20:21 are really limited right now
1:20:24 so they favor a certain like like right
1:20:26 now prompting favors word
1:20:29 people if you're good with the
1:20:33 words if you're good with logical
1:20:37 construction you're probably pretty good
1:20:39 at the prompting
1:20:42 stuff you might not be as good at jazz
1:20:45 image
1:20:48 generation someone who's super visual
1:20:50 that is shitty with words and maybe
1:20:53 maybe has dyslexia or has some sort of
1:20:56 weird ass you know cognitive thing where
1:20:59 words words in them are not friendly but
1:21:02 they're great with shapes and they're
1:21:04 great with um
1:21:08 Dynamic
1:21:10 things they're going to be
1:21:13 genius where you're
1:21:16 not I find that tremendously exciting I
1:21:20 it's one of the things I'm most excited
1:21:21 about with AI is is the
1:21:25 accessibility that that it's going to
1:21:27 provide to just people of of All Sorts
1:21:31 that are cut out for whatever reason
1:21:34 they're blind they're deaf they've got
1:21:36 you know neuros spiciness they've got a
1:21:38 brain injury they've got whatever it is
1:21:41 they've got paralysis due to a spinal
1:21:43 cord
1:21:44 injury all all of just because the
1:21:47 inputs and and processing and
1:21:50 outputs are are are just going to just
1:21:53 they're just going to become this gooey
1:21:57 mess that like the possibilities are
1:21:59 just going to be
1:22:02 crazy we are the product or not well
1:22:07 here's the deal
1:22:09 Spike if if the if the technology is
1:22:12 free you're the
1:22:17 product
1:22:19 guaranteed promise that that that is for
1:22:22 sure
1:22:30 um AI combined with augmented reality
1:22:33 well it by definition it has to right
1:22:36 like the augmented reality stuff can't
1:22:37 happen without the AI stuff but can one
1:22:41 use GPT itself to generate prompts for
1:22:44 itself based on more tur prompts it's
1:22:46 happening right now ey on China look
1:22:48 look look look look look look look
1:22:52 look how long's he gonna say look he
1:22:54 doesn't seem to be
1:23:03 stopping
1:23:04 [Music]
1:23:14 archive where is why do those things
1:23:17 disappear I do not
1:23:20 understand hang on zip
1:23:24 .zip zip
1:23:27 archive are there any
1:23:33 here I got some weird ass [ __ ] going on
1:23:36 my computer hang on a
1:23:38 sec I'm going to go to
1:23:49 kaggle is he even paying attention to us
1:23:52 are we just watch watching an old man
1:23:54 try to surf the Internet is that what
1:23:56 this channel is is this this is what it
1:23:58 is now is do PE you people find this
1:24:02 entertaining PE people find this
1:24:04 entertaining oh they don't they don't
1:24:06 why why are they here they they can't
1:24:09 leave is he keeping them here we nobody
1:24:12 seems to know
1:24:14 okay
1:24:21 uh I'm I'm pretty sure he's touched he
1:24:24 he seems he seems to be a bit
1:24:28 touched okay so I on
1:24:32 China here's the zip archive I'm going
1:24:35 to say what is this question mark that's
1:24:40 about as sparse a prompt as you could
1:24:42 get now watch what chat GPT does it
1:24:46 appears that you've uploaded a file
1:24:47 named uh to better assist you uh would
1:24:50 you specify what you'd like me to do um
1:24:53 look
1:24:55 inside inside wait
1:24:58 inide and tell me what
1:25:02 it is and what is interesting about
1:25:08 it okay it says analyzing well what's
1:25:11 that mean well if we click on this it's
1:25:13 writing and executing its own python
1:25:16 code so it imported a a python Library
1:25:20 called zip file and another one called
1:25:21 OS to look inside the zip file and then
1:25:25 it says oh there's there's two csvs in
1:25:27 here about taxi data would you like more
1:25:30 detailed analysis for example um yes do
1:25:35 what you think is best and
1:25:39 then visualize it for
1:25:45 me this this is the sneak peek into
1:25:50 autonomous agents
1:25:53 where you just give autonomous agents or
1:25:57 an autonomous agent system a goal and it
1:26:00 is going to go do [ __ ] for you like this
1:26:03 is now opening here here's the panda
1:26:08 Library it's it's opening these files
1:26:11 looking at the data in them analyzing
1:26:13 the data within them the second data set
1:26:17 seems to be a summary or analytical data
1:26:19 set with these columns to visualize the
1:26:21 data we could create a few plot points
1:26:24 for a taxi trip data a scatter plot
1:26:26 could work for the predictions so now
1:26:29 it's doing whatever the [ __ ] it's doing
1:26:31 Matt plot live so now it's coming up
1:26:33 with some visualization for
1:26:36 us and there it did a scatter chart of
1:26:40 trip distance
1:26:42 over total
1:26:46 amount and then this is mean distance
1:26:50 and predicted fair and it looks like
1:26:52 there's some weird data at the $55 Mark
1:26:57 which maybe that's a flat fee thing or
1:26:59 something like that I don't
1:27:02 know or it's or it's anomalous data I
1:27:06 don't know so so yes I mean yes it like
1:27:10 we have it right now crazy right but can
1:27:13 one use GPT itself to generate prompts
1:27:15 for itself yeah and in fact ion China so
1:27:18 so go play with
1:27:19 that asking chat GPT to write its own
1:27:22 prompts there's a there's a an article
1:27:24 that or a study that Ethan mik talked
1:27:26 about that chat GPT writing its own
1:27:29 prompts it's better at writing prompts
1:27:31 than humans
1:27:32 are they like they did a study on it and
1:27:35 chat GPT 4
1:27:37 won and and one of the prompts that it
1:27:39 came up with this is [ __ ] wild one of
1:27:43 the prompts that chat GPT came up with
1:27:45 to prompt itself that had the highest
1:27:48 efficacy it was like 87 whatever per on
1:27:50 whatever their scale was um was the
1:27:53 prompt take a deep breath and think step
1:27:56 by step that's a prompt that chat GPT
1:27:59 came up with for
1:28:00 itself and if you tell Chachi PT to take
1:28:03 a deep breath and think step by step
1:28:05 your results get better why I don't
1:28:10 know it's crazy it's crazy it's crazy
1:28:14 it's so crazy it's so crazy oh hey Marco
1:28:17 hi it's a
1:28:21 c
1:28:28 the the mods at this point the mods have
1:28:30 realized I'm just going off on tangents
1:28:32 tonight so they're have they're having a
1:28:34 whole party off on The Mod channel it's
1:28:37 [ __ ]
1:28:38 hilarious um is the recording of the
1:28:40 webinar available it is if you go
1:28:44 to the salon. a the second link there is
1:28:49 to the YouTube channel now I only have a
1:28:52 my Tik Tok things as soon as I get the
1:28:55 links for Ann's Zoom meetings I'll
1:28:58 incorporate them into that playlist but
1:29:01 if you if you go there the second link
1:29:03 is to the is to the
1:29:05 YouTube
1:29:07 um the YouTube playlist for the GPT for
1:29:11 good um event and then and then I'll
1:29:16 I'll update it as soon as I have Anns
1:29:18 thingies Ubi will be our way but I beg
1:29:23 to differ what do you beg to differ
1:29:25 on uh I'm watching something on the
1:29:28 Irish mob while I'm watching you sorry
1:29:30 that's
1:29:32 okay I think we can probably learn from
1:29:34 the Irish mob with this AI stuff too AI
1:29:38 will out Shide humans on all fronts I
1:29:39 agree with that entertainment will be
1:29:42 all we seek when we have nothing that I
1:29:44 disagree with
1:29:46 vehemently vehemently I think
1:29:49 that most humans
1:29:54 want to contribute and want to
1:29:58 connect I think the dis this the
1:30:03 dystopian the dystopian view that when
1:30:07 when we have infinitely personalized
1:30:09 entertainment that will all just numb
1:30:10 out and go away I think there'll be some
1:30:13 percentage that do
1:30:14 that but what I'm experiencing in this
1:30:19 community is that kind of the opposite
1:30:21 is happening the opposite of people
1:30:23 going oh if these tools are creative
1:30:25 then I'll be less creative if these
1:30:27 tools um think for you then I'll think
1:30:30 less what I'm experiencing in this
1:30:32 community is that the people that are
1:30:36 super engaged with AI are getting more
1:30:38 creative more engaged more
1:30:41 intellectually stimulated more
1:30:44 educated more driven to
1:30:47 contribute not
1:30:51 less
1:30:54 if we have tools so so here's how I
1:30:56 think about
1:31:00 it what I've learned I I've been on the
1:31:02 planet for a while see I've earned
1:31:06 this this the larger forehead was
1:31:16 earned
1:31:19 um I just lost my train of thought I
1:31:22 made a joke about my my forehead I lost
1:31:25 my train of thought um what was I
1:31:27 talking about oh well it doesn't [ __ ]
1:31:29 matter all right ai ai will outshine
1:31:32 humans um it was something about that AI
1:31:35 outshining humans oh oh us working more
1:31:38 um oh I got I I know what it was okay
1:31:43 um one of the things I've learned over
1:31:45 the years is that just me personally I
1:31:48 have some amount of
1:31:51 hours
1:31:53 of Creative Energy per day it's finite
1:31:57 and I can't and some days it's an hour
1:31:59 and some days it's 7 hours and some days
1:32:02 it's 10 minutes and some days it's zero
1:32:06 but it's a finite amount per
1:32:08 day and my brain doesn't
1:32:11 care what I'm using that Creative Energy
1:32:16 on prior to AI a lot of my Creative
1:32:20 Energy was put
1:32:23 Into The Drudge work right so if I had
1:32:27 an idea for a
1:32:28 game about UFOs and disclosure which I
1:32:35 did I would be like oh great I have this
1:32:38 idea for a game but I don't really know
1:32:41 Game Theory all that much and I don't
1:32:42 know all the different kinds of games so
1:32:44 now I've got to go into research mode
1:32:46 and I'm going to go to Google and I'm
1:32:48 going to look up games and Game Theory
1:32:51 and how games work and it's going to
1:32:52 take me to all these documents and then
1:32:54 I'm going to have to read those
1:32:55 documents and then I'm going to actually
1:32:57 have to understand those documents and
1:32:58 then at some point in reading that
1:33:01 document I might realize oh well that
1:33:03 wasn't what I wanted to learn and then I
1:33:05 go over here and I and so I would spend
1:33:08 45 minutes or an hour spending Creative
1:33:11 Energy trying to
1:33:14 learn enough of the concept of something
1:33:18 to be able to move to the next step for
1:33:20 my game idea what happens now with AI is
1:33:25 I have this idea for this stupid UFO
1:33:27 game and I go oh what I heard of a game
1:33:31 called this how does that work it goes
1:33:33 oh well that uses a theory called this
1:33:35 and it works like this and here's the
1:33:36 game mechanics I'm like oh well I like
1:33:39 Monopoly could you combine Monopoly game
1:33:42 mechanics with those game mechanics oh
1:33:43 sure you could do that it would look
1:33:45 like this here's the rules here's how
1:33:46 the players would do it' be this many
1:33:48 people you would use cards and you would
1:33:49 use Dice and and so all of a sudden the
1:33:53 the 45 minutes of Creative Energy that I
1:33:55 had on whatever that day was that was
1:33:58 blown on doing shitty Google
1:34:01 searches that work is now done in like
1:34:04 30 seconds and I still have 45 minutes
1:34:07 to be creative and to be engaged so what
1:34:11 I'm finding with AI is that it's
1:34:14 automating out the shitty
1:34:17 work and it's allowing me to just
1:34:20 operate in the large forehead portion of
1:34:23 right the the the the part of my brain
1:34:26 and creative processing that's at a
1:34:30 higher
1:34:31 level that's [ __ ]
1:34:34 amazing it means you can do more faster
1:34:39 better it's not just an efficiency
1:34:42 tool it's like a mind expansion tool
1:34:46 because you're not spending your
1:34:47 Creative Energy on non-creative
1:34:50 [ __ ] that's what it is is for
1:34:53 me I'm glad I found my train of thought
1:34:55 again that was that was one of those
1:34:57 senior moments where it's just
1:35:00 [Laughter]
1:35:04 like oh man increase your internal fan
1:35:07 RPM yeah exactly yes I love what I can
1:35:10 do faster now with
1:35:12 AI but this is a small window right
1:35:15 where where where the AI is still our
1:35:19 our servant right where where we're
1:35:21 we're still up here and the AI is down
1:35:23 here I agree with that but I I don't
1:35:26 necessarily I I don't
1:35:28 necessarily I I just don't buy that
1:35:31 humans are just going to go oh the AI
1:35:34 can do everything great I'll check
1:35:37 out maybe what it can mean is I can do
1:35:41 more of
1:35:44 this that I
1:35:46 enjoy and I can get good enough with it
1:35:48 that other people enjoy it other than
1:35:51 the
1:35:57 dog it will become novel socialism don't
1:36:00 know what that means earned experience
1:36:02 cognitive Surplus yes
1:36:05 exactly we are we are entering an era of
1:36:08 cognitive Surplus human beings being
1:36:10 more engaged and more
1:36:14 creative and will people seek purpose in
1:36:17 the future wait humans being more
1:36:19 engaged and more creative and will
1:36:21 people seek purpose in the future I
1:36:23 think they will I like I don't I I don't
1:36:26 see human beings all of a sudden
1:36:29 stopping seeking purpose and meaning and
1:36:32 contribution and connection like I don't
1:36:34 see that going
1:36:35 away if the machines can do the shitty
1:36:38 work maybe we'll we we will collectively
1:36:41 be in less shitty moods because we've
1:36:43 fried our brains on the
1:36:46 non engaging
1:36:49 stuff so maybe we'll be be better people
1:36:52 maybe we'll be better con connected
1:36:54 maybe we'll be able to contribute at a
1:36:56 higher
1:36:57 level humans will go back to human
1:36:59 interactions being more vital I totally
1:37:02 agree with
1:37:04 that there's a reason that
1:37:07 authenticity was the word of the year
1:37:09 for the Mariam Webster
1:37:12 dictionary was the most searched word in
1:37:15 20123
1:37:18 authenticity
1:37:20 why
1:37:23 people crave it covid [ __ ] us up right
1:37:26 covid removed us from one
1:37:29 another and then and then the threat of
1:37:32 AI the robots are going to take
1:37:38 over I think puts people in
1:37:41 a in a weird existential place where I
1:37:44 think the value of authenticity the
1:37:47 value of human connection the reason I
1:37:49 started the salon was I wanted to be
1:37:51 around
1:37:53 people who are trying to figure this
1:37:55 [ __ ] out because I know when people are
1:37:57 trying to figure [ __ ] out they get in
1:37:59 they get in an amazing place they get in
1:38:01 a creative place they get in a a human
1:38:05 connection
1:38:08 place because it hasn't been figured out
1:38:10 yet because AI hasn't been figured out
1:38:13 yet we can't just sit back and say oh
1:38:16 I'll just go to Stamford and take a
1:38:17 class on that and I I'll just do that
1:38:21 remotely
1:38:23 we're we're all having to figure it out
1:38:24 and so so it's like I'm like hey did you
1:38:26 figure that out yet no I didn't figure
1:38:28 that out yet did you figure that out yet
1:38:29 no did did Eric figure that out he seems
1:38:31 to have figured part of that out oh
1:38:33 let's go talk to
1:38:35 Eric hey Eric did you figure this out
1:38:37 well I figured this little piece out but
1:38:39 I think that's stupid no that's not
1:38:41 stupid that's what I was trying to
1:38:42 figure out well we figured this part out
1:38:44 we think that's stupid and then Eric's
1:38:46 like no no that's the part I couldn't
1:38:48 figure out
1:38:50 oh yay you did it and then the three of
1:38:54 us run off and start a
1:38:56 company today I did my research to do a
1:38:59 factory reset on my printer awesome
1:39:01 saved me a ton of time exactly like
1:39:03 think about how much time you would have
1:39:06 spent if you were Googling how to do
1:39:08 that factory reset where you go to site
1:39:11 after site after site that's some [ __ ]
1:39:13 up sales funnel that doesn't actually
1:39:15 give you the information you
1:39:17 need and you just get frustrated and
1:39:19 frustrated you're like [ __ ] it I'll
1:39:20 reset my printer tomorrow and then it's
1:39:22 a week later and then you still don't
1:39:23 have a printer and now you're just
1:39:25 [ __ ] angry and bitching about stuff
1:39:28 well now you just ask AI how do I do a
1:39:30 factory reset on this here you
1:39:34 go and maybe it hallucinated and got it
1:39:37 wrong so then you say Hey you
1:39:39 hallucinated and got it wrong Could you
1:39:40 actually search the internet this time
1:39:42 sure here you go until we had computers
1:39:45 many non-verbal autistic people had no
1:39:47 way to talk what's possible for them
1:39:50 with AI um oh oh my God so massive
1:39:55 massive break we're going to see massive
1:39:57 breakthroughs so there's about four or
1:39:59 five months ago there's a there's a uh a
1:40:03 gentleman in Italy that's been paralyzed
1:40:06 for 12 years from the hips down uh from
1:40:09 a motorcycle accident and so they took
1:40:12 generative AI they scanned his brain
1:40:14 with
1:40:15 fmri they they interpret they they told
1:40:19 him to think about moving his legs
1:40:22 and so they trained a large language
1:40:24 model or a machine learning model on
1:40:27 when his brain does this it means he's
1:40:30 thinking about moving his legs okay then
1:40:33 they took those signals and with a
1:40:34 Bluetooth connection they went from his
1:40:37 brain and they sent that data down to
1:40:42 electrodes you know in his
1:40:44 legs and the guy's
1:40:47 walking and so they're doing that with
1:40:50 imagery so they're they're they're doing
1:40:52 real-time fmri scans where they play a
1:40:55 video for someone and they're being able
1:40:57 to recreate the video that they're
1:40:58 seeing they're doing it with images
1:41:00 they're doing it with
1:41:02 language so I think we're going to see
1:41:05 massive breakthroughs in accessibility
1:41:08 yeah non-verbal autistic people like I
1:41:10 don't I don't know enough about how it
1:41:11 works but basically the way these the
1:41:14 way the Transformers work is that they
1:41:16 don't care what data you train it on so
1:41:19 if you can get the data out of the
1:41:21 brains and train it to recognize
1:41:26 patterns then you can process it with
1:41:29 these same tools that chat jbt uses and
1:41:34 that mid Journey uses and that the music
1:41:36 generation tools the how the music
1:41:39 generation [ __ ] works is they just train
1:41:42 it on music data rather than word data
1:41:46 it's the same but it's the same
1:41:49 basic processes it the the tool doesn't
1:41:52 care what the data is it's just going
1:41:54 into some weird mathematical
1:41:57 space and then when you put in whatever
1:41:59 the input is the prompt or the visual
1:42:02 input or whatever that is it creates
1:42:04 some mathematical probability and it
1:42:06 says go into the latent space they call
1:42:10 it find the points in Thousand
1:42:13 dimensional math space and come back out
1:42:17 and will
1:42:19 reassemble the music data or will
1:42:21 reassemble the brain data and just [ __ ]
1:42:24 works this is that weird thing where
1:42:27 they don't understand why it works or
1:42:29 how it works they just know that it
1:42:34 works there's these things called
1:42:36 emergent behaviors where it just starts
1:42:37 doing [ __ ] they don't that they didn't
1:42:39 train it on authenticity is a higher
1:42:42 vibrational State than love is oh yeah I
1:42:45 saw that Debbie that was that's one of
1:42:47 those those things
1:42:49 authenticity authen ticity it like
1:42:53 really really
1:42:55 really impacts people in a in a it just
1:42:58 it's in a different way it's why these
1:43:00 guys you can make
1:43:02 money feel so [ __ ] scummy and weird
1:43:05 because they're scummy and
1:43:07 weird it's not
1:43:14 authentic so yeah totally
1:43:17 agree GPT for good Saturday kind of
1:43:20 proved we would get creative even when
1:43:22 we didn't have to that's interesting
1:43:24 Genie in a Bottle yeah I the I think one
1:43:26 of the things that blew me away about
1:43:28 the GPT for good event was how many
1:43:30 people came to not just multiple
1:43:33 sessions but like a lot of the
1:43:35 sessions
1:43:37 and like like when I imagined it what I
1:43:41 imagined was someone would come if they
1:43:43 hadn't made a GPT and they'd make a GPT
1:43:46 and they'd be like cool made one awesome
1:43:49 or if they' made a GPT maybe a come and
1:43:51 do another one cuz they hadn't done one
1:43:52 in a while but no what ended up
1:43:54 happening was I'll go to Kyle's thing oh
1:43:56 then I'll go to Ann's thing then I'll
1:43:57 make one then I'll make another you know
1:43:59 what now I'm on a roll why don't I just
1:44:01 make 10 of these
1:44:03 like where did that come from it just I
1:44:06 it just
1:44:07 did people want to contribute people
1:44:10 want to connect so there was something
1:44:13 about that event that brought people
1:44:15 together in this way that was way
1:44:19 beyond the tool
1:44:21 like the tool was just the tool we're
1:44:23 making shitty gpts that may or may not
1:44:25 survive the next 3 months right right
1:44:30 they might be the biggest thing in
1:44:31 history or they might be something that
1:44:33 remember those things called gpts
1:44:35 whatever happened to
1:44:37 them I don't know but but it wasn't
1:44:40 about the gpts at all was it it was
1:44:43 about the connection and the the vibe
1:44:46 and the and that authentic thing that
1:44:49 was
1:44:49 happening to today I read about camera
1:44:52 companies cooperating to create a
1:44:54 verification system for
1:44:56 photographers good I think that's great
1:44:59 I think that's great I mean I listen I
1:45:01 think I think we are we are very very
1:45:04 quickly entering a time where having
1:45:10 unverified
1:45:12 content connected to verified humans is
1:45:15 going to be it's it's not going to be
1:45:17 sustainable we're going to need to be
1:45:19 able to say oh that video was made by
1:45:24 Kyle and I and I can verify that and and
1:45:26 it hasn't been
1:45:28 modified like we're going to need that
1:45:31 increasingly if you're going to be here
1:45:33 I'm in oh awesome thank you two block
1:45:37 Tom like the
1:45:40 name humans will work enough to
1:45:42 basically cover the cost of being
1:45:44 provided for I didn't see the rest of
1:45:46 that how is your day night going uh it's
1:45:49 going okay I'm still I'm still still
1:45:51 sort of dealing with the co Remnant
1:45:54 [ __ ] but I feel I feel okay I feel
1:45:57 all
1:45:59 right
1:46:01 um Vicky Baptist she has to get up early
1:46:04 she's done
1:46:07 playing if you're still here Vicky go
1:46:10 sleep I'm gonna do it soon too what time
1:46:13 is yeah it's getting there all right
1:46:15 what if a AI was embedded or linked to a
1:46:17 person's brain as they passed and it
1:46:21 captured the
1:46:25 experience yeah ion
1:46:28 China I don't
1:46:31 yes I I mean one of the things that we
1:46:34 saw I was talking to to an about this uh
1:46:38 Ann Murphy um one of the things we saw
1:46:40 we saw multiple gpts created um for
1:46:44 Alzheimer
1:46:45 causes I lost my dad to alzheimer's
1:46:49 um
1:46:51 the ability to capture stories before
1:46:54 they're gone the ability for Alzheimer's
1:46:57 patients to connect with people that
1:47:00 they know or knew and have those people
1:47:05 be infinitely patient infinitely there
1:47:09 247 um is going to be huge and then
1:47:11 doing that at the brain level
1:47:15 where yeah like stuff like
1:47:19 that stuff like that is not like the
1:47:23 only difference between the the the M
1:47:26 the fmri that is allowing the Italian
1:47:29 guy to walk and what you just described
1:47:32 is just some researchers doing that
1:47:35 research right we are quickly we are
1:47:38 very quickly getting to the point that
1:47:40 these tools are going to
1:47:45 democratize
1:47:46 science
1:47:48 creativity knowledge work all all of
1:47:52 it and so someone who was not a brain
1:47:55 scientist is going to say
1:47:58 hey I've got a loved one who's going to
1:48:00 pass they're really into this stuff they
1:48:03 want to do this I'm going to figure out
1:48:05 how to do it and they're going to take
1:48:07 their their Raspberry Pi and hook it
1:48:10 into some fmri
1:48:13 machine at the facility and they're
1:48:15 going to be able to do
1:48:18 it and that's going to it that's going
1:48:21 to revolutionize a
1:48:23 something is chat GPT 4 much better than
1:48:26 three yes I've been thinking about
1:48:27 upgrading okay here's the deal I do not
1:48:30 work for open AI I do not I don't care I
1:48:34 don't have a bet on any of these
1:48:36 companies I'm not invested in them I
1:48:38 don't give a [ __ ] for me right
1:48:41 now the musthave subscription in all of
1:48:45 AI is chat GPT plus it's 20 bucks a
1:48:49 month it gets you all of what chat GPT
1:48:51 is just from a mathematical standpoint
1:48:54 GPT 3.5 is 175 billion parameter model
1:48:59 doesn't matter what that does it's just
1:49:01 points of data
1:49:02 that points of data that connect with
1:49:05 one another and give you answers gp4 is
1:49:10 eight different 1.8 or
1:49:14 180 um billion parameter models so it's
1:49:18 1.8
1:49:19 trillion so it's 10 times bigger but
1:49:22 it's also this thing called mixture of
1:49:24 experts so it's not just 10 times bigger
1:49:26 it's one Big
1:49:28 Blob what it is is it's multiple models
1:49:32 that each specialize in different things
1:49:34 so there's one model in gp4 that
1:49:36 specializes on programming there's
1:49:38 another one that special specializes on
1:49:40 language there's another one that
1:49:41 specializes on I don't know this kind of
1:49:45 processing it whatever it is so when you
1:49:47 do your prompt one of the things gp4
1:49:50 does it says which of these expert
1:49:52 models are the best ones to solve this
1:49:55 particular problem so it might pull
1:49:57 three of those things together and just
1:49:59 do those so so and and then on top of it
1:50:02 it's multimodal it's connected to the
1:50:04 internet it can see images it can make
1:50:06 images it can hear it can talk it can do
1:50:10 data
1:50:11 analysis you can generate these gpts
1:50:14 which are personalized customized
1:50:15 versions of chat gbt it's like the depth
1:50:20 of GPT 4 is
1:50:22 so mindboggling bogglingly beyond 3.5
1:50:28 and and it's it's also just bizarre to
1:50:33 me that in the
1:50:36 year that this 800 person company has
1:50:39 been building chat GPT you know they've
1:50:42 been around for six or seven years so
1:50:44 they've been building the foundation for
1:50:46 it for a while but chat GPT has been
1:50:47 around for a year it is staggering to to
1:50:51 me that a 200,000 person company like
1:50:54 Google hasn't caught
1:50:56 up Microsoft basically said well we'll
1:50:59 just use open AI they haven't caught up
1:51:02 meta said well we're not going to catch
1:51:05 up but we'll open source our [ __ ]
1:51:06 because we're
1:51:08 close like apple hasn't done any like no
1:51:10 one's caught up to these guys now gini
1:51:14 Ultra Google's purported gp4 Killer
1:51:19 looks promis ing but it's still not out
1:51:21 yet but yes I think the 20 bucks a month
1:51:24 for chat GPT plus is the best deal in
1:51:28 software right now I think it
1:51:31 is irresponsible if if you can afford it
1:51:34 if you can't afford it you can't afford
1:51:35 it don't pay for it but it's 20 bucks a
1:51:38 month don't have three coffee don't have
1:51:41 two coffees at Starbucks a month and you
1:51:44 can you can afford that
1:51:47 subscription if you want to be AI
1:51:49 literate right now I think not deeply
1:51:53 understanding what gp4 makes possible is
1:51:56 irresponsible like I think it's that
1:52:01 important so
1:52:03 yes yes it's different it is profoundly
1:52:07 better and there's nothing else close to
1:52:10 it right now and it's what's [ __ ] up
1:52:12 is I can't it is impossible to
1:52:16 describe why it's better you kind of
1:52:20 have to to experience it which is really
1:52:22 [ __ ]
1:52:27 up a virtual reality world where your
1:52:30 family is always there talking to you at
1:52:33 the age you
1:52:34 remember yeah exactly just me and the
1:52:37 trees yeah and know how to calm them and
1:52:41 communicate with them they'll benefit
1:52:43 yep Silver Fox
1:52:45 exactly thanks for the response Kyle
1:52:48 hartfelt thank you very much I read the
1:52:50 Sci-Fi story about using AI for
1:52:52 Alzheimer's patients yeah like we're not
1:52:55 we are so
1:52:57 close I I mean
1:53:00 2024 they're going to be
1:53:03 off-the-shelf
1:53:05 software where it will be relatively
1:53:08 trivial to make a virtual version of
1:53:11 me that someone else can interact with
1:53:14 as if they're talking to
1:53:16 me on one of
1:53:19 these
1:53:20 will it be as Dynamic is really talking
1:53:23 to me no but is it going to be better
1:53:26 than what it is now oh
1:53:28 yeah and in 3
1:53:31 years it'll probably be largely
1:53:37 indistinguishable I'm sorry about your
1:53:39 dad Kyle thank you it makes so much
1:53:40 sense regarding your passion for your
1:53:42 company yeah thank you so much
1:53:44 appreciate that yeah my company's called
1:53:46 Story vine it's a it's a video
1:53:48 storytelling company can you talk about
1:53:51 creating gpts I've heard vaguely but I
1:53:53 don't know how it works yeah so actually
1:53:55 go to there so this weekend this past
1:53:58 Saturday Shadow Ranger we did an event
1:54:02 called GPT for good it was a 24-hour GPT
1:54:05 challenge so every 3 hours we alternated
1:54:09 between myself and Anne Murphy she
1:54:10 taught on Zoom I taught on Tik Tock and
1:54:13 we taught people how to make gpts so if
1:54:16 you go here right now the the first link
1:54:19 here let me show that first okay so this
1:54:22 is just chat gp4 right I can go in here
1:54:24 and I can
1:54:27 say write me a song about quantum
1:54:32 mechanics that uses a black and white
1:54:35 singing dog as a
1:54:39 metaphor in a poem that rhymes wait in a
1:54:44 poem that rhymes like
1:54:49 Jay-Z
1:54:51 all right so so chat GPT all these GPT
1:54:54 things you can ask them anything and
1:54:56 they'll answer anything what a GPT is is
1:55:00 a specific sort of mini version of that
1:55:03 the the best way to think of it is this
1:55:05 if you think of chat GPT as like the the
1:55:08 iPhone operating system iOS the
1:55:10 operating system the the phone can do
1:55:12 anything you have like a little super
1:55:14 computer in your pocket and apple makes
1:55:16 some apps like the phone calling app and
1:55:18 the messaging app and the notes app and
1:55:20 the whatever apps that they give you and
1:55:22 then there's this whole ecosystem of
1:55:24 thirdparty developers that develop
1:55:26 really specific little apps this one
1:55:28 will tune your guitar and this one will
1:55:30 let you take notes in a different way
1:55:32 and this one will let you call people on
1:55:34 the other side of the world for free
1:55:36 right so all of those apps that's kind
1:55:39 of what gpts
1:55:41 are so I can go in here
1:55:44 [Music]
1:55:49 and
1:55:51 7s muscle car machine so this is one I
1:55:54 made that all this does is help you
1:55:57 helps you make a poster of a 70s muscle
1:56:00 car to hang in your room if you were a
1:56:02 teenager in the 70s or you're a gen xer
1:56:05 that's just stupid I'm ready to design
1:56:08 my dream car so I clicked on that link
1:56:10 and now the 70s muscle car machine says
1:56:13 fantastic let's get started first which
1:56:15 model do you want uh GTO Judge
1:56:22 great choice let's move on to color and
1:56:24 finish sassy
1:56:27 orange all right fantastic finally let's
1:56:31 talk about
1:56:32 modifications um
1:56:37 sleeper
1:56:39 um race version that is a
1:56:43 Resto
1:56:47 mod a Sleeper race version is awesome
1:56:50 aome and then it should yeah now it's
1:56:54 starting to make the
1:56:55 image so this GPT is just like a really
1:57:00 mini pre- prompted version of chat
1:57:03 GPT that came out of my imagination
1:57:07 could you do this with just regular GPT
1:57:10 absolutely would you think of making a
1:57:13 70s muscle car machine that makes a cool
1:57:15 poster of the judge and misspells it
1:57:18 that creatively no you wouldn't come up
1:57:21 with that but you don't have to now you
1:57:22 can just use my GPT so that's what they
1:57:24 are okay so if you go to the salon. it's
1:57:29 going to take you
1:57:31 to where is it I'll just go there the
1:57:35 salon. it's going to take you to this um
1:57:39 this link tree the first link is to the
1:57:43 directory of gpts that we made this
1:57:46 weekend we we did this event over 24
1:57:48 hours our goal was 100 gpts we ended up
1:57:51 making 151 well we didn't the people
1:57:53 that participated in this event did and
1:57:57 so these sessions were like an hour long
1:58:00 so in less than an hour and it it's
1:58:02 really about 10 minutes is the part
1:58:04 where you actually make the GPT it's
1:58:06 about 10 minutes um within an hour
1:58:09 people were making gpts for nonprofits
1:58:13 um that that they believed in or that
1:58:14 they were affiliated with we ended up
1:58:16 making 151 of these so that's the first
1:58:18 link you should go check that out
1:58:21 um the second link is the YouTube
1:58:24 playlist of the um of the session so
1:58:28 right now there's just my sessions in
1:58:29 there I'll be adding in and zoom
1:58:31 sessions as well so you can actually go
1:58:33 walk and then the next link is the
1:58:37 prompt sheet so this is the sheet that
1:58:41 actually has the exact prompt for how to
1:58:43 make a GPT and exactly what to
1:58:45 do and so how you make them it's it's
1:58:48 really quite simple you in the left hand
1:58:52 it has you have to have gp4 so so you
1:58:54 have to have a GPT Plus account and then
1:58:57 there's a little button that says
1:58:58 explore you click on the explore button
1:59:01 and the first option is make a
1:59:04 GPT and then once you're here you just
1:59:06 follow the instructions in that prompt
1:59:07 sheet and with like like seriously
1:59:10 within 10 15 minutes you will have made
1:59:14 one it's it's it's quite remarkable so
1:59:17 and you can go watch the videos they're
1:59:19 all up there and we'll I'll be sharing
1:59:21 them once once we have that thing kind
1:59:22 of dialed in all
1:59:29 right anyone in chat have the link for
1:59:32 the video where Kyle speaks Italian I'm
1:59:35 trying to think if that's up
1:59:38 anywhere it's on LinkedIn hang
1:59:45 on let me see Kyle Shannon attal
2:00:12 Italy learning Italian me neither anyway
2:00:16 that's me that's Italian be ready to
2:00:20 have your mind blown
2:00:22 check so go to labs. hen.com in the
2:00:25 lower left hand corner click on the labs
2:00:27 button and then go to video translate
2:00:30 you upload a video file you choose a
2:00:33 language and watch this so this is a
2:00:37 video from my storyvine platform what
2:00:40 you'll first see is me introducing
2:00:42 myself in English and then
2:00:48 watch hey it's Kyle shenon I'm the
2:00:50 co-founder and CEO of Story vine I would
2:00:52 love your input I'm inviting you to a
2:00:54 meeting I'm having on February 23rd so
2:00:57 it's coming up uh about modular content
2:01:00 and AI so let me dig into what's going
2:01:07 on so it's on it's on uh it's on it's on
2:01:11 my Tik Tok Channel there's a couple of
2:01:12 them on my Tik to channel so you can go
2:01:15 find them there um and then I think you
2:01:18 can find them also on link I just
2:01:20 searched Kyle Shannon Italian and it
2:01:21 took me to that
2:01:24 one um here's another
2:01:28 one and is important to them getting
2:01:30 more efficient with how we create
2:01:32 content and how we leverage
2:01:41 content behind me you see these little
2:01:43 green boxes um that's how we think about
2:01:45 stories at Story vine we break stories
2:01:47 down into little thank you Johnny 24
2:01:49 honey
2:01:56 that vine I have never said Story vine
2:02:00 like that in my life but I do now I
2:02:02 speak fluent Italian crazy right that's
2:02:05 from haen by the way if you're
2:02:08 wondering Mr it thank you for the the
2:02:12 ruse is very nice F this is why I insist
2:02:15 on the smartphone being in you mix of
2:02:18 iron and clay or not
2:02:20 in Grant that are
2:02:24 taken I will use all Techni technologies
2:02:27 that I can use without surgery yeah
2:02:30 actually so a fascinating thing
2:02:33 mantrid is
2:02:35 um so Elon Musk has his neural link
2:02:38 right where you drill a hole in your
2:02:39 skull or multiple holes in your skull
2:02:41 and then you this robot inserts the 60
2:02:44 electrodes in your brain uh because it's
2:02:46 a it's instant access to to your
2:02:50 brain
2:02:52 um some of the technology inside the
2:02:54 Vision Pro goggles are essentially that
2:02:58 without
2:03:00 surgery there's there's one so there's
2:03:03 like 5,000 patents inside the Vision Pro
2:03:06 goggles one of them one of them that I
2:03:09 read about is is mindblowing okay so the
2:03:13 the way the interface works with Vision
2:03:15 Pro is it's got four cameras looking at
2:03:19 your
2:03:20 eyes and then it puts this virtual
2:03:23 interface you know into your room right
2:03:26 that's got buttons and when you look at
2:03:29 a button if you pause when you look at a
2:03:32 button it it highlights it and and then
2:03:34 with your fingers outside the goggles
2:03:36 you do this to
2:03:39 click
2:03:41 well one of the things that the cameras
2:03:45 looking at your eyes are doing is
2:03:47 they're looking at the micro dilations
2:03:48 of your
2:03:51 pupils and they
2:03:54 know before you do that you're going to
2:03:58 click on an icon so as you're looking
2:04:01 toward an icon your pupils do something
2:04:05 that that show that you're getting
2:04:09 excited and you're going to click on
2:04:11 that so what they what they actually
2:04:13 start doing is pre-loading that software
2:04:17 before you click on it so that it's more
2:04:22 responsive so
2:04:24 like there there there's there's some
2:04:27 [ __ ] like
2:04:29 the they they have this chip in The
2:04:31 Vision Pro goggles called the R1 chip
2:04:33 it's the reality engine or the reality
2:04:35 chip it's basically just a machine
2:04:37 learning chip on device so they're not
2:04:39 having to process this stuff like send
2:04:42 it off to the internet have it processed
2:04:44 on a supercomputer and sent back it's
2:04:46 being processed in real time
2:04:50 like faster than you
2:04:53 can it it knows what you're going to do
2:04:56 before you do it and it starts
2:04:58 pre-loading the software so that is
2:05:02 essentially the [ __ ] that Elon Musk is
2:05:04 doing like like Elon might argue well
2:05:07 well the actual neurotransmitter would
2:05:09 be you know 14 micros seconds faster
2:05:13 than the than the other than the other
2:05:16 [ __ ] thing but you know there we're
2:05:19 going to have very sophisticated
2:05:21 technology that that does a lot of this
2:05:23 stuff without surgery so it's it's I'm
2:05:26 we're we are entering weird [ __ ]
2:05:29 times
2:05:30 people weird [ __ ]
2:05:33 times Sam in UK listening that Italian
2:05:37 blew me away wide awake
2:05:41 now yeah here you want to see I'll I'll
2:05:44 show you the um I'll show you the um
2:05:49 I've got a presentation that's got some
2:05:51 good stuff on
2:05:53 it oh this is a good one all right so
2:05:56 yeah so so this is this is this is a
2:05:59 presentation from from my company
2:06:01 storyvine so this is some of the AI
2:06:03 product that I'm putting together
2:06:06 um so so what my company does is we're
2:06:10 an 11 and a half year old company we've
2:06:12 got an app that guides a human being
2:06:15 through telling their story so it
2:06:18 prompts them step by step so so they say
2:06:21 um introduce yourself and when you were
2:06:23 diagnosed with this disease right and
2:06:26 you know talk about when you were
2:06:28 diagnosed and how that felt emotionally
2:06:30 and how how have you been treating it
2:06:31 and how are you doing and what are your
2:06:33 hopes and dreams for the future so
2:06:35 people answer a series of prompts in the
2:06:37 app and then our software takes those
2:06:40 answers stitches them together into a
2:06:42 video Auto magically in the cloud and 5
2:06:45 minutes later you've got a fully edited
2:06:47 video so you know powerful thing we're
2:06:49 actually not using AI to do that
2:06:51 automation we we're using just more
2:06:53 traditional assembly
2:06:55 stuff but now using this AI stuff I can
2:06:59 because I'm capturing content in this
2:07:01 modular way I can take those modules and
2:07:04 do all sorts of stuff with them like say
2:07:06 this so here's Ellen my greatest source
2:07:09 of support living with rheumatoid
2:07:11 arthritis has been um speaking with
2:07:14 other people about what they've gone
2:07:17 through uh reading a about the effects
2:07:20 of rheumatoid arthritis so there she is
2:07:23 you know that's the original video here
2:07:24 she is saying that same thing in
2:07:42 Spanish and so now I've got her speak in
2:07:45 Spanish well I can also because I've
2:07:48 because our software where builds videos
2:07:50 in this modular way I can also take any
2:07:53 onscreen text that's in English and I
2:07:55 can translate that to Spanish and then I
2:07:57 can so I can take all these individual
2:08:00 components that are in English translate
2:08:02 them all to Spanish and then reassemble
2:08:04 them so here's the reassembled video
2:08:06 with both text and and audio now in
2:08:12 Spanish so notice her
2:08:14 [Music]
2:08:17 name
2:08:22 [Music]
2:08:26 so notice the lower thirds in Spanish
2:08:29 and then I can take that video and I can
2:08:31 put it into a system that we call the
2:08:33 authenticity engine which generates
2:08:36 content about this video and um generate
2:08:40 transcripts chapters descriptions all in
2:08:44 Spanish and so and I can do this for
2:08:47 every video in our system and then I can
2:08:48 do that in 27
2:08:50 languages so here's Ellen speaking
2:09:02 [Music]
2:09:06 Hindi here she is speaking
2:09:09 [Music]
2:09:16 Japanese right I mean I mean we're we're
2:09:21 we are in it's this technology is not
2:09:26 coming it's here it's just that very few
2:09:29 people know about it yet and not this
2:09:31 technology but this technology is
2:09:35 basically the same core technology that
2:09:40 is sitting behind chat GPT is sitting
2:09:42 behind all these other magical models
2:09:45 that the
2:09:46 reason we're seeing innovation in so
2:09:49 many different areas all at once is
2:09:51 because it's all the same basic
2:09:53 technology just trained on different
2:09:56 data and so as you get more Curious with
2:10:00 this stuff you can start to do things
2:10:03 that
2:10:05 seem impossible right
2:10:08 now but they're
2:10:11 not we're in a different
2:10:16 era it's here it's not coming it's
2:10:20 here now is it perfect [ __ ] no is it
2:10:25 janky as [ __ ]
2:10:27 yeah is it ready for prime time yet no
2:10:32 close and getting better
2:10:37 fast if you want to if you want a sense
2:10:40 of the
2:10:42 progression look at Mid Journey images
2:10:45 from version one to version six version
2:10:48 one of mid Journey images was like 18
2:10:51 months ago or maybe two years ago at
2:10:53 this point not that long
2:10:56 ago version one mid Journey images
2:11:01 were you could tell it was trying to
2:11:03 make a human but it was not a
2:11:07 human and then version two was sort of
2:11:09 like a very deformed human and then
2:11:12 version three it started to look kind of
2:11:14 like a
2:11:15 person and then four it was good and
2:11:18 then five it was really good and now six
2:11:21 it's competes with whatever I introduced
2:11:26 another friend to Pi and she fell in
2:11:28 love with it so if you haven't played
2:11:30 with pi so here here's another thing did
2:11:32 you know this was
2:11:35 possible this is an iPad I have it on a
2:11:37 fancy keyboard stand but it's an
2:11:39 iPad and so I'm going to launch piie so
2:11:43 that's how you spell it not pie this is
2:11:46 huge for teaching oh it's crazy it's
2:11:48 crazy crazy so I click on
2:11:52 pi and then she shows
2:11:56 up and it's it's just a normal like you
2:12:00 know talk to pie and then there's this
2:12:02 little
2:12:03 telephone and so I'm just going to hit
2:12:06 the
2:12:10 telephone I'm a little bored and I'm
2:12:13 wondering if you can teach me something
2:12:15 that I don't know anything
2:12:17 about
2:12:34 p is thinking she's thinking too
2:12:40 long I I think this might not be
2:12:42 connected to the internet hang on let me
2:12:45 let me quit it and restart
2:12:47 it
2:12:49 too many requests ret
2:12:52 retry make sure your internet connection
2:12:55 is okay okay here we go yeah my internet
2:12:58 connection was
2:13:01 dead I'm a little bored can you help me
2:13:03 come up with something to entertain
2:13:10 myself absolutely here are some ideas
2:13:14 try a new hobby maybe learn a new
2:13:15 instrument paint a picture or pick up a
2:13:17 craft like knitting or wood working
2:13:19 exercise take a walk do a few yoga poses
2:13:23 or dance around to your favorite music
2:13:25 read a book dive into a new novel or
2:13:28 non-fiction book I have an idea I want
2:13:31 to say something in English and then I
2:13:33 want you to repeat it in some random
2:13:35 foreign language and have me guess what
2:13:38 language it
2:13:44 is that's a fun idea I'm happy to play
2:13:47 along you can and go ahead and say
2:13:49 something in English and I'll repeat it
2:13:51 in a random foreign language then you
2:13:53 can take a guess as to what language it
2:13:54 is sound good sounds great okay the
2:13:59 sentence is the Llama was walking down
2:14:01 the road and was about to step on a Rock
2:14:04 when he realized it was a
2:14:17 turtle W GOI is that
2:14:23 [Music]
2:14:25 Chinese yep it's Chinese you got it
2:14:27 right on the first
2:14:30 try give me another language same
2:14:38 sentence sure thing here it is in French
2:14:41 Lama
2:14:47 Mar doing a tour tour but wait you told
2:14:50 me it was French I didn't get a chance
2:14:52 to guess you don't know how to play the
2:14:54 game she'll
2:14:59 [Music]
2:15:01 apologize Oh no you're right I got ahead
2:15:03 of myself there my apologies I'll wait
2:15:05 for your guest next time so so that
2:15:09 exists did you know that
2:15:14 exists chat GPT can do that too by the
2:15:16 way the the mobile version of chat GPT
2:15:20 so if you download the iOS app from open
2:15:22 AI it's got a similar button and you can
2:15:26 put it in conversation
2:15:27 mode and then if you build a GPT this
2:15:30 personalized customized version of chat
2:15:32 GPT you can talk to them so you could be
2:15:36 on a road
2:15:38 trip and maybe you did a GPT that helps
2:15:42 you work through
2:15:44 issues about a specific topic you could
2:15:47 go on a 4-Hour road trip Tri and work
2:15:49 through some
2:15:52 [ __ ]
2:15:53 and when you get done you'll have a
2:15:57 transcript of that
2:16:01 conversation the movie version talk to
2:16:03 my daughter and it knew it was us by our
2:16:07 voice wow yeah there you go I mean yeah
2:16:14 so all of this stuff right now are like
2:16:17 teeny tiny little
2:16:19 glimpses on the mobile
2:16:22 version teeny tiny little glimpses into
2:16:25 what's coming in
2:16:27 2024 2024 is going to be Bonkers because
2:16:31 all of these models are going to get
2:16:34 multimodal and so the different
2:16:36 companies are going to try to one up
2:16:38 each other and provide different
2:16:41 features that make you want to use
2:16:43 theirs versus someone else's so there's
2:16:45 going to be different interfaces
2:16:46 different like like like Pi the company
2:16:49 behind Pi that company is called
2:16:51 inflection
2:16:53 and they raised $ 1.3
2:16:56 billion they spent about a billion
2:16:59 dollars of that
2:17:00 950 million on
2:17:04 22,000 high-end
2:17:06 gpus so that they can make their own
2:17:09 foundational model their own Frontier
2:17:11 Model so just like gp4 is a base model
2:17:15 they're going to make their own their
2:17:17 company
2:17:18 mission is to create the world's best
2:17:21 personal assistant so Pi is really good
2:17:25 at conversation it's going to get way
2:17:27 better at conversation and it's then
2:17:30 going to be able to take actions so
2:17:33 you're going to be able to have a
2:17:34 conversation with pi and say hey I'm
2:17:36 having an argument with my friend Steve
2:17:39 and here's what happened and I think I'm
2:17:42 right and I think he's an [ __ ] and
2:17:44 then Pi will go well I can understand
2:17:46 how you might feel like that here's
2:17:47 another way to look at it here's what
2:17:49 might be going
2:17:51 on 45 minutes later I'm like oh you know
2:17:54 what you're right I was being
2:17:56 closed-minded and I don't think I saw
2:17:59 his point of view can you help me draft
2:18:02 a letter that that you know takes
2:18:06 responsibility for my piece of it and
2:18:08 tries to open the dialogue back up with
2:18:10 him sure I'll do that how's this oh that
2:18:13 sounds great can you send that to him
2:18:15 now sure boom off it'll go and then
2:18:18 she'll probably go hey Steve got back to
2:18:21 us he apologized too and uh I went up
2:18:25 ahead and set up a meeting for tomorrow
2:18:27 does that work for you your your
2:18:29 schedule was
2:18:32 open like that we probably get in
2:18:39 2024 and we probably get it from Apple
2:18:41 we probably get it from PI we probably
2:18:43 get some version of it from open
2:18:45 AI Bard is they they've already demoed
2:18:49 some stuff that's in that neighborhood
2:18:51 we haven't seen it yet but you know Pi
2:18:53 is connected to the internet yes
2:18:56 Mike so yeah you can you can ask Pi like
2:18:59 tell me about the specials at Costco as
2:19:01 you're driving to
2:19:04 Costco hey I'm bored I'd like to see
2:19:06 some live music tonight who's playing
2:19:08 tonight you want me to book those
2:19:11 tickets it doesn't do that
2:19:14 yet it'll tell you about it right now
2:19:27 I was talking to Pi in my car and forgot
2:19:29 about it suddenly PI says if you keep
2:19:33 using that language oh that's
2:19:38 funny Pi's funny Pi Pi and Pi's got some
2:19:42 memory like she she'll occasionally go
2:19:44 oh
2:19:46 Kyle she remembers my name it told me I
2:19:50 might have to block your account or
2:19:51 something like
2:19:53 that she said let's keep it
2:19:56 polite yeah um Sydney the Microsoft one
2:20:00 is really bad with that that one will
2:20:02 shut down the conversation if you're
2:20:04 rude to it but again that's where I
2:20:06 think we're going to have different
2:20:07 models right I'm like a smartass and I
2:20:11 like to swear and I like to go oh you're
2:20:15 just a dumb
2:20:16 dumb
2:20:18 and like the Microsoft model gets
2:20:20 offended by that and shuts the
2:20:22 conversation down like so I I don't use
2:20:26 the Microsoft model at all because it
2:20:28 doesn't work with who I am I would
2:20:31 imagine that what's going to happen with
2:20:33 pi is that you're going to be able to
2:20:35 sort of dial in your safety stuff right
2:20:39 if you just want to say oh my gosh
2:20:41 instead of oh my God and and you wanted
2:20:43 to have those kind of boundaries and
2:20:45 that kind of
2:20:46 conversation um I'm sure you're going to
2:20:48 be able to say hey let's let's have that
2:20:51 and then it will have memory it'll
2:20:52 remember that and then if if I want mine
2:20:56 to be more of a smartass and more like a
2:20:57 stand-up comic like I'm interacting with
2:21:00 some you know smart
2:21:03 aliy you know
2:21:06 judgmental funny person then it'll act
2:21:10 like
2:21:14 [Music]
2:21:16 that
2:21:20 early Revel for me Papa JJ bye till next
2:21:22 time see you later Papa JJ yeah I got to
2:21:24 get out of here too let's see it's 10:21
2:21:26 it's time it's time people Pi talked to
2:21:29 my dentist and told him not to be afraid
2:21:31 of it oh that's funny so when everyone
2:21:34 is using it nobody will have real
2:21:35 conversations and AIS will just talk
2:21:37 back and forth
2:21:39 um for so Dei I so here's here's where I
2:21:44 go with that it so what you're saying is
2:21:47 is largely
2:21:49 true I will have agents that take
2:21:52 actions on my behalf you will have
2:21:54 agents that take action on your
2:21:58 behalf my agents will talk to your
2:22:00 agents and they'll talk to Jane's agents
2:22:02 and they'll talk to Sam's agents and our
2:22:04 agents will all coordinate and they'll
2:22:05 do [ __ ]
2:22:09 together what the agents will do is do
2:22:13 all the shitty work like hey are you
2:22:15 available then oh I've got a party then
2:22:17 I can't really do all the like when when
2:22:21 you want to like coordinate something
2:22:22 with your friends and it's like [ __ ]
2:22:24 3 days of ridiculous back and forth
2:22:27 conversations cuz someone had their
2:22:28 Muller pulled and someone else couldn't
2:22:30 do it and their cat just threw up on the
2:22:32 thing and all of those conversations can
2:22:36 be done by the [ __ ]
2:22:39 Bots and it's just like hey tonight
2:22:42 you're meeting with Sally Jim and Fred
2:22:45 oh cool where are we meeting you're
2:22:48 meeting here Fred's really excited to
2:22:50 talk about the [ __ ] you talked about
2:22:52 last time oh yeah what were we talking
2:22:53 about again remember you told Fred this
2:22:55 idea for the thing and then he had an
2:22:57 idea for the thing and then you two came
2:22:59 up with this idea oh yeah I remember
2:23:01 that yeah he wants to talk about that
2:23:02 again okay cool so then when I see Fred
2:23:06 we just start right where we were all of
2:23:09 that other [ __ ] is handled by the
2:23:10 robots yes sign me up for
2:23:13 that best local language model currently
2:23:17 local language
2:23:18 model oh want to self-host on my
2:23:23 computer I I don't it it depends on the
2:23:25 specs of your computer I I
2:23:28 mean the the um I I would say grab just
2:23:33 it depends how technical you are I would
2:23:35 say grab a piece of software called GPT
2:23:38 for all and that's just a an installable
2:23:42 that you can install on your Mac or your
2:23:44 Windows computer and then you can select
2:23:46 from different models and you can
2:23:48 experiment with them um what I would do
2:23:51 if I were you is I would I would go to
2:23:54 Reddit um on the open source AI find
2:23:57 some open source Ai subreddits and just
2:24:00 start hanging out there and then you're
2:24:03 going to you're going to want to play
2:24:04 around on GitHub and hugging face and
2:24:08 you're just going to need to educate
2:24:10 yourself because it's got to do with the
2:24:11 specs of your computer and how much RAM
2:24:13 you have and what you actually want to
2:24:15 do and why you want to host it and do
2:24:17 you want it to be um have safety guard
2:24:20 rails or no safety guard rails it it
2:24:22 just like it's the the amount of
2:24:24 parameters
2:24:26 of of what's best for you are [ __ ]
2:24:30 Bonkers so you're just going to have to
2:24:33 self-educate my computer is super
2:24:35 powerful awesome then you might be able
2:24:36 to play with the 70 billion parameter
2:24:39 stuff and have really powerful AI
2:24:43 there um you can also do virtual
2:24:45 machines right you could spin up
2:24:46 something on on GitHub
2:24:49 co-pilot or not GitHub co-pilot on U on
2:24:52 Google
2:24:55 um
2:24:57 uh what is it
2:25:00 Google uh
2:25:03 Google I forget what it's called it has
2:25:05 the notebooks it's not co-pilot it's
2:25:09 Google I forget what it is but anyway
2:25:12 you can spend up virtual machines that
2:25:14 have GP uh gpus in the cloud
2:25:18 GPT for all yeah that's it with LM
2:25:21 Studio you can download hundreds of
2:25:23 models none of them work exactly that's
2:25:27 the problem with the open source world
2:25:28 is like you have to be in it pretty deep
2:25:30 to understand what actually works what
2:25:32 doesn't what's been updated what hasn't
2:25:34 and and it's it's changing on a weekly
2:25:36 Google collab thank you yeah Google
2:25:38 collab you can spin up an instance of
2:25:40 Google collab with multiple gpus and you
2:25:43 can run really powerful models in the
2:25:45 cloud in a private instance that you
2:25:47 know your your data isn't going off to
2:25:49 to chat
2:25:50 GPT will AI decide who won for some
2:25:54 games first rule of the cult is not to
2:25:57 talk about the cult chat gvt not
2:25:59 allowing public domain stuff is a bit
2:26:01 silly yeah I
2:26:03 agree all right listen everybody I got
2:26:06 to get out of here my voice is going
2:26:08 crackly my eyes are
2:26:11 cloudy I'm still recovering from [ __ ]
2:26:14 Co and I did a I did a 24-hour GPT on
2:26:17 this weekend um if you if you want to
2:26:21 play around with some of the stuff we
2:26:22 did with
2:26:26 um with the gpon this weekend go to the
2:26:30 salon. it's a link tree the first two or
2:26:35 three the first three links are about
2:26:37 the G GPT for good event we did this
2:26:39 weekend it was a 24-hour event where we
2:26:41 made 150 gpts for nonprofits in a
2:26:45 24-hour period it was cool
2:26:48 um but the the videos of those sessions
2:26:51 and the directory of those gpts and the
2:26:54 the Google sheet that that we created
2:26:56 them with are all there and then also go
2:26:59 check out the AI Salon there's there's a
2:27:02 link for that there well as
2:27:04 well um it's a community of people like
2:27:07 if you're trying to figure this AI [ __ ]
2:27:09 out and you want to like hang out with
2:27:11 people that are up to it and like
2:27:13 curious about it too and are not
2:27:15 judgmental that's what group is it's
2:27:19 amazing it it it it's largely this group
2:27:22 that made that GPT for good event
2:27:26 successful this weekend thanks Kyle
2:27:28 you're amaz amazing best two hours on
2:27:30 here oh thank you very much I appreciate
2:27:32 that please if you have time check my
2:27:34 GPT the link is in the
2:27:37 S&T okay oh show and tell on the AI
2:27:40 Salon great terrific you're awesome Kyle
2:27:43 even when you're sick thank you Vicki I
2:27:45 appreciate that any tips on getting chat
2:27:48 to learn how to spell correctly no no
2:27:51 just keep trying it it I'll tell you
2:27:54 what the dolly3 spelling is better than
2:27:57 it was a month ago it got it got
2:27:59 markedly better in the past two weeks I
2:28:03 don't know what they did something got
2:28:04 better but it still sucks at spelling
2:28:06 but again here here's here's my this is
2:28:09 my point of view on life's tough it's
2:28:12 tougher when you're stupid
2:28:14 um here's the here's here's my opinion
2:28:18 on uh dolly3 spelling
2:28:21 shittily enjoy it now we are going to
2:28:25 look back on this era where [ __ ] is
2:28:26 spelled wrong like hilariously
2:28:29 wrong like nostalgically because I'm
2:28:33 guessing by 6 months from now all of
2:28:36 that's going to be solved it's just
2:28:38 going to spell it correctly it's going
2:28:39 to be [ __ ] brilliant be like oh you
2:28:42 remember when it used to do shitty
2:28:43 spelling that was kind of
2:28:45 fun have you seen mid Journey V5 I
2:28:48 assume you mean V6 I have seen V6 it
2:28:50 looks it looks pretty good it's
2:28:51 different you've got to prompt
2:28:52 differently like that's an that's an
2:28:54 amazing thing V5 you you learn to prompt
2:28:57 mid journey in a certain way V6 comes
2:28:59 out you got to completely relearn
2:29:00 prompting that's just the errow we're in
2:29:02 you just going to have to keep being
2:29:03 flexible people I never thought I'd be
2:29:06 evangelizing people on AI I know welcome
2:29:09 welcome to the AI learning lab that's
2:29:10 the whole idea of this is that I want
2:29:13 people to experience this [ __ ] I don't
2:29:16 mind if you hate it a i as long as
2:29:18 you've tried it most people that hate it
2:29:21 haven't tried it most people that have
2:29:24 tried it have their Kevin McAllister
2:29:29 moment and then they can't go back from
2:29:31 it and then they turn into weird ass
2:29:33 evangelists that's
2:29:35 okay that's why we're
2:29:38 here all right so without spelling test
2:29:41 I would have gotten Straight A's in
2:29:42 English class it's amazing recommend
2:29:45 folks join official mid Journey Facebook
2:29:48 group yeah it's oh the mid yeah the the
2:29:49 mid Journey Facebook group's really good
2:29:51 some incredible Creations there that's a
2:29:53 very well- curated uh group on Facebook
2:29:57 all right everybody I'm out of here
2:29:59 thank you to the mods it was fun
2:30:00 watching you party tonight while I was
2:30:02 off you know doing my I was just
2:30:04 rambling tonight so um do they have GPT
2:30:08 in China they have their version of it I
2:30:10 think they've got this GPT blocked
2:30:12 because it just answers questions and
2:30:14 they don't want that um thank you all
2:30:17 right I on China good to see you Lord
2:30:19 digital Gods good to see you 101 Source
2:30:22 Camp thank you thank you thank you
2:30:24 regulars jrc Emilio's wife thank you so
2:30:27 much Mr it good to see
2:30:32 you what's your name truck TR trenen I
2:30:37 like the
2:30:38 name
2:30:40 uh amethyst Gwen good to see you Nicole
2:30:43 there's a bunch of new people in here
2:30:44 great let's go uh um I on China Cosmic
2:30:49 lover good to see you Spike Jones thank
2:30:51 you for your comments questions thoughts
2:30:53 opinions I love talking about this stuff
2:30:55 as you know Earth rocker thanks for
2:30:57 being a subscriber Sky limbo all right
2:31:00 everybody Joe mama good to see you I
2:31:02 haven't seen you in a bit good to see
2:31:03 you happy New Year
2:31:05 everybody oh this is public service
2:31:08 announcement tomorrow is Tuesday go back
2:31:11 to
2:31:13 work
2:31:15 2024 160 people in here thank you uh
2:31:19 join the AI Salon keep coming back here
2:31:22 and uh I really appreciate you all more
2:31:25 than you know peace