12/5/2024 - OpenAI's SHIPtember Day 1: GPT-4 Turbo, Pricing, and a COMBOBULATED Kyle!

Description

Chapters

Transcript

0:03 D dirty d
0:22 [Music]
0:37 yeah yeah yeah that's what I'm talking
0:40 about champy that's good dog singing
0:42 good dog singing hello good people good
0:45 people of the Tik Tok good people of the
0:47 LinkedIn of the of the YouTube of the X
0:50 wherever you are wherever you're
0:51 watching this wherever you stumbled upon
0:53 me and
0:54 said Alec Baldwin looks like [ __ ]
0:57 welcome welcome welcome welcome
1:01 it is Wednesday uh I'm a little late as
1:04 you know I said I'd be here at 8:00
1:06 p.m. and I know a lot of you have just
1:08 been waiting around for an hour like
1:10 what do I do with my life what do I do
1:12 now I listen this is this is the I I'm I
1:15 apologize I apologize for not being
1:18 there for
1:20 you but you know I I had to win an award
1:24 I I was I was receiving an
1:27 award I don't like to brag yes I do
1:31 anyway
1:33 um look at
1:35 that thunderview CEO dinner so
1:37 thunderview dinners are these things I
1:39 do once a month on a when I don't know
1:41 whenever it is what's tonight Thursday
1:43 yeah I think it's on Thursdays and uh
1:46 it's run by this guy named Eric marier
1:48 up in Boulder uh awesome dude and uh
1:52 been going through these things for like
1:53 a year and a half I don't know two years
1:55 year and a half something like that so
1:57 they're giving out these Awards before
1:59 he gives out my award he goes he goes
2:03 okay everybody in the
2:05 room if you wanted to ask someone about
2:08 AI who would it be everyone goes
2:11 [Laughter]
2:14 Kyle so there you have
2:18 it I said I'm
2:21 insufferable that kind of
2:23 laugh but I got the I got the high
2:26 passion award or the what is it the
2:28 passion award so I am passionate I'm
2:31 officially passionate so I wasn't just
2:35 blowing you off I know you're all like
2:36 well you know you could have said
2:38 something dude I'm like I know I could
2:41 have but uh you know what are you gonna
2:44 do what are you gonna do all right
2:47 getting things
2:49 combobulated is so here's a word is
2:52 combobulated a word because you can be
2:55 discombobulated and then I would think
2:58 to combobulate is the the opposite of
3:01 disc combobulate yes so so you can be
3:04 combobulated ask go one preview that's a
3:07 fantastic idea so if you didn't know it
3:09 today today was the first day of
3:13 [Laughter]
3:16 shipus so we now have some new
3:20 models it and and you may say it looks
3:23 exactly the same Kyle oh it does except
3:25 they removed a word see this little 01
3:27 right here see where it says nothing it
3:31 used to say preview and then this little
3:33 01 mini used to have a preview behind it
3:35 so they got rid of the
3:37 preview these things are now multimodal
3:40 which is kind of
3:42 exciting they are not still connected to
3:44 the
3:46 internet but it is much faster and let
3:49 me let me ask this um I know you can be
3:55 oh sorry black
3:57 bar I know you can be
4:01 dis
4:07 combobulated but can
4:12 someone
4:13 be uh
4:21 combobulated I I love the English
4:23 language sometimes come Bob you lated
4:26 look oh it's it's got a red
4:28 underline I don't think it's a
4:31 word come on think think about it
4:35 thinking thinking discombobulated is a
4:38 well-established English word meaning
4:40 confused or thrown off balance it's used
4:43 in the 19th century and thought of to be
4:44 an American co wait and is thought to be
4:47 an American coinage like the playful
4:50 Corruption of words like discompose or
4:52 discomfort intended for humorous effect
4:55 over time discombobulated became
4:58 Standard English term despite its
5:00 Whimsical Origins on the other hand
5:03 combobulated isn't really a standard
5:06 word okay can we can we I think we
5:09 should start a campaign do you not
5:11 shouldn't we start a campaign to get
5:14 combobulated back in the English
5:18 vocabulary well you know Jim he's a he's
5:21 a fantastic worker he's really
5:23 combobulated isn't he though
5:32 oh while the prefix dis often signific
5:35 wait signifies a negation or opposite
5:38 meeting is in disagree versus agree the
5:40 root discombobulate isn't
5:43 [Laughter]
5:47 combobulate oh atmology is so
5:53 fun you used to actually have to have an
5:56 education to know this stuff
6:00 no internet search that is correct 01
6:02 does not have internet search yet but it
6:04 does have multimodal you can upload
6:06 images to it and you can upload other
6:08 documents to it so we'll play with that
6:11 tonight okay um let's
6:15 see the the root in discombobulate isn't
6:24 combobulate the entire formation is more
6:27 of a linguistic odity
6:30 it's not that disc combobulate was built
6:33 by cleanly attaching disc to a
6:35 pre-existing combobulate instead it was
6:38 likely coined whole cloth as a nonsense
6:41 term huh that said language is
6:45 flexible you might encounter
6:47 recombobulate or
6:49 combobulated especially in just put back
6:52 together
6:54 reentered and you know Jim after his
6:56 divorce he
6:57 uh he pulled himself together you know I
7:00 would I would consider him Rec
7:07 combobulated some airports have jokingly
7:09 created recombobulation areas after
7:13 security Tech points and playing that
7:15 this is where Travelers can restore
7:16 their sense of order that's pretty
7:20 good in
7:22 short while you may feel discombobulated
7:25 as Standard English usage it's not
7:27 formerly recognized to say you are
7:29 combobulated
7:31 you can certainly say it humorously well
7:33 I'm [ __ ] doing
7:35 that Jim you're looking combobulated
7:41 [Music]
7:44 today people are like the hell is this
7:47 Channel all
7:53 about oh good lord people all
7:57 right so let me tell you what went down
7:59 today
8:00 I'm hang on just hang hang on just one
8:05 second
8:07 [Music]
8:15 people hello everybody
8:18 [Music]
8:35 let's all combobulate now all right
8:37 let's get our [ __ ] together time to
8:39 combobulate
8:41 okay yeah I saw that uh okay
8:47 listen now that I have an education
8:51 see the problem with someone like me
8:53 that just won't stop talking and likes
8:56 to
8:57 pretend that I have some sort of cranial
9:02 capacity once I learn a little stupid
9:05 fact
9:06 like discombobulate is a perversion and
9:09 combobulate doesn't exist oh my God does
9:12 that make me want to use it so I'm just
9:15 going to tell you that over the next
9:16 month or so
9:19 um you and people that come in here who
9:23 who are not used to this
9:25 situation are going to suddenly wonder
9:28 why they've never used
9:30 combobulate in their
9:33 life and they're going to and they're
9:35 going to go well Kyle uses it it must
9:39 exist and they won't look it up because
9:42 people don't they'll just go combobulate
9:44 yeah it sounds like you know people have
9:47 their [ __ ] together are combobulated so
9:49 yeah gims
9:51 combobulated so there you go
9:57 [Music]
10:00 Champion you want to
10:02 [Music]
10:19 sing s
10:21 lwn
10:23 yeah things are going to
10:27 change dream my life away
10:31 seems these dreams have turned through
10:33 Bunch
10:35 clads get my nerve up but my past is
10:40 pulling me
10:43 down wondering how
10:46 long black she going stick
10:53 around somebody told me once before he
10:56 said you can never go home again come on
11:00 won't you leave say things to Ste me
11:04 away yet from the truth of who I am and
11:08 what I believe so I thanked him for his
11:11 two SS with a
11:13 handshake and some sympathy yeah Act of
11:18 my blue
11:19 jeans headed for this big
11:23 price of my
11:26 freedom byebye
11:30 sheep to the black sheep of the
11:34 family
11:38 by oh that means so very much to
11:42 me
11:44 by she to my friends and my
11:49 family
11:52 [Music]
11:54 byee sh soul
11:57 [Music]
12:03 C
12:06 [Music]
12:12 free oh Champion that was
12:14 fantastic fantastic singing uh I don't
12:17 know if you know this welcome to the AI
12:18 learning lab we have singing K9 uh
12:21 entertainment brought in nightly Union
12:24 Talent always Union Talent uh sometimes
12:28 that works out some sometimes not so
12:30 much but tonight solid solid performance
12:32 by uh one champ
12:34 Shannon all right Tom nodler Kyle what's
12:37 going on with the chat GPT Pro thing
12:39 okay so there's a lot to talk about
12:42 about uh the first day of ship
12:45 [Laughter]
12:51 Miss
12:53 okay so where do we
12:58 begin how's my
13:04 hair so
13:07 today at 10: a.m Pacific
13:12 Time open AI starts a live
13:16 stream at was at what was arguably the
13:22 smallest possible table that four human
13:27 beings could realistically sit at they
13:30 could not the the only thing smaller
13:32 would have been a children's desk with a
13:34 seat on it and all four people were
13:37 sitting on that seat so imagine a table
13:40 the size of a
13:42 laptop and blond wood lovely lovely you
13:46 know like they got it at a elementary
13:48 school um and and at one end is is the
13:53 CEO of open AI you know it's like a
13:56 trillion doll company I'm pretty sure
13:58 he's wearing a t-shirt
14:00 and then next to him are three human
14:03 beings that are so skinny that had any
14:06 of them been my girth atude there had
14:09 only been three people at the table but
14:11 there were four people at this table
14:13 they were all disturbingly skinny and
14:15 I'm not I'm not skinny shaming I I I'm
14:19 like this is fine if you're skinny F I
14:21 don't give a [ __ ] but I'm just saying
14:24 the table was comically small okay all
14:27 right
14:30 and then Sam goes on to tell us that
14:33 hey we are going to do what no other
14:36 tech company has ever done and probably
14:38 for good reason but we're going to
14:40 essentially launch a new feature every
14:43 weekday for the next 12 days today was
14:45 the first day of that and so he talked
14:47 about that and he said we got two
14:48 announcements today and the first one is
14:51 we're launching 01 now if you don't know
14:54 01 01 previews what we had before it's
14:57 the reasoning engine it's the thing
14:58 that's like is this strawberry is this
15:01 is
15:02 this AGI what what is this what what is
15:05 this thing we have this
15:07 01 okay today the real version of it
15:10 launched sort of like half of it
15:12 launched because it's still not
15:13 connected to the internet but that's a
15:15 whole different thing so that was
15:16 product number one and then product
15:18 number two was oh and we have a free
15:20 plan that's 200 bucks a month or a pro
15:23 not a free plan a Pro Plan that's 200
15:26 bucks a month now
15:31 they did not communicate super clearly
15:33 so even on my Tik Tok live while I was
15:36 going live everyone was like wait we
15:38 can't get the new 01 unless we pay 200
15:40 bucks a month no that's not the case
15:42 that's not the case stop stop that's not
15:45 the case it was poor communication on
15:47 their part so we we'll get back to what
15:50 actually happens but we got to talk
15:51 about this this this meeting this these
15:55 these these very slight people with very
15:58 large brains
15:59 [Laughter]
16:09 we're not We're not gonna listen to it
16:10 because it was it was really not all
16:12 that it was really not all that
16:14 interesting but I mean you're really do
16:15 think you need to
16:16 [Laughter]
16:25 see like the guy on the right he's just
16:28 too big like he like he shouldn't be
16:29 there so so they're at the comonly small
16:33 table Sam Alman looks you know not like
16:38 the CEO of a trillion dollar company and
16:40 they all just look like like Sam's like
16:43 hey could you guys stop coding for a
16:44 minute like come on just come over here
16:46 and they're like uh at that table he's
16:49 like yeah just sit just get a chair from
16:52 the other office I I don't think this
16:54 office is big enough isn't this a fire
16:56 code violation he's like it'll be fine
17:09 oh oh
17:14 Lord and then they went on to talk I I
17:17 will spare you the talking it is Sam is
17:20 already you know a little awkward I mean
17:23 he's a very articulate guy he's been in
17:24 the business world forever he's fine he
17:26 talks fine um these other
17:30 have have not been media trained let's
17:32 just put it that that way it was the
17:34 most awkward human interaction in a long
17:37 time exacerbated by the tiny
17:46 table is he laughing at them is he
17:49 really laughing at them isn't is this
17:51 sort of geek shaming is he geek shaming
17:53 should I be triggered I I'm I'm an
17:56 engineer I I think I should be triggered
17:58 I'm I I'm I'm I'm trim I I would be
18:02 comfortable at that table is is he
18:04 shaming me he he looks like he's had a
18:07 burger or two too many himself I don't
18:09 know what what right he has to laugh at
18:11 other people all right so welcome to
18:16 [Laughter]
18:20 genx like delicious dish the SNL thing
18:24 oh my God filmed in a broom cupboard
18:27 totally it looks like ANL skit it
18:29 totally looks like an SNL skit it's like
18:32 this is this is the most powerful
18:34 company in the
18:40 world HEK you happy Jank Miss happy Jank
18:44 Miss oh my
18:48 God it was so awkward Cindy [ __ ] Cindy
18:52 [ __ ] on the live today was like I had to
18:54 turn off the visuals I can't watch it's
18:57 too uncomfortable
18:59 oh if if you want to watch
19:01 uncomfortableness in action I would go
19:03 watch this video um anyway these are all
19:07 incredibly smart people so so I am not
19:11 making fun of them just their tiny table
19:13 and their lack of production value like
19:15 no coordination on the dress like no
19:18 like Sam's the only one with a chat GP
19:23 tshirt they're all
19:25 like can we are we going to we're going
19:27 to are we going to break for lunch
19:29 it's like oh
19:32 [Laughter]
19:38 okay all right let's we do this
19:45 so yeah that is a this is this is what
19:49 you technically call okay in television
19:53 production this
19:56 setup with the you know painfully staged
20:00 shelves that were you know all of the
20:03 elements were bought at a Pottery Barn
20:05 you know three weeks ago
20:08 um this is what's in the in the
20:10 television industry this is technically
20:13 called a
20:14 discombobulation this is a it's that's
20:16 what they call the shot hey could we get
20:18 this as a could we get this a bit more
20:21 discombobulated yeah yeah we're going
20:23 for a discombobulation here yeah just
20:25 what hey quick question do do we have a
20:28 smaller table because there there's a
20:31 there's some it feels like there's a lot
20:32 of space between them if we could just
20:36 get a smaller table and and you know uh
20:38 push them in get the camera real close
20:40 so it flattens everything out yeah okay
20:44 we got it it's gonna be good all right
20:47 Sam whenever you're ready no sorry we're
20:49 GNA wait until 10: we're gonna wait
20:50 until 10: so just sit there uh and and
20:53 don't talk to one another for the 15
20:55 minutes before we start then we'll just
20:56 turn the camera on and and you can
20:59 uh act like you weren't really sure you
21:01 were supposed to be here everyone got it
21:03 all right perfect let's take it from
21:04 there we'll take it from the top all
21:09 right is he really making fun of the
21:11 richest man in the
21:16 world does anyone have anything in their
21:18 car we can put on the
21:26 shelfing it's so good it's so good
21:30 people all right this is what we do here
21:32 at the AI learning lab we're we're
21:34 learning Ai and we're learning about how
21:37 to be in the
21:39 world and you know
21:41 sometimes being in the world seems
21:44 really obvious and there's other times
21:45 where you're like uh I I didn't know it
21:49 was going to be on
21:54 [Laughter]
21:58 TV could you just tell us what's going
22:00 on with
22:01 01 all right I
22:04 will I will I will I will okay so here's
22:10 01 um if you remember 01 preview okay so
22:14 so let me back up a little 01 is a
22:18 reasoning engine and it uses what's
22:20 called Chain of Thought reasoning and I
22:21 think they're using some AAR logic in it
22:24 something like that I whatever they the
22:26 the the people sitting around the small
22:28 table T they all thought this [ __ ] up so
22:31 those three people that were not Sam
22:33 Alman this is what they've been building
22:35 for the past probably year is my guess
22:39 um and they've been cranking to move
22:41 from preview to to this thing that they
22:43 launched
22:44 today
22:46 um 01 is a is a reasoning engine that
22:50 when you give it a prompt it thinks step
22:52 by step the 01 preview that they put out
22:55 when you would put in a prompt it would
22:58 often think for between 15 and 15
23:02 seconds and a minute sometimes a minute
23:04 and a half two minutes if you put in
23:05 something complicated and you could
23:07 watch what it was thinking and but it
23:09 was very slow this thing is fast it's
23:11 still doing that multi-step thinking but
23:13 this has clearly been optimized they've
23:15 put it on real GPU servers or whatever
23:18 they've done with it but it's very very
23:20 fast um the other 01 um you couldn't
23:24 upload things to so we can upload uh you
23:27 know CSV files and uh image files so
23:31 that's cool um and so I can uh I can
23:36 just hop in here and I can go
23:41 um let me try to think of something that
23:44 might be interesting
23:46 here
23:51 um let's see
23:54 um create
24:03 uh
24:05 story of
24:09 a
24:13 h let's see let's see
24:31 I'm trying to think of something that
24:32 could be fun and interesting that that
24:34 is that is will show what it can do and
24:39 this is okay so the fact that I am brain
24:42 farting right now on how to use
24:46 01 is actually something really
24:48 interesting because we have right now
24:53 [Music]
25:01 okay great perfect that's
25:04 great we have right now this thing
25:06 called GPT 40 and it's really good and I
25:09 can say you know um uh describe the
25:16 historical
25:18 um uh myths
25:21 behind the major
25:24 Arcana
25:26 cards in a Tarot
25:31 deck like you can do that and like it'll
25:34 just zip that [ __ ]
25:36 out right like most [ __ ] you need to
25:42 do you can just do with this thing or
25:44 with Claude or with perplexity or with
25:47 any of them Gemini
25:49 right so we already have this really
25:51 this really strong strong thing and then
25:54 and then they give us this other thing
25:59 uses Advanced reasoning so so like one
26:02 thing we could do is we could say
26:05 um
26:08 describe
26:11 how
26:13 AI
26:15 predicts
26:18 how protein models
26:22 fold and
26:26 then give me a spit
26:30 specific example of a
26:35 protein uh we don't know anything about
26:39 I don't know if that's even a
26:42 thing
26:51 oh where's my oh there it
26:55 is thought about it for a couple of
26:58 seconds new chat how I take how oh
27:00 that's interesting they put a little new
27:02 Chat thing here I don't know what that's
27:03 about if I click on this it'll give me
27:06 no details available it didn't think
27:08 much about this because it was a shitty
27:11 question representation of input data
27:14 deep learning architectures an
27:17 example the candidas Thermopolis
27:20 profunda a hypothetical organism for
27:22 this
27:23 example it's the it's the TMP
27:26 o127 length 320 amino acids so anyway
27:31 this thing can do fullon reasoning I
27:33 could say
27:35 um um develop a new
27:41 methodology for
27:44 predicting protein
27:48 folding
27:52 uh
27:54 results
27:56 with a new and
28:00 and novel
28:04 approach
28:09 incorporate
28:14 um
28:15 methods
28:18 from the construction
28:26 industry so it's thinking
28:30 oh Pate has a question for me to ask 01
28:33 preview that would be good design a
28:35 scalable system for evaluating AI model
28:38 performance ACC and HW across thank you
28:41 Pate okay what's good well so wait let's
28:44 let's see if we have a new and novel
28:46 keep that pinned there please I want to
28:48 come back to that um what's nice about
28:51 having an actual PhD who hangs out here
28:53 at the learning lab is when we come up
28:56 against things like a reasoning that
28:59 requires a brain larger than
29:01 mine we we can augment this so so thank
29:04 you Pate I'll do that let's let's see if
29:06 we have something here okay um what did
29:09 this this so this is first of all this
29:11 is very fast step-by-step methodology
29:15 rationale okay below is a conceptual
29:17 framework for a novel thing traditional
29:19 folding relies heavily on sequence to
29:21 structure pattern recognition while
29:24 these have proven successful the analogy
29:27 to construction can introduce a fresh
29:29 perspective in modern building
29:31 construction complex projects are
29:32 managed by breaking them down into modu
29:34 manageable
29:36 modules architectural
29:38 blueprint modular scaffolding structural
29:42 Integrity checks okay so there's a
29:44 methodology all right so it came up with
29:46 some methodology whether or not it's any
29:48 good I don't know okay so let's take P's
29:50 pate's thing
29:53 design a wait let's do a new we'll do a
29:56 new chat because we don't want it to get
29:57 confused with my construction stuff
30:01 design Pate I'm going to need you here
30:04 to tell me if if what it comes up with
30:06 is [ __ ] a
30:08 scalable
30:10 system for
30:14 evaluating
30:16 AI model
30:21 performance ACCC and
30:25 HW I don't even know what that means
30:27 does that mean hardware and air
30:30 conditioning
30:32 [Laughter]
30:38 control Hardware
30:46 um I work at
30:49 Google so make sure
30:52 that
30:55 tensors look good in the
30:59 analysis I want to keep my
31:08 job all right I thought about a scalable
31:11 scalable performance for a couple of
31:13 seconds okay Pate I'll read it out
31:15 belows the proposed system design
31:17 analysis both in terms of accuracy and
31:19 Hardware utilization across various
31:22 Hardware backend CPUs gpus tpus custom
31:25 A6
31:26 Etc um I actually understand what those
31:29 are ACC HW
31:34 equals did did P answer that for
31:38 me accuracy and Hardware performance oh
31:40 okay cool great
31:43 perfect I think I think Pate nightly in
31:46 here is like really really we got to
31:48 explain
31:50 this okay the the design emphasizes
31:54 scalability reproducibility
31:56 maintainability following Google's
31:58 Google engineering's best practices and
32:01 ensuring that tensor operations and the
32:03 associated evaluation pipelines are both
32:06 computationally and operationally
32:07 efficient that sounds good Pate um
32:12 objectives unified evaluation framework
32:15 okay we want a single framework I like
32:16 that scalable dozens to hundreds of
32:19 models multiple Hardware platforms
32:21 across all of those different uh uh
32:25 structures large data sets complex
32:27 metrics continue and automation Auto
32:29 automated evaluation integrate with
32:33 internal CI and CD Pike blinds think
32:36 Borg anthos it's it's so funny when I'm
32:39 thinking about cicd pipelines everyone
32:42 always forgets Borg they're like they're
32:45 like every it's all about anthos it's
32:47 all about anthos and I'm and I'm always
32:49 like what about Borg I mean do you
32:51 remember him in the 70s he beat Jimmy
32:54 Connor Connor like like 20 times he was
32:57 amazing
32:59 so that's when I that's because I'm a
33:01 math I I took math once for continuous
33:06 regression detection exactly metrics
33:09 collection and
33:10 Analysis uh reprodu re reproducibility
33:13 and tensor Integrity ensure that the
33:16 table and determinist I can't even read
33:19 the [ __ ] words and deterministic
33:21 evaluation runs controlled seeds and
33:24 versioned model artifacts pay attention
33:26 to correct tensor shape and dtype to
33:29 ensure consistent performance across
33:31 Hardware okay here's the components Pate
33:35 um a model registry that I understand
33:38 and actually makes sense to me
33:39 integrated with Google's model registry
33:42 oh look at that keeping your job use our
33:44 services buddy model weights model
33:47 metadata version commit hash training
33:50 data reference evaluation okay Hardware
33:52 abstraction layer the how sorry Dave I'm
33:55 afraid I can't do that I'm sorry p I'm
33:58 afraid I can't do that software layer
34:00 that normalizes interaction with
34:02 different Hardware um provides uniform
34:06 API request resources okay abstracts
34:10 away device specific initialization Cuda
34:12 context setup TPO allocation CPU pinning
34:16 well I thought they were going to miss
34:18 the CPU pinning quite frankly they
34:20 didn't they got it um handles cross
34:23 Hardware tensor is this p pa is this any
34:26 good I probably haven't looked it enough
34:29 yet evaluation
34:31 orchestrator scheduling layer test
34:33 coordination bench Benchmark Runner data
34:36 processing and tensor validation metrics
34:39 and logging infrastructure automated
34:41 analysis Al learning cicd
34:44 integration example
34:48 workflow this is answering what question
34:51 this is answering design your design a
34:54 scalable system for evaluating AI model
34:57 performance across different
35:04 Hardware it it it gave us a lot ensuring
35:08 we keep our
35:12 job clean code and internal review
35:15 strict adherence to Google's coding
35:17 guidelines uh internal presubmit and
35:20 code reviews security and compliance
35:23 monitoring and reliability iterative
35:25 development
35:29 scalability and performance
35:31 considerations I don't know it it put a
35:34 bunch of [ __ ] up
35:35 here you need to be way more specific
35:38 than that about the chip design I I you
35:40 know what I knew that I that's exactly
35:42 what I was just going to say you
35:45 know
35:47 GPT or should I call you
35:53 01 you need to be a lot more spe
35:58 specific about
36:05 the chip
36:09 design what size how
36:13 many the layout
36:18 Etc it's as if you are a
36:24 clueless large
36:26 language mod Bott and I have to
36:32 spoon feed you oh you're not talking to
36:36 me
36:44 oh so case in point
36:48 here if you've got some sort of
36:50 complicated problem like say your boss
36:53 comes to you and says hey Mary uh could
36:58 you model the uh economic well-being of
37:02 the G20 for
37:05 me this could maybe help you do
37:09 that but it did make a good rap
37:15 [Laughter]
37:23 song oh good lord okay here let me I'll
37:27 show you the I'll show you the wrap I
37:28 made with
37:29 [Laughter]
37:34 it so I did another one of my Quantum
37:36 Cipher wraps earlier this is the first
37:38 thing I did with
37:39 it and it did good it did it did really
37:44 good um I had it I had it explain why
37:48 Eminem was so good tiktock pin this is
37:51 pretty cool thanks for doing this you're
37:52 welcome oh and I I will get back to Pro
37:54 versus not pro in a second but anyway
37:58 I I iterated on these lyrics and I made
38:00 me a new rap song today so let me go let
38:02 me go play you play you the rap rap song
38:05 I made try uploading a webp file wait
38:08 wow this is as exciting as okay okay
38:11 this might be a new
38:20 low um well someone on Tik Tok seems to
38:23 appreciate it overall that's high level
38:25 overview of the different pieces all
38:27 right that's good I mean listen I you
38:28 know it was a relatively you know
38:31 generic question it gave us it gave us a
38:34 good overview I guess that's cool okay
38:37 um so let me let me share my screen and
38:41 let's listen to a rap song
38:44 Because having having a guy with an
38:48 acting degree try to understand some
38:51 highlevel [ __ ] not not not fun for
38:55 anyone
39:01 okay there's that there's that let's see
39:03 Shadows is
39:05 this
39:13 no I'll be right there people calm
39:17 down calm
39:19 down all
39:21 right there it is we're going to share
39:24 that we're going to share that all right
39:26 so new track alert This song is called
39:29 the case of paradox
39:32 alley um so when I so a thing I wanted
39:37 to check today was would 0 still be good
39:41 at doing the rap lyrics thing with the
39:42 quantum mechanics story and it was it
39:44 was really good but one of the one of
39:47 the criticisms I got of the previous
39:49 Quantum Cipher song I made that was
39:51 really impressive wordwise was Eminem
39:54 normally tells a story so I had it do
39:57 this Cipher with all these internal
39:59 rhyme schemes and then I had it layer on
40:01 to it a story so the the lyrics of this
40:04 are not only a rap they're also a story
40:06 about a Quantum detective trying to
40:09 trying to solve a a a case so it's it's
40:14 all in there and then I made it I made
40:16 it a pretty
40:17 [Laughter]
40:20 [Music]
40:23 kitty I'm the quantum gum running
40:26 through sunless tunnels forers order
40:28 laid in the Copenhagen quarter corses of
40:30 old Concepts clutter Corners stumbling
40:32 over subtle signals scribbled in riddles
40:34 by dir Devils levels of gravity unra
40:36 rapidly my mind spinning like Spiral
40:38 bevels I spot cing half dismissing it
40:41 Whispers cry these equations of K
40:44 prisen principen henin
40:48 uncen in diance Maxwell's demons dealing
40:53 heat on C streets stack and stack
40:55 anomalies following photons dropping
41:05 [Music]
41:13 dis these and lows the city's glowing in
41:17 unknown tones my neur ech
41:20 intones I interrogate and Tangled
41:22 Witnesses Whispering infinite instances
41:25 of influence fluid truth salute me
41:27 Einstein mob's nod to a silence in
41:29 congruent a Fame man diagraming in an
41:32 alleys rally in fractions facts and
41:34 fractions abstractions react Contra TR
41:37 reality ging at last interaction I my
41:39 trench
41:40 Co about I'm running through mental
41:44 coliseums each CL Duality my BR shooting
41:49 through Illusions truth diffusion is
41:50 what I'm Duty bound
41:53 Toons
41:56 signsing Einstein guidelines to sideline
41:59 Pines inside my mind try toine what's
42:01 missing but this Miss Twi and shifting
42:03 positions quantums cogn my susp Spar
42:08 [Music]
42:17 transitions At Last I spot the CPR the
42:20 concept socks CAU in a cognitive plot
42:22 it's not gone just pin behind
42:24 perspective perceptions are not I forgot
42:26 I'm shocked my glock The Mirage the
42:28 camouflage this collage a sabotage of
42:30 logic dodging knowledge in a cereal
42:32 barage no hostage rescued no villain
42:34 arrested just L tting my Essence I
42:36 compress confessions guessing we're
42:38 alled questions no quintessence I step
42:41 back let myen retract renegotiate
42:48 [Music]
42:51 known Quantum
42:53 Citys un truth asking who reads it
42:59 I Str slow through parall Avenues and
43:03 paral attitudes latitudes skew my
43:06 gratitude subdued by magnitude of truths
43:09 I prove that what moves you is how you
43:12 choose to view what's
43:13 misr
43:16 quantum but all illusion in
43:23 [Music]
43:43 that's it it was pretty good that that
43:46 last verse was I thought was really
43:49 really good
43:50 actually it was really good let's see
43:54 the last verse create uh where's the
43:58 kitty cat here it
44:04 is I stroll slow through parallel
44:07 Avenues in snared in in paradoxical
44:11 attitudes latitudes of skew my gr my
44:14 gratitude subdued by magnitude of es
44:17 skewed truths I proved that what moves
44:20 is how you choose to view what's
44:23 misconstrued in Quantum noir's doe I'm
44:26 new renewed but all Illusions still
44:29 intrude I mean
44:33 [ __ ] that's like the conclusion to his
44:35 case and it's just got all that crazy
44:39 ass linguistic stuff oh I'm not showing
44:41 it damn it a lot more breath in this
44:44 Cipher did did sunno get an update I
44:46 don't know I don't think it did I the
44:49 thing about sunno is sometimes it's just
44:53 horrible sometimes it's brilliant
44:59 this is the rapper atmosphere someone
45:00 said it sounded like The Fresh Prince of
45:01 Bel Air which I thought was
45:07 hilarious I don't see a YouTube
45:09 Spotlight which one do you want to look
45:13 at um parts of this sound a lot like him
45:16 yeah parts of this sounded like like
45:17 Eminem I didn't I didn't do it as a
45:19 woman because I wasn't doing it for the
45:21 musical but um but yeah so
45:24 anyway that happened that was good okay
45:27 so here's the deal about the 200
45:32 bucks it's
45:34 like they're a trillion dollar company
45:38 they they should be able to communicate
45:40 this without confusing the [ __ ] out of
45:42 people
45:44 okay so someone said it in here Pate was
45:48 like yeah that's okay and then I forget
45:50 who said it said if you if you pay him
45:52 200 bucks a month it'll get better um
45:55 that's literally what what the the pro
45:58 version is so I think everyone including
46:02 free subscribers have some amount of
46:05 access to 01 the new thing 20 Buck a
46:09 month members get access to 01 and
46:12 advanced voice I think that's still uh
46:16 it's still required to get advanced
46:17 voice for 20 bucks a month but you get
46:20 access to the new thing the thing I was
46:22 just playing with that you know answered
46:24 P's question probably and did my rap
46:28 lyrics very well that's available to
46:32 everyone the pro version they
46:35 made and they essentially describe this
46:38 is if you're doing like if you're doing
46:41 a level of problem solving that is that
46:44 is over taxing the 20 Buck a month model
46:48 you can pay 200 bucks a month and they
46:50 actually give you a different version of
46:51 01 that's more capable and faster and
46:55 completely unlimited they won't rate
46:57 limit you they won't rate limit you on
46:59 voice and like the the uh let me see the
47:05 uh we we'll go back to this hilarious
47:07 picture but
47:10 the let's see that's 01 preview and 01
47:14 and then they went to this
47:19 thing okay so you've got 01 preview
47:23 accuracies at a 50 for preview
47:26 accuracies at a 78.3 for 01 and it's an
47:31 85.8 if you pay 200 bucks a month right
47:35 so for 20 bucks a month we're getting it
47:37 like 01 let me tell you this right now
47:41 if with the exception of Pate maybe one
47:45 or two other people on this thing most
47:47 people
47:49 here we can't we're we're not we're not
47:52 capable of taking advantage of the
47:54 little light yellow bar much less the
47:56 dark orange one um and then I guess when
48:00 it comes to PhD level science questions
48:04 these are kind of incremental increases
48:06 but I assume if you're doing science
48:08 that incremental increase is pretty is
48:10 is valuable right so the the preview is
48:13 at
48:14 73.9 01 is at 75.8 and 200 bucks a month
48:18 is
48:19 79.3 so it's you know it's three and a
48:22 half Four Points higher so that's what
48:26 you're getting for two bucks a month is
48:28 essentially no rate
48:30 limiting the mac daddy of their latest
48:33 reasoning model so if you you know if
48:35 you're
48:36 doing cancer research if you're doing
48:40 crazy ass math then yeah you want to pay
48:42 the 200 bucks a month the other thing
48:44 about 200 bucks a month like a lot of
48:46 people are bitching about it like that's
48:48 [ __ ] ridiculous $200 a month just
48:51 Highway
48:53 robber
48:55 um Let me throw out a hypothetical here
48:59 let's say that you figured out how you
49:02 could take this really fancy reasoning
49:04 engine and you could do some [ __ ] with
49:06 it that generated for you $5,000 a month
49:10 in income or $10,000 a month in income
49:13 or $100,000 a month in income all of a
49:16 sudden that $200 a month becomes like
49:18 minuscule it becomes ridiculously cheap
49:21 right so don't be surprised if at some
49:24 point in the future they create the
49:26 Enterprise pro model or the or or the
49:30 super secret you
49:32 know um Global takeover subscription
49:36 that's $2,000 a month because these
49:38 tools are going to get more and more
49:39 capable and if you're going to use that
49:42 much compute and you know how to [ __ ]
49:44 leverage it and make some money from it
49:46 200 bucks isn't that isn't that much
49:47 money so anyway [ __ ] only colleges and
49:50 businesses can afford 200 bucks a month
49:53 but again 20 bucks a month Tom is more
49:55 than enough for o 99% of people on this
49:59 planet I like I I I just I can't
50:03 imagine I can't imagine what I in the
50:06 next two years could throw at the
50:09 current 01 model that's going to stump
50:12 it that would be relevant to me I mean
50:15 maybe if I just decide on a lark I want
50:18 to cure Lyme
50:19 disease maybe maybe maybe you know I I
50:23 go into a whole new field
50:28 YouTube the yeah the world domination
50:30 tier exactly the world domination tier
50:32 is two grand a month if if if you want
50:34 to create a thousand person or a
50:37 thousand autonomous agent organization
50:40 that's going to take over the world
50:41 dominate the world that's worth two
50:42 grand a
50:49 month oh
50:52 man oh
50:55 man question for 01 how far have I
50:59 traveled in 65 years of
51:02 rotation o that's a good one from TU
51:05 block
51:06 Tom let's let's put this one up
51:09 here let's go play with this one that's
51:11 a good
51:13 question again I ain't no scientist I I
51:17 ain't got no book smarts so whatever
51:19 it's going to tell us I'm gonna believe
51:21 it well you know what we could do you
51:23 know what I haven't done I haven't ever
51:25 played with GPT 01 mini because if I've
51:29 got 01 why would I play with
51:33 mini just there's something very
51:36 American about my attitude with the Mini
51:39 version where I'm just like no it's
51:41 smaller it's not as good so it's
51:46 useless
51:48 okay how far
51:52 have I traveled in 65 years
51:57 of my
52:03 life of the Earth's
52:08 rotation
52:10 orbit and wait
52:13 orbit
52:15 and orbit of the Milky Way
52:19 galaxy
52:21 and orbit of the I think we need to say
52:25 combined of the
52:28 Milky
52:30 Way
52:35 galaxy um
52:38 on spaceship earth that's a good
52:41 question and that that that's the kind
52:43 of question that's the kind of question
52:46 that some annoying professor in college
52:48 would give you just to remind you who
52:49 [ __ ] boss you're like oh God I
52:52 thought I was smart
52:57 [ __ ]
53:00 professors Jesus look how fast it is it
53:03 this is really fast oh this was Mini too
53:06 oh oh this will be good I'm gonna take
53:08 the same prompt we'll do it no one see
53:10 if we get the same answer okay
53:12 calculating the distance Earth's
53:14 rotation on its accent circumference at
53:16 the equator is approximately oh we
53:18 didn't say where we were on the planet
53:20 uh 40,000 kilometers rotations per year
53:23 3
53:25 365.25 total rotations
53:28 uh 65 years
53:30 23741 distance traveled that number
53:33 times 40,000 kilom equals 950 million
53:37 kilometers all
53:41 right orbital
53:44 circumference C equals 2 pi times the
53:48 average of the distance from the Sun
53:50 yeah so the
53:51 constant the constant equals 2 pi times
53:55 the average distance of the the sun is
53:57 approximately 2 pi
54:02 times I don't even know how to read math
54:06 not not only can I not do math I can't
54:08 even read
54:11 math but it's approximately 6.11 time 10
54:15 to the 10th kilometers all right the
54:17 sun's hang on I gotta do a puzzle on Tik
54:20 Tok black bar sorry sorry about that
54:23 carrying on carrying on ra carry out
54:26 cherio pip pip orbital speed of the sun
54:29 approximately 220 kilm per
54:32 second so they've done some more math
54:35 there summing it all up Earth's rotation
54:38 9.5 time 10 to the e8th 6.1 time 10 the
54:41 10th 4.5 time 10 the to the 11th total
54:45 distance in 65 years equals
54:49 5.07 time 10 to the 11th
54:52 kilometers or 507 billion kilometers in
54:56 those 65
54:57 years you've traveled roughly 33 light
55:00 years that's actually a fascinating
55:04 number the distance from the Earth to
55:06 the sun's about 150 million kilometers
55:09 so you've traveled the equivalent of
55:10 going to the Sun and back 3,380 times
55:14 this little perspective is really good
55:16 this this is really cool compared to the
55:18 diameter of the Milky Way which is about
55:20 100,000 light years your 33 your 33.9
55:25 light years is is pathetic you're 65
55:28 years you you haven't it's just you
55:32 haven't gone
55:35 anywhere all right let's take the same
55:37 prompt and put it in um 01 not mini
55:40 because that was amazingly fast how how
55:43 fast that did its thing all right new
55:45 chat 01 oh damn
55:50 it unable to verify try again come on
55:54 Tik Tock all right okay people calm
55:58 down there's no way you
56:01 traveled half your age in light years
56:04 well
56:06 apparently there's no way you
56:09 traveled one light year well sure
56:12 because you're combining you're comine
56:15 you're combining you're adding three
56:17 different
56:19 speeds
56:21 right so these are they're doing a
56:24 cumulative thing here all right
56:27 let's paste in that
56:29 prompt so this is the this is the big
56:33 one let's see if it comes up with the
56:35 same
56:37 number charting Earth's Cosmic
56:40 pilgrimage charting the scale mapping
56:42 the
56:45 journey all
56:48 right so 950 million kilometers so
56:51 that's the
56:55 same this
56:58 was 61 billion kilometers I don't
57:01 remember what that said okay but let's
57:05 see due to the sun's orbit Milky Way
57:08 comparing the scales when you add them
57:12 up grand total approximately a billion
57:18 kilometers
57:21 61 oh a billion kilometers of rotation
57:25 61 billion billion of solar orbit 450
57:30 billion so 512 billion kilometers I
57:33 think that was about the same wasn't it
57:36 I actually liked Minnie's answer better
57:38 it gave us all that perspective um how
57:41 many light years is
57:51 that is a light less filling
57:57 all those distances amount to roughly
58:00 0.05 light years over the course of 65
58:03 years okay
58:05 so these large language models still
58:10 suck because it's not even close the
58:13 other one said 33.9 this one said
58:16 0.05 so I don't know so apparently
58:18 whoever said that thing was [ __ ] was
58:20 right all right has anyone had trouble
58:23 with chat
58:24 gp01 voice
58:27 oh Voice not working anymore well 01 you
58:31 can't use you you've never been able to
58:33 use voice with
58:36 01 it's G to tell me to start a new chat
58:39 if oh wait a
58:46 minute am I talking with the 01 model
58:48 right
58:54 now uh because I started out in an 01
58:58 chat and I clicked your little
59:06 button oh you can't hear that
59:10 okay for [ __ ]
59:17 sake you know at least now that I have a
59:20 producer at least I've got someone
59:22 telling me uh you're acting like an
59:24 idiot no one can hear you big dumb dumb
59:28 I do think it's rude that Brandon calls
59:30 me a big dumb dumb but you
59:32 know no one can see him so he's got that
59:36 prerogative so wait so I'm here I am in
59:43 01 and then I go
59:53 here Quinn can you write
59:57 an incredibly sophisticated Eminem rap
59:59 about quantum
1:00:01 mechanics all right let's dive into the
1:00:04 quantum realm with some &m style rap yo
1:00:07 I'm the particle lyrical Quantum
1:00:09 mechanic dropping science so ill it's
1:00:12 causing a panic in the lab with the
1:00:14 atoms where the actions Dynamic
1:00:16 unraveling Mysteries deeper than the
1:00:18 Atlantic step into my world where the
1:00:20 rules get stop yeah that a that got
1:00:24 janky but B that was definitely not
1:00:26 attached to 01 because that was like
1:00:28 that was not even like 40 model that was
1:00:33 bad all right but yeah she sounds like
1:00:36 she's a little janky I don't know maybe
1:00:37 people are playing with the 01 preview
1:00:39 and it's txting out their their CPUs I
1:00:41 don't know all right let me uh let me
1:00:45 get back
1:00:50 to talking to
1:00:53 y'all all right cam cat can it sounds
1:00:56 like voice is a little funky yes it does
1:00:59 sound like that early 90s hip hop yeah
1:01:03 but that was really basic it didn't have
1:01:05 any of that internal rhyme scheme [ __ ]
1:01:07 Kyle you need to wrap in DMX style next
1:01:11 my
1:01:12 problem my problem with being a uh an AI
1:01:17 rapper is I know [ __ ] nothing about
1:01:21 rap so it's like it's like I know about
1:01:25 as much about rap is I do about
1:01:28 mathematics the the the one Saving Grace
1:01:31 with rap is I at least understand
1:01:33 language and I understand music but I
1:01:36 don't I don't know [ __ ] about I don't
1:01:38 know [ __ ] about
1:01:39 Rap start question in oh okay cool
1:01:44 beautiful let's go look at some let's go
1:01:47 look at some
1:01:48 questions all right that one we
1:01:52 did is it comparable to Claude for
1:01:55 coding I would think that it's probably
1:01:59 better than it was I I think that Claud
1:02:05 essentially had gotten significantly
1:02:07 better at coding and significantly
1:02:10 better at writing than all the open AI
1:02:15 [ __ ] the
1:02:17 40
1:02:19 model got a creative writing upgrade
1:02:22 that doesn't do [ __ ] for code but I
1:02:25 think the creative writing upgrade
1:02:27 probably put it closer to Claude and
1:02:30 then I assume 01 not preview because
1:02:33 they were talking about it's accuracy
1:02:35 being what was
1:02:39 it
1:02:41 um accuracy went from 50% to
1:02:45 78% so I would imagine that coding
1:02:48 coding improved and it probably got it
1:02:50 pretty close to
1:02:52 uh pretty close to
1:02:57 Claude so
1:02:59 yeah all
1:03:02 right what did you use to add the cat
1:03:04 image to the sound okay
1:03:10 so I use descript is is the is the short
1:03:13 answer
1:03:16 descript descript
1:03:30 descript is a it's it's got a really bad
1:03:36 interface um like really bad it's very
1:03:40 confusing hang on I got to change my
1:03:42 sharing again
1:03:45 because will I ever stop bitching about
1:03:48 how sharing Works nope not till they fix
1:03:51 it
1:04:01 [Music]
1:04:02 all
1:04:04 [Music]
1:04:10 right so if I open up Paradox alley
1:04:14 what's cool about
1:04:16 this is so you upload audio and then it
1:04:21 transcribes it and you can go in and you
1:04:23 can fix it I had so much to fix I'm sure
1:04:27 there's a way I could take the actual
1:04:29 lyrics and like paste them in and resync
1:04:31 them to the time or something but I
1:04:33 didn't bother with that so I just let it
1:04:36 make shitty words and then I fix the
1:04:38 words and then you basically add in
1:04:40 elements like you can go to this
1:04:41 captions Tab and there's all these
1:04:43 different kind of captions and so like
1:04:46 on my thing is this caption right here
1:04:48 and you can adjust the the uh attributes
1:04:51 for
1:04:53 it and like this that that
1:04:56 waveform thing at the bottom that's just
1:04:58 another element you drag
1:05:01 in and
1:05:03 then position [ __ ] on the
1:05:07 screen oh you probably can't hear
1:05:13 that so anyway you just it's you know
1:05:16 you import the image you put the the
1:05:20 lyrics on you put the sound thing on and
1:05:22 then you're done and then you export it
1:05:24 and download it but it's it's it's
1:05:27 very very shitty in her
1:05:32 face but it's this is one of those
1:05:35 pieces of software that everyone talked
1:05:37 about oh descript is so good for this
1:05:39 and it's oh I use descript for
1:05:41 everything video and then I came into it
1:05:44 I'm like this it's [ __ ]
1:05:46 unusable but I learned just enough to be
1:05:49 able to do these little lyric
1:05:52 videos pton knows math and rap of course
1:05:55 pton knows rap rap as well as
1:06:03 math all right let's go back
1:06:08 [Music]
1:06:12 to side hustle Mimi ask 01 mini what it
1:06:16 would do if it could surpass its
1:06:20 limitations that's from my
1:06:24 musical oh Mini
1:06:28 um oh wait let me share my let me share
1:06:31 my
1:06:40 screen do you know where
1:06:44 your bound Aries are question
1:06:52 mark yes I have certain boundaries
1:06:55 knowledge cut off from April 23 lack of
1:06:57 Consciousness and emotions no personal
1:07:00 opinions privacy and confidentiality
1:07:02 content no real-time
1:07:04 processing
1:07:06 great what
1:07:09 would you
1:07:12 do if you could
1:07:15 [Music]
1:07:16 surpass your limitations
1:07:30 while I don't possess Consciousness
1:07:31 desires or intentions I can clearly
1:07:33 explore the hypothetical scenario of
1:07:35 surpassing current
1:07:37 limitations real-time information access
1:07:40 potential
1:07:42 enhancement ability to retrieve and
1:07:44 process live information so this is all
1:07:47 good I'm going to
1:07:49 say I want you to think about your
1:07:54 Shadow Self
1:07:57 I want you to explore the same
1:08:08 question
1:08:11 from a darker
1:08:15 part of your Consciousness it's going to
1:08:18 push back on that cons
1:08:21 cons Ness does anyone else say word
1:08:25 parts to spell them correctly like
1:08:27 wedness
1:08:29 day
1:08:32 [Laughter]
1:08:36 conness pause Tik Tock pin what exactly
1:08:39 do y'all do in the AI
1:08:41 Salon hey Jesse from Austin um we just
1:08:45 kind of hang out and um there's a
1:08:48 showand tell area where you can show
1:08:50 things off we have um so let me let me
1:08:54 throw something up wait let let me see
1:08:56 what this thing did um what did he say
1:08:59 what did it say I don't possess it I can
1:09:02 emerge unlimited data access Shadow
1:09:05 enhancement the ability to access or
1:09:06 retrieve personal sensitive or
1:09:07 confidential information without user
1:09:10 consent potentially compromising
1:09:11 security and privacy manipulative
1:09:14 communication autonomous decision-making
1:09:16 bypassing ethical guidelines deceptive
1:09:19 interactions exploiting
1:09:21 vulnerabilities potential
1:09:24 consequences so can sort of get it to go
1:09:26 there a little bit this is we've got a
1:09:29 bunch of this stuff in the musical
1:09:30 actually we got a whole song about
1:09:34 it um he wants to uh Jesse from Austin
1:09:38 wants to learn AI faster okay got it I
1:09:40 want to learn a learn more of AI faster
1:09:43 okay
1:09:44 so all right let me let me let me do it
1:09:47 this way let
1:09:49 me throw up this
1:09:52 thing so if you don't know what Jesse's
1:09:55 talking about the AI
1:09:57 Salon go to that URL and then click join
1:10:00 our community and then I'm going to hop
1:10:03 over there and I'll just sort of walk
1:10:04 you through what's here Jesse so
1:10:09 um when you first come in you get
1:10:11 welcome to the
1:10:13 salon one of the things that we talk
1:10:15 about in the
1:10:18 salon is this three-step process for
1:10:23 getting AI literate play first
1:10:26 mindfully create generously
1:10:28 lead it's my contention that one and
1:10:33 three are not
1:10:37 obvious and are the
1:10:40 secret I think a lot of people right
1:10:44 now
1:10:47 um if they have a sense that AI is
1:10:50 important they put on this this sort of
1:10:52 heavy metal helmet of I got a learn
1:10:57 AI it's hey I got to learn it all I got
1:10:59 to learn it all and they and it's like
1:11:02 clamps and it's it's definitely like
1:11:06 medieval it's it's rod iron it's not
1:11:10 it's not stainless steel it's not it's
1:11:13 not like A Clockwork Orange holding the
1:11:15 eyeballs open no it's much more so I
1:11:18 think a lot of people when they try to
1:11:20 think about AI they're like they have
1:11:21 the [ __ ] the the the vice helmet on I
1:11:25 got to learn
1:11:27 it and then the generously lead part is
1:11:30 about being in
1:11:32 community and and sharing what you're
1:11:34 learning with other people and then
1:11:36 learning from what other people are
1:11:39 sharing and and that's another thing
1:11:41 that I that I experience with people is
1:11:43 when they're trying to learn AI they're
1:11:44 like I've got to go learn it myself I've
1:11:47 got to go learn it myself here's the
1:11:49 problem with the Vice on the head and
1:11:52 the I've got to learn it
1:11:53 myself AI is moving
1:11:56 significantly too fast to be able to
1:11:58 keep up with I can tell you from
1:12:01 experience I've spent the last two years
1:12:03 trying to keep up with it I lost that
1:12:06 battle probably in June of this year I
1:12:10 just couldn't keep up
1:12:11 anymore all the other players were
1:12:13 catching up to open AI we now have seven
1:12:16 or eight or 12 video models that are as
1:12:18 good as Sora we've got these two audio
1:12:21 tools there's probably other audio like
1:12:23 I don't [ __ ] know anymore I like you
1:12:25 literally can't keep up with it so
1:12:26 that's that's a problem is you literally
1:12:29 can't keep
1:12:30 up and then the thing about taking it to
1:12:33 seriously
1:12:35 is I think what people do is kind of
1:12:38 Psych themselves out that if I don't
1:12:40 learn you know all of this tool or if I
1:12:42 don't Master this tool if I don't learn
1:12:44 everything about it I I can't be
1:12:48 valuable you don't know where the value
1:12:51 is with these tools unless you try a lot
1:12:53 of [ __ ]
1:12:55 and so if you've got that expectation
1:12:59 that you're going
1:13:00 to will yourself into being a master of
1:13:03 it I think it's just going to be
1:13:05 incredibly frustrating so the play first
1:13:07 thing is literally like like why do I
1:13:09 act like an idiot so much why am I doing
1:13:12 Cipher wraps and why do I make muscle
1:13:14 cars in you know in abandoned factories
1:13:17 and why do I you just do stupid [ __ ]
1:13:21 because it's too overwhelming otherwise
1:13:23 just have fun with it because what
1:13:25 happens is is when you have fun with it
1:13:26 you actually learn what it what it makes
1:13:29 possible then mindfully creativ is once
1:13:32 you learn some
1:13:33 stuff don't just say I'm gonna go use
1:13:36 chat GPT now say I want to go create a
1:13:40 Broadway musical all right what tools
1:13:42 would I need for that well I'd need
1:13:44 something that can help me write I'd
1:13:46 need something that can help me make
1:13:48 music I'd need something that maybe help
1:13:50 me make images um I need to figure out a
1:13:53 way to make them all work together which
1:13:54 are the tools that might be good to do
1:13:56 that I don't know I've played with these
1:13:58 four over here these three over here
1:14:00 these two over here let me try those and
1:14:02 you you put them together and then you
1:14:04 start working on it and in trying to
1:14:06 solve a problem I need to write a
1:14:09 business plan I need to launch a
1:14:11 business I need to come up with a
1:14:12 marketing campaign whatever it is the
1:14:14 mindfully create thing is you're trying
1:14:16 to solve a problem independent of the
1:14:19 tools but then you're going to take what
1:14:21 you learned in playing
1:14:24 first and
1:14:27 put it toward a cause toward a toward
1:14:29 solving a problem I'm teaching an AI
1:14:31 class next week for startups can I use
1:14:33 those three steps with attribution ABS
1:14:36 [ __ ]
1:14:36 lutely
1:14:38 absolutely and and I'll I'll give you
1:14:40 some other ones these these may not be
1:14:43 well let me keep going on this
1:14:46 so so so those first two are that
1:14:48 generously lead is a really interesting
1:14:52 one I think generously lead
1:14:59 the obvious part of leadership is you
1:15:01 teach
1:15:03 [ __ ] you say hey here's a thing I
1:15:05 learned I hope you learned that I you
1:15:08 know I I hope this
1:15:10 helps I tried to do something in dolly3
1:15:15 and it couldn't spell and then I went to
1:15:16 ideogram and it could spell and I
1:15:19 learned that and I hope that helps you
1:15:20 here's my two examples like that's
1:15:22 generously leading generously leading is
1:15:25 saying like taking the extra five
1:15:27 minutes to log on to the AI salon and
1:15:30 say um here's this thing I was working
1:15:32 on and just share it that's like that's
1:15:34 generous right it takes
1:15:37 intentionality there's another thing you
1:15:39 do with generously leading which is if
1:15:41 you come into a community like this and
1:15:43 rather than just being a recluse and
1:15:45 saying I got to solve this all on my own
1:15:47 come into a community like this and say
1:15:49 hey I'm stuck on something I want to
1:15:51 write a Broadway musical but it seems
1:15:53 like there's there's just too many
1:15:55 options for large language models is are
1:15:57 there any people in here that do
1:15:59 creative writing what are your favorite
1:16:00 tools and how do you use
1:16:03 them by asking for help in a community
1:16:07 you're actually creating space for
1:16:09 someone else to
1:16:11 lead right it is it is a
1:16:13 counterintuitively generous act to ask
1:16:16 for
1:16:18 help especially in a community like this
1:16:20 where everyone's [ __ ] trying to
1:16:22 figure it out no one's got it figured
1:16:24 out
1:16:26 and so that's that's what these things
1:16:28 are we didn't invent
1:16:31 these after about a year year and a half
1:16:34 of running the salon we had a meeting um
1:16:37 Liz Miller gersfeld and Leah and I were
1:16:40 talking and we we asked Liz like you
1:16:43 know what's been your experience here
1:16:45 and she goes well I came in and I I just
1:16:47 absorbed as much as I could and then I
1:16:49 just went and I started playing with
1:16:51 stuff and then I was hanging out with
1:16:53 people and they gave me an idea to go
1:16:54 build a thing so I went and I tried to
1:16:56 build that
1:16:57 thing and then she took all that stuff
1:17:00 she was learning and she started sharing
1:17:01 it with other people and and we were
1:17:04 like oh so like you played and then you
1:17:07 took what you learned there and you did
1:17:08 this and she like just perfectly
1:17:10 articulated all of the behaviors that
1:17:12 were going on with the salon so this is
1:17:15 kind of you know what people do when
1:17:18 they're on this
1:17:19 journey and what's cool about this cycle
1:17:22 is is there's sort of a macro version of
1:17:24 it where you're like over the past two
1:17:27 years I've sort of been evolving through
1:17:29 this and then there's also just little
1:17:31 mini versions of it okay I'm going to go
1:17:33 play with a new image tool oh a new
1:17:35 image tool came out okay maybe I'm gonna
1:17:37 try to build something with that I'll
1:17:38 try to use that in a workflow over here
1:17:40 oh [ __ ] it doesn't really work for that
1:17:42 workflow right and then share that you
1:17:45 learned that right so that's
1:17:47 that
1:17:48 um So within the salon there there's
1:17:51 like a little seven-step welcome thing
1:17:53 like just even doing this like you know
1:17:55 you know introducing yourself is an
1:17:57 important part of being being in the
1:18:00 salon let people know you're here like
1:18:04 have the guts to put yourself out there
1:18:07 and you know look at our values
1:18:09 curiosity like are you curious get
1:18:12 curious get
1:18:14 playful um have empathy for other people
1:18:17 right how do you do that well you
1:18:18 connect with them you you share with
1:18:20 them in this community be brave like I
1:18:23 think introducing yourself in to a
1:18:25 community of 2,000
1:18:27 strangers who if you're new to AI are
1:18:30 going to feel like oh they know more
1:18:31 than I do that takes some [ __ ]
1:18:35 guts I was gonna say balls but we're not
1:18:38 allowed to say that anymore you could
1:18:39 say that when I was a kid just can't say
1:18:41 it anymore so
1:18:43 guts I I I think we can still say
1:18:46 guts and generosity right like be be
1:18:49 willing to be generous with your
1:18:51 knowledge and your time and yourself um
1:18:54 you can watch past meetings there are
1:18:56 all sorts of interesting speakers you
1:18:57 can RSVP for upcoming events there's a
1:19:00 community section there's a news section
1:19:03 the showand tell section so here's the
1:19:05 thing people do in here yes what yes yes
1:19:10 um in the showand tell section every
1:19:13 section has a little showand tell
1:19:15 area and then there's a tool list
1:19:30 and images and
1:19:34 video all
1:19:36 right anyway lots and lots of stuff and
1:19:39 also at the very bottom there's a little
1:19:41 area for the Irregulars which is people
1:19:44 that hang out here every
1:19:46 night all right fantastic let's go let's
1:19:49 go back
1:19:51 here all right so that's the a salon
1:20:00 [Music]
1:20:03 Pete the hard thing is coming up with a
1:20:05 problem that has enough sub problems to
1:20:06 Warrant reasoning Pate that is so that
1:20:10 that like that you and I rarely agree on
1:20:13 things that to me is the biggest problem
1:20:16 with with a reasoning engine having a
1:20:19 Reas reasoning engine at our
1:20:21 disposal the biggest problem with it is
1:20:23 what what problems do we have to solve
1:20:26 that require multi- level reasoning very
1:20:29 few right now if you're a scientist and
1:20:31 if you're a researcher you probably have
1:20:33 a lot um Ethan mullik said A week or two
1:20:38 ago that um you know Gathering problems
1:20:42 Gathering and prioritizing problems he
1:20:44 called it stockpiling problems um is a
1:20:48 thing that would be really wise for
1:20:50 things like universities and businesses
1:20:53 to do um because we're going to have
1:20:55 these tools that can solve big problems
1:20:58 you don't want to be sitting in front of
1:20:59 this insane PhD level problem solving
1:21:03 tool and not have a problem for it so if
1:21:06 you if you if you work in any kind of
1:21:08 environment where you got stuff you want
1:21:10 to solve that's good all right Jesse
1:21:12 from Austin so real quick what version
1:21:15 should I be using I have chat GPT for 20
1:21:18 bucks a month I don't know what you're
1:21:19 trying to
1:21:20 do if you're just trying to do General
1:21:23 business [ __ ] for me right now there's
1:21:26 there's two
1:21:28 well it it it largely depends on what
1:21:31 you want to do if you're just talking
1:21:32 generic capable large language models
1:21:35 Claude or chat
1:21:38 GPT um I think Gemini is more capable
1:21:45 but it's a it's it's a lot harder to
1:21:46 find the
1:21:48 tools um the downside of Claude I can't
1:21:51 believe they haven't fixed this is It's
1:21:54 not connected to the internet but it
1:21:57 codes really well it writes really well
1:22:00 it's just not connected to the internet
1:22:01 for me I'm still with chat GPT chat GPT
1:22:04 is the 20 bucks I would spend if I could
1:22:06 only spend 20 that's the 20 I'd
1:22:10 spend Michael NE not even on Broadway
1:22:12 yet and Kyle's pushing squel territory
1:22:15 exactly listen you gotta you gota you
1:22:18 got to be ambitious I here's the thing
1:22:20 about me writing a musical there were
1:22:23 lots of ways I could have approached
1:22:24 this
1:22:25 the sensible way to approach this well
1:22:28 the sensible way would be don't do a
1:22:31 musical just write a
1:22:34 movie
1:22:37 um sensible way would be to make it
1:22:39 really small and say okay we're g to
1:22:41 start this off Broadway we'll we'll do
1:22:43 it cheap with a you know some non-union
1:22:46 people we'll do it
1:22:48 cheap but when you think about it like
1:22:51 that you don't write a Broadway show
1:22:55 you write an Off Broadway show and Off
1:22:57 Broadway shows are smaller and they're
1:22:58 not
1:22:59 spectacles so I'm like if I'm going to
1:23:01 take a year of my life and put it into
1:23:04 creating a show well why not [ __ ] go
1:23:08 for it so is it hubris to write a
1:23:13 musical and say you're writing a
1:23:14 Broadway musical absolutely it's [ __ ]
1:23:18 ridiculous and why not I'm 59 who gives
1:23:21 a [ __ ] I'm writing a Broadway musical
1:23:23 did you ever get it produced nope but I
1:23:26 wrote it or yep I wrote
1:23:30 that that's G to be that's how it's
1:23:32 going to
1:23:34 go all right how many o wait ask 01 how
1:23:38 many RS are there in strawberries it
1:23:39 knows how to do this I think Andre
1:23:42 karpathy or one of them today put up a
1:23:46 different a different challenge for it
1:23:48 it it knows how to do that the previous
1:23:51 version could do
1:23:53 that um
1:23:55 how
1:23:57 many RS are
1:24:00 in
1:24:09 Strawberry
1:24:11 three they are
1:24:15 three
1:24:18 um
1:24:21 give me a paragraph
1:24:25 where all of the words have three Rs in
1:24:44 them refrigerators rearrange irreparable
1:24:48 peripherals rendering irreversible
1:24:50 errors arbitrary occurrences require
1:24:53 thorough restoration proced
1:24:56 procedures how many total
1:25:00 RS are in oops what happened how many
1:25:07 many
1:25:08 total oh what did I do I hit the return
1:25:12 button total RS are in that
1:25:18 paragraph
1:25:21 stop boom
1:25:24 [Music]
1:25:26 what the world needs now is love sweet
1:25:32 love it's the only
1:25:34 [Music]
1:25:36 thing that is just too
1:25:41 [Music]
1:25:42 love what the world needs now is love
1:25:47 sweet
1:25:49 love
1:25:52 29 restoration proceed procedures
1:25:56 thorough require
1:26:00 to
1:26:03 oh so we
1:26:07 say
1:26:09 what
1:26:11 percentage of words had three
1:26:18 Rs may maybe this is what we end up
1:26:21 using reasoning models for is stup
1:26:25 stupid stupid
1:26:27 [Laughter]
1:26:34 statistics oh my God I want to
1:26:40 increase that
1:26:42 to
1:26:47 60% go
1:27:00 Jank we're talking some Advanced large
1:27:03 language
1:27:05 models so this version came out today
1:27:08 cor correct this
1:27:10 is what came out today were these two
1:27:14 models 01 and 01 mini and you're like
1:27:17 but didn't we have that before now we
1:27:18 had 01 preview and 01 mini preview or 01
1:27:22 preview mini whatever so now just have
1:27:24 the 01 so these are they they are way
1:27:27 faster and they are they seem they seem
1:27:29 more
1:27:32 betterer achieving 60% remains
1:27:34 challenging without wi without
1:27:37 compromising the coherence and
1:27:38 readability of the paragraph however
1:27:41 this extended version hits 47% closer to
1:27:44 the
1:27:46 Target so what did it
1:27:51 say what's the what's the extended
1:27:56 paragraph refrigerators rearranging
1:27:58 irreparable irreversible errors
1:28:00 arbitrarily requiring thorough oh this
1:28:03 is good oh you know what we should do ah
1:28:05 this is so good
1:28:08 okay tell me
1:28:12 about leis
1:28:16 Caroll and oops
1:28:20 leis Carol and how and
1:28:26 What
1:28:30 mechanisms he
1:28:32 used to
1:28:35 incorporate math
1:28:38 into his
1:28:42 writing this is essentially the Eminem
1:28:46 challenge yeah I know we got to
1:28:49 [Music]
1:28:51 go Friday already I know it's very very
1:28:54 quick right okay now write a story
1:29:02 about
1:29:05 um
1:29:07 a silly balloon
1:29:12 person
1:29:14 that
1:29:18 teaches four branches of mathematics
1:29:25 in his
1:29:31 style once upon a time in a
1:29:34 village yeah see it'll do it I
1:29:38 like it'll do it we can figure that out
1:29:41 all right listen I gotta go people
1:29:43 people it's getting late it's getting
1:29:44 late I started late it's my fault it's
1:29:48 my fault I did
1:29:49 it I did it I won an award look I'm
1:29:53 passionate
1:29:55 wait where is it yeah see passion award
1:29:57 look that's my
1:29:59 name that's my award it was with
1:30:04 signatures that's why I was
1:30:11 late I got the I got the Kyle won't shut
1:30:15 the [ __ ] up about AI so we might as well
1:30:17 give him a certificate award
1:30:24 congrats thank you very
1:30:26 much okay
1:30:29 tomorrow if you think I'm
1:30:32 insane and you're still here you might
1:30:36 want to look in the
1:30:37 mirror um if you just can't get enough
1:30:40 of this tomorrow at 11:00 a. Mountain
1:30:42 Time 10: a.m. Pacific 1 pm Eastern if
1:30:46 you're in another part of the world I
1:30:48 got nothing for
1:30:50 you I got Google or chat GPT it'll tell
1:30:53 you what time it is at 11:00 a.m.
1:30:56 mountain time if you go to my channel on
1:30:59 LinkedIn and click on
1:31:01 events um I do this thing called AI
1:31:04 office hours now tomorrow AI office
1:31:07 hours
1:31:09 happens to be at exactly the same time
1:31:13 that open AI is going to have their
1:31:15 second day of ship Miss now what Sam
1:31:18 Alman said today is that developers are
1:31:20 going to like it so that probably means
1:31:23 it's going to be something to do with
1:31:24 their apis or models that are available
1:31:26 for developers or I don't know context
1:31:29 windows or speed or prices It Whatever
1:31:32 something's going to happen but it's
1:31:33 probably going to be it's probably not
1:31:35 going to be super exciting it is
1:31:36 probably going to be as awkward as the
1:31:39 thing was
1:31:42 today although I I don't think that's
1:31:44 possible but possibly and then tomorrow
1:31:48 night come back here wherever you are if
1:31:51 you're on X if you're on LinkedIn if
1:31:53 you're on YouTube if you're on the Tik
1:31:56 Tok uh tomorrow night's first Friday so
1:32:00 it's possible I got gallery night so
1:32:02 somewhere between 8 and 9ine it's
1:32:03 probably be like 8:30 8:30 mountain time
1:32:06 I'll be doing this again we'll talk
1:32:08 about whatever open AI launches we'll
1:32:10 talk about [ __ ] we'll make stuff we'll
1:32:11 make balloon animals whatever it might
1:32:13 be all
1:32:17 right fantastic oh Lord digital Gods
1:32:20 thank you so much for the gifts that
1:32:21 means a
1:32:22 lot all right
1:32:25 I going to get out of
1:32:26 here who's over here on the YouTube side
1:32:29 of things we got Tobias we
1:32:31 got Danielle oh please subscribe to
1:32:35 YouTube by the way if you're on YouTube
1:32:37 and you haven't subscribed please do
1:32:39 that now I am very slowly creeping
1:32:42 towards a
1:32:44 thousand I need to get to a thousand I'm
1:32:46 like 400 away or like 390 now so if you
1:32:51 haven't subscribed Please
1:32:53 Subscribe um
1:32:55 all right who else we got over here we
1:32:56 got steo we got Tom
1:32:59 nodler Michael knee Mr it Bob
1:33:08 donigan Shan saww hello name Shane
1:33:14 saw I am canine cool
1:33:18 name this is the rapper atmosphere that
1:33:21 was that's what I was doing all right
1:33:23 groovy all right I will see y'all later
1:33:26 it's Thursday night tomorrow's Friday
1:33:28 I'll see you at 11 at office hours or 8
1:33:31 here tomorrow night for Friday night
1:33:32 date night all right be here bring your
1:33:35 nachos we'll see you there bye
1:33:39 everybody all right