0:00

Let's do it .

0:03

You have taste in a

0:05

way that's meaningful to software people .

0:08

Hello , I'm Bill Gates . I

0:13

would recommend TypeScript

0:15

. Yeah , it writes a lot

0:17

of code for me and usually it's slightly

0:19

wrong . I'm reminded , incidentally

0:21

, of Rust here Rust , rust .

0:25

This almost makes me happy that I didn't

0:27

become a supermodel .

0:29

Cooper and Ness . Well

0:31

, I'm sorry guys , I don't know

0:33

what's going on .

0:35

Thank you for the opportunity to speak to you today about

0:37

large neural networks . It's really an honor to be

0:39

here .

0:40

Rust Rust Data Topics Welcome to the Data Topics

0:42

.

0:42

Welcome to the Data Topics Podcast . Rust

0:44

Data Topics . Welcome to the Data Topics Podcast

0:46

. Hello and welcome to Data Topics Unplugged

0:49

, your casual corner of the web where we discuss

0:51

what's new in data every week , from

0:53

llamas to AI licenses

0:55

, everything goes . Check

0:58

us out on YouTube . I don't

1:00

even know where we are else anymore , but we

1:02

do have a video version of this , so feel free to go

1:04

there check us out . We also share

1:06

the screen here and there on the episode , so

1:08

feel free to have a look or have a look at the show

1:10

notes . Feel free to leave a comment

1:13

or question or send us via email

1:15

. We'll try to get back

1:17

to you in a timely manner

1:19

, but no promises there . Today

1:21

is the 4th

1:24

of November of 2024

1:26

. My name is Murillo . I'll be hosting

1:28

you today , joined by the one and only Bart

1:30

, hi . Woohoo , he's back

1:33

. Actually , yeah , but I feel like , for

1:35

the people listening , I'm not sure if no , maybe , yeah

1:37

, it's been two episodes , right that Bart hasn't

1:39

been there . Two released episodes

1:42

, no . Yeah that's true . Yeah , yeah , yeah

1:44

, and a recorded one and

1:47

one of the recording has to be released right indeed

1:49

, indeed , indeed , indeed . So , uh , what's

1:51

your absence ?

1:52

you want to explain yourself , mister I went to

1:54

uh kratz to do a

1:56

uh , so an island in greece to do

1:58

a trail run . Basically nice how ?

2:00

was it it was fun yeah

2:02

, did you beautiful island ? Did

2:04

you ? Um , how do you say your performance

2:07

? No performance . I don't know how did you place in the run

2:09

? I was , uh , eight place

2:11

oh okay , so the

2:13

seventh loser , basically .

2:15

Well , exactly maybe next time , bart just

2:18

kidding , that's kidding .

2:19

How many people ? Eight , no , just nine

2:22

. Just

2:24

kidding . No

2:26

, I'm just kidding , I'm just , I'm just jealous because I couldn't

2:28

do it .

2:29

I couldn't do it , if I'm being honest , but it's really

2:31

nice , like uh , it was uh , like

2:34

you can do a lot of elevation there . Like it goes from

2:36

crap , goes from sea level to

2:38

2500 meters , wow , yeah , I didn't know it was that high actually

2:40

.

2:40

Wow , but it's crazy , like , because you said like , oh , it's really fun , no , what did you say in high

2:42

, actually , Wow , but it's crazy , like , because you said like , oh , it's really fun

2:45

, no . What did you say in the beginning ? How did you describe it ? I said it's really

2:47

nice , it's a beautiful island . No

2:49

, no , no . You said it's really nice because you have a lot of elevation

2:51

, but to me that sounds like the opposite . You know

2:53

, it's like fuck .

2:54

That is true ?

2:54

Yeah , they say that . No , never mind , I'm not going to

2:57

go there . So cool . We

3:06

also had Halloween last week .

3:08

Did you do anything special ? I went

3:11

trick-or-treating with the kids .

3:12

Nice , nice , nice , nice , nice . What about

3:14

you ? We went . It

3:17

was a long weekend , right ? So me

3:19

and my wife , we rented a sleeper

3:21

van . We went camping as well . Oh

3:25

, I heard this . Yeah , you heard . We'll talk later

3:27

about it . It was fun

3:29

. It was fun , it was a different experience , it

3:32

was cool .

3:33

Where did you ?

3:33

go In the German community

3:35

of Belgium . So we didn't go far , because also it

3:37

was the first time . We're like , yeah , let's see how it goes

3:40

. And so we rented a van

3:42

with the black sheep van . So they have

3:44

a lot of different ones and you can just like the sleep prevents

3:46

. They have like a kitchen and some , some of them have

3:48

like a shower and bathroom and stuff . Um

3:51

, so we just rented one and went to the german community . It

3:53

was good weather . It was a bit

3:55

like it didn't rain no , it was

3:57

good weather to camp I thought

4:00

it was a bit cold , but

4:03

I'm from brazil , so what do I know ? But

4:05

yeah , it didn't rain . It didn't rain

4:07

. We were able to do some hikes For November in .

4:09

Belgium , it was good weather .

4:10

Yeah , that's true , that's true , but I think the best day was

4:12

Sunday , when it was sunny , but that's the day we're coming back

4:14

, so it was like okay . But , yeah , it was good it

4:28

was good .

4:29

What do we have for today , bart ? I see here uh gpt search . What is this about ? Uh

4:31

, chat gpt release a new functionality which basically

4:33

is a search functionality , right like how you would

4:35

uh use google . Yeah

4:38

, you can now use chat gpt and you can

4:40

basically like . There's a small icon which you can

4:42

click and you're showing it . It on the screen now

4:44

. It's like a

4:46

globe with search next to it . If you

4:48

click that , it will actually search

4:51

the interwebs instead of just

4:53

relying on its Training

4:56

weights , exactly , yeah

4:58

, what I understand is that they

5:01

leverage Bing quite a lot to do this .

5:04

But it's all under Microsoft umbrella , because Bing

5:06

does that too .

5:07

No , Well , Bing is just a search engine .

5:09

But it does also use LLMs behind and all these things

5:11

.

5:11

Right , yeah well they now also have an

5:13

AI version to that yeah .

5:16

Have you tried it ?

5:18

I tried it over the weekend and

5:21

I guess it works quite kind

5:23

of okay .

5:25

Better than Google .

5:26

Just recap that Google had the like

5:28

things that you would uh , that you , I think

5:31

, where you before you

5:33

had a lot of hallucinations like , if you have a specific

5:35

search , like I want to know uh

5:37

, which , uh which , shops

5:40

are open on Sunday that also

5:42

provide this service , like very specific , yeah

5:44

, right , like you , you save a huge amount of hallucinations

5:46

. Um , and

5:49

now it's just uses actual search

5:51

results to come to uh

5:53

an answer . Yeah , I actually use it . Like

5:55

it was very random , like when I tried it , it

5:57

uh , it was just released and I was looking

5:59

for uh climbing

6:02

areas in the Ardennes that

6:04

had a lot of routes that were suitable for kids

6:06

. So , very specific , right

6:09

. If I would have done this before , I'm sure there

6:11

would be a lot of hallucination .

6:12

And now it's quite okay . Oh

6:14

, great that's good .

6:16

When you go to the page that you're showing now . You

6:18

also have different types

6:20

of widgets .

6:22

Sometimes they show a map and stuff like that but

6:30

I couldn't really reproduce that , to be honest , like so yeah , for the people that are just listening

6:32

, on the announcement page that we'll put on the show notes they have some like tabs kind of , so they

6:34

show different examples for weather and then there's

6:36

like icons for weather stock . So

6:38

you have the classic time series plot

6:40

thingy , sports with

6:42

the , the game schedules , news and

6:45

maps . But you can get it

6:47

to work well with the example

6:49

search I did .

6:50

I asked can you show this on a map ? And didn't work

6:52

. But maybe you need to have a specific prompt

6:54

to to show results on a map , right

6:56

?

6:57

yeah , I think one thing that I I'd like using

6:59

chat gpt's for , almost for like brainstorming . So actually

7:01

when we're camping we're like , okay , what are

7:03

? Like me and my wife , we have two

7:05

dogs . Um , it's a rainy

7:08

day , can you give me 20 ideas of things that we

7:10

can do around here ? And then , yeah , like a

7:12

lot of times it gives some hallucinations . But if I say

7:14

20 , even if 10 are non-hallucinated

7:17

I think it's also good , but a lot of times

7:19

they still say I need to check with the local establishment

7:21

to see if they accept dogs and all these things and

7:23

I think maybe this could be , uh , this should be more

7:25

, should be yeah , yeah . So it's cool

7:27

, really , really cool . Um , maybe

7:31

also one thing I did use chat gpt

7:34

for , while we were on the chat

7:36

gpt topic , um , last

7:40

week I had a Not

7:45

an issue , I guess , it was a bit of an issue . So

7:48

it was about networking stuff , right

7:50

, and I'm not I think I'm not super comfortable

7:52

with networking and all these things . So I just kind of said

7:54

, okay , there's an issue , and then someone I'm

7:56

just putting the conversation here , I'm

7:59

not going to read through it all , but basically

8:01

that was an issue Someone was trying to deploy

8:03

, maybe before that setting the

8:05

scene , someone was trying

8:07

to deploy um two

8:12

applications in one vm , so it's just a vm

8:14

on the cloud . Okay , uh

8:16

, back end and front end , so like past api

8:18

and streamlit , right , um

8:20

, they're using docker compose and when they're

8:23

trying to deploy , both

8:26

addresses were working , so you can

8:28

access the documentation on the FastAPI

8:31

, you could access the streamlit . But whenever they

8:33

tried to talk to each other , something

8:36

was going wrong and just this connection refused , okay

8:39

, and I was

8:41

like , okay , I had to kind of solve

8:43

it and I wasn't sure where to start , right , and I was like Googling stuff and I was like , okay , I had to kind of solve it and I wasn't sure where to start

8:45

, right . And I was like Googling stuff and I was like you know what ? I'm

8:48

just going to ask ChatGPT , so also

8:51

as like a learning tool , right , and

8:53

this is the chat that I was showing before . So then

8:55

the first thing I just kind of said , hey , you're an

8:57

expert , blah , blah , blah , and then it gives

8:59

me a blob about ports

9:01

and addresses and all these things there were some questions

9:04

about , like the port , because sometimes the port was specified

9:06

and sometimes it wasn't . So then it explained a bit

9:08

some things like that . And then here it kind

9:10

of explained exactly what I mentioned here

9:12

with some dummy information

9:14

, right . And then it gave me a whole

9:17

bunch of examples , like , of possible

9:19

issues , right . And the

9:21

first thing they said is the course , cross-origin

9:23

resource sharing configuration , which

9:25

I've heard a few times but I never

9:27

quite understood . And

9:30

then they say firewall , group settings , protocol mismatch , and

9:32

I was like , ok , this is probably not correct , not correct

9:34

, but let's try this one , right . And

9:36

then I try some some things Right , like

9:38

, for example , I try to go in the

9:40

VM and curl the endpoints and I say , well , that's

9:42

working . Does that prove that it's course ? And

9:44

then I said , oh , actually , no right , because

9:47

this only happens on the browser . Blah , blah , blah

9:49

. And

9:54

then they also mentioned to go in the developer tools . I tried that

9:56

. I couldn't see anything . But then I also went on the browser

9:58

and I used the , the yeah , basically the console stuff , right

10:01

, to do a curl request and then

10:03

I got the course error and then I said , does this

10:05

confirm the issue ? And then it was like , yes , this confirms

10:07

the issue . So I was like , okay , now I need to . Okay

10:09

, how can I solve this now ? And then he gave

10:11

me instructions for the fast api . You can

10:13

do this , you can admit aware , blah , blah . And it actually

10:16

worked right and I just

10:18

thought it was a bit of a . It

10:20

was a bit of a mix right , like it did instruct me , quite

10:22

a . But I also had to use a bit of what I already

10:24

knew to try to confirm the issue and

10:26

then take steps there . And

10:28

I was actually super happy . I don't think I would have been able

10:30

to do it without something like Chachapiti to just give

10:33

me ideas , to just bounce back , to just like rubber

10:35

duck my way through it kind of I mean more than rubber

10:37

duck really , you know , yeah

10:46

, I thought like from it was really like wow , like that that's what it was , you

10:48

know , and something that I yeah , like you know , I don't know it was it felt very

10:50

magical . You know , it was just like oh , this

10:53

is super cool , um , and I think

10:55

it's also mixed a bit with like what I knew

10:57

already and trying some different things to confirm

10:59

. And I also think chat , gpt is really good for this

11:01

task that you can kind of verify yourself

11:04

right . And I think coding is a very

11:06

good example of these things because it gives you something and

11:08

you can try and either it will work or either it won't work

11:10

right . And I think for things

11:12

like I don't know if you're sick and

11:14

you ask ChagPT , it's probably not a good example because

11:17

you cannot verify the things that it's saying

11:19

right . So I think there's

11:21

a big use case where we talked about before , about

11:23

brainstorming , like giving ideas , but

11:26

also about these verifiable questions

11:29

, that if it hallucinates , and also

11:31

low stake a bit . So I think a doctor

11:33

is also something that is high stakes For coding . Maybe

11:36

it's lower stake , like if

11:38

ChagPT gives me a command to drop a database

11:41

, maybe I wouldn't just use it , I would do

11:43

some more research . But I think for like low stakes

11:45

thing that you can verify yourself , chatgpt is a really good

11:47

use case for it .

11:48

Yeah , and especially like how you use it here , like very

11:50

iteratively right , Like not just

11:53

generate me the answer , because that is often wrong

11:55

.

11:55

Yeah , indeed , indeed . But I feel like whenever

11:57

you hit a wall , sometimes it's good to just kind

11:59

.

12:00

Yeah , to get a bit of feedback . You would normally

12:02

reach out to a colleague .

12:03

Exactly you can with a lower threshold

12:06

ask something like ChatGPT

12:08

. Faster response as well . Right , If you send a message

12:10

, sometimes you have to wait .

12:12

But I think , like how you

12:14

describe it , a bit rubber ducking approach , that really

12:16

helps . Yeah , that's valuable .

12:17

Indeed , and I think , even

12:20

if the ChatGPT hallucinates and it gives you something way off

12:22

maybe that's some , that's something that is way off

12:24

gives you an idea of something that is more relevant

12:26

. Maybe

12:29

, maybe so , for example , when I was uh

12:31

doing something with rust and then they said

12:33

, so this , this , two traits , and I was like , ah , that actually

12:36

is not right , but maybe I should look more

12:38

into the type system and maybe this and maybe that , you

12:40

know , and actually it gave me a new

12:42

perspective , you know , and I think sometimes , at

12:44

least for me , I get so focused on one thing

12:46

, like , okay , I think that's the problem

12:48

, so how to try to solve this , this and this and this and this , and I

12:50

couldn't get it . And then it's like , well , what if this is not the

12:52

problem ? And that is the problem , and

12:55

having something that just like

12:57

a parrot , that just says something , uh

12:59

, helps me a lot . Okay , yeah

13:02

, yeah , so quite happy with it , not

13:04

, yeah , I think there's other ais , right , like findcom

13:06

as well , that gives more um , because

13:09

it also searches the web , right . So

13:11

that may be better , depending on what you're trying

13:14

to look for , because I also think that this was very

13:16

generic , right , it was like web

13:18

standards and stuff like that . So it's

13:20

only when you use it consciously to

13:23

debug or something it works .

13:24

I think what it doesn't work is when you use it consciously to debug or something it works . I think what it doesn't work is when you

13:26

use it as a , as a bit of a shortcut yeah

13:28

.

13:28

Yeah , yeah , yeah , yeah , yeah , yeah , yeah yeah .

13:29

Yeah , yeah , yeah , yeah , yeah , yeah , yeah , yeah , yeah , yeah , yeah

13:39

, yeah . Something like javascript and html

13:42

intermixed yeah and I just asked gpt

13:44

to to add a feature to it , change

13:47

it , and it didn't work at all , but

13:49

you knew what that ?

13:51

you knew that it wouldn't work when you read it I

13:53

didn't read it , like it was a big component like I just

13:56

tried to copy paste it just to see it just

13:58

as a test to see what it gives .

13:59

Yeah it doesn't work at all , yeah it was way

14:01

off , but then I tried it with cloth and

14:03

it was actually better . Oh really , yeah , I think

14:05

the performance was better , but still not not very

14:07

difficult but I think , as again , like if you

14:09

do this a bit more consciously , not trying to have a

14:11

shortcut , yeah , like it works to to very

14:14

easily adjust something that you're

14:16

not like . I don't , I never write liquid right

14:18

like this helps me to very quickly

14:20

uh make adjustments by by

14:22

interactively asking stuff .

14:24

Yeah , true , I also think that I don't know how

14:26

specific liquid components are

14:28

.

14:28

Well , the problem , I think , with liquid is that there's

14:31

probably not a huge amount of open source

14:33

liquid out there , right , exactly , there

14:35

is on Python .

14:36

Yeah , yeah , there is on Rust , but it's also like what

14:39

is in the training set right

14:41

. Yeah , indeed . Yeah

14:44

, I also wonder these things , things like if I'm asking something super

14:46

specific or something that I feel like it's very new

14:49

, um , maybe

14:52

I don't ask hbt because I know that

14:54

I need the freshness you know , like the

14:56

fresh information , um , but

14:59

yeah , I think indeed , it's kind of like knowing what

15:01

to what , to prompt where , right

15:04

, right , but I

15:06

think it's a valuable tool . But

15:09

, yeah , I wouldn't ask ChatGPT

15:11

for something like Python 3.13 , right , because

15:13

it's very new . Well

15:16

, with search you could do it , right . Ah , that's true

15:18

, with search you could do it , but actually that's what

15:20

I was using findcom for right

15:22

, which is a bit the search , but I think findcom for right , which is a bit the search , but I think

15:24

findcom , I think they brand themselves as like for developers

15:27

.

15:27

Let's see yeah , findcom to me

15:29

is a bit like perplexity , but then

15:31

for developers what do ?

15:33

you mean perplexity for developers

15:35

like perplexity . Yeah , like it's just advanced

15:37

search ah , yeah , yeah , yeah , yeah , but

15:39

that's the thing so here . So I'm just showing the ui

15:41

for findcomcom

15:43

Just put your question here but it says from

15:45

idea to product , but they also

15:48

have like a playground in code , so it does

15:50

feel more catered towards

15:52

developers , right , so that's what I was

15:54

using , but maybe it's like chat , gpt search will

15:56

actually fulfill that need

16:00

now , right , but

16:02

yeah , maybe , as I mentioned

16:04

, python 313 , python 313

16:06

is out since october

16:10

2024

16:12

, so it's been a little while , but I don't

16:14

think we got the opportunity to talk about it . It

16:17

made some noise on the python community for

16:19

two main reasons , I would say the

16:22

free threaded mode and

16:24

the just-in-time compiler . I

16:27

think those two things they made a quite a bit of

16:29

noise . Pep 703 and pep 744

16:31

um

16:34

, why is this a big deal ? Do you should

16:36

I can I throw you under the bus , or should I

16:39

just take a crack and then you can correct me ?

16:42

um , it's , uh

16:44

, it's a big deal because it's probably

16:47

should be a

16:49

performance enhancer the

16:52

free threaded , or probably both . Yeah

16:54

, this should both um , the free threaded is

16:56

. Uh is maybe a bit more specific

16:59

, because python has always had

17:01

a global interpreter lock , which

17:03

means that you're that if you do multi-threading

17:05

, that you're basically limited

17:07

to a single core .

17:10

Yeah , so basically , the global

17:12

interpreter lock basically locks

17:15

objects , right . So if you have a

17:17

value you can think of like a shoebox , right

17:20

, there's something in that shoebox , and

17:22

then if you have multiple processes

17:25

trying to access or change

17:27

what is in that box , then you have an

17:29

issue of what to do . So the lock basically

17:31

says I have that box , no

17:33

one else can touch it . But that also means that

17:36

only one thing can run at a time . So

17:39

Python is known for having

17:42

this issue . Yeah , and

17:44

yeah , like there are workarounds , right , like python's written

17:46

in c .

17:47

Some people write some stuff in other languages that don't

17:49

have this limitation , and but it's not really an issue

17:51

like it's a certain design

17:53

choice which allows that to be a very efficient garbage

17:55

collection and yeah , that's true , that's , um

17:57

, but because of this , indeed , like

17:59

, like , you had this where you can

18:02

, you can discuss whether or not that was true

18:04

threading or not , um , or that

18:06

it was actually single threaded . But , uh

18:08

, when you use multi-threading , it was , it

18:10

was scoped to a single core and now , with free

18:13

threading , you can basically thread across multiple cores . Yes

18:15

, um , and you had ways around this

18:17

with multi-processing and stuff like this , but , uh , this

18:19

is the first time , um , that we can do this

18:22

. That should , uh , especially

18:24

for multi-threaded stuff , it should speed stuff up and

18:27

the , the multi-processing get you mentioned .

18:29

Basically you start different Python processes

18:31

, yeah , so basically each

18:33

process will have its lock , but then

18:35

you just run it in parallel . Basically , exactly

18:38

.

18:39

So this is today an option . It's

18:41

not on by default . You need to enable it and

18:46

it's there . Let's see

18:48

what it gives . Yeah , it gives . Yeah , the big question is

18:51

that there is no guarantees

18:53

for uh , backwards compatibility on

18:55

all the libraries that were not set up , but

18:57

I think the community today has a very good view

19:00

on what the impact will

19:02

be . This is a bit of a let's

19:04

see what it gives . Yeah , the benefits are . Benefits

19:06

are more than downsides ?

19:09

Yeah , indeed , I did hear that

19:11

just because free

19:14

thread and mode is available doesn't mean that your

19:16

code is compatible with it . So

19:19

people need to change their code to keep these

19:21

constraints in mind . And

19:23

I also heard that this is not

19:25

a final thing

19:27

, like it's still experimental , right , but I think

19:30

python has a yearly release

19:32

cycle kind of , so this is still like

19:34

half baked , but it's a bit on purpose

19:36

because they want people to get their hands on early

19:38

. Yeah , file bugs and all these

19:40

things , right , um

19:42

, so , yeah , very cool . I haven't I

19:45

haven't heard a statement , let's

19:47

say , a personal experience with the free thread

19:49

of python , but , uh

19:51

, it looks cool . What about you have you ? Do you know anyone

19:53

that tried this or any first thoughts

19:55

, experiments , disappointments , maybe ?

19:58

I haven't tried it myself , yet and you know anyone that tried

20:00

.

20:00

Do you have you heard any ? Any statements

20:03

? I don't know if a statement is the right word , but , like the for any

20:05

testimonials , um

20:08

, no , no

20:10

, yeah no , yet , aside from the online

20:12

people and

20:15

the just-in-time compiler . What is this , uh

20:17

?

20:19

uh , it does a code optimization just

20:22

in time , and it should make sure that your code

20:24

runs faster for

20:26

certain scenarios . Again , that

20:29

is also

20:31

again not on by default . There

20:33

might be backwards compatibility issues . Let's

20:37

see . Yeah , To

20:40

me these things are most likely . Both

20:42

, by the way , are a bit more suited

20:45

for low-level libraries

20:48

. You're probably not going to get

20:50

much of a performance enhancement from

20:52

your typical use case

20:54

. Your typical use case is not

20:56

a huge optimization algorithm , right ?

20:58

Yeah , yeah , yeah .

20:59

Where this is really key , these last-minute

21:01

enhancements . Yeah , I

21:04

also feel a bit like that If you're building a data pipeline

21:06

, you're probably not going to benefit from this right , but

21:09

maybe the underlying libraries that you use too Indeed

21:11

, but I think that's a bit .

21:12

The Python is a slow language

21:15

, blah , blah blah . But the things that need to

21:17

be fast , people figure out a way already to make it

21:19

fast . I feel Like machine learning is very compute

21:21

hungry , right , but most of

21:23

the libraries underneath underneath they work with c++

21:26

or c , right . So

21:28

, um , yeah , but I

21:30

agree , I mean , I think it's . It's

21:33

not bad , right , like it's not . There's

21:37

no downside . Let's say this

21:39

is experimental , it's an option exactly so

21:41

foster is better indeed , and I

21:43

think it's a bit of an experiment , like people say , like

21:45

let's do this , let's see what gives . Yeah , right , um

21:48

, and yeah , let's see what gives

21:50

. I think typing hints was a bit the same , right ? They just kind

21:52

of put it there and then they all found these like very

21:54

cool use cases for it . So so I'm excited

21:56

, and maybe , uh , just talking about , uh

21:59

, jit compilers , so just in

22:01

time , I think the most famous

22:03

one is a pipe I pipe pi , right

22:05

, um , which

22:07

uh , basically takes your , your code and

22:09

right before it runs it

22:11

will compile to something that is very specific

22:13

. So an example is if you have

22:16

X is one , so

22:18

basically it's a number then the computer

22:20

thinks it needs to allocate certain memory and then afterwards

22:22

it realizes that it's 1.227

22:25

, whatever , and then you have to allocate another place

22:27

. If you can actually scan your code one time

22:29

, you can allocate it once and then it can make it faster

22:31

and all these things . Well , I'm not an expert

22:33

in these things , but that's how I I understand

22:36

them , but , um , so I think it's cool , let's see

22:38

. Let's see what gives as well , because I also think even

22:40

pipe has a lot of trade-offs , right ? So that's what held

22:42

people back from implementing python

22:45

, but uh , now

22:47

it's like an alternative python

22:49

. Yeah , implementation right , indeed

22:52

, indeed , indeed . So yeah , maybe

22:54

I don't want to get too much

22:56

into it , but like python

22:59

, I guess , is like the language syntax , right

23:01

, the way you write code and how you

23:03

understand it . There are different implementations

23:05

that try to comply to this . Pypy

23:08

is one of them , but I'm not sure if it follows

23:10

everything . So I think the core

23:12

is there , but maybe , if I don't know , like

23:14

a Walrus operator , I'm not sure if it's supported

23:17

, right . So there's some different implementations

23:19

in the different languages .

23:20

And I think the typical one that everybody uses

23:22

, if you just get started with Python , is CPython

23:25

, cpython , cpython it's by far the most popular

23:27

and that's written in C .

23:29

But then this PyPy is written in RPython

23:32

, I think , which is something that looks a lot like Python

23:34

itself . There's also a

23:36

Rust implementation of Python . There's a NET

23:38

implementation of Python . There are different ways

23:41

, right ? So

23:45

basically , you write a program that reads a file that

23:47

looks like a Python file and then you , yeah , but that language underneath

23:50

can be different things . Some other

23:52

small things that I came across

23:54

Python 3.13 , what didn't make

23:56

the headlines ? So

23:58

a lot of people made a lot of noise about these things , but

24:01

I thought , well , I'll get to it

24:03

in a bit . Yeah

24:05

, basically there were some changes to the PDB , right

24:08

, which is the debugger thing , right

24:10

? What does PDB

24:12

stand for ? Actually , completely

24:14

forgot , sorry , the

24:16

PDB . I

24:19

know it's for debugging , yeah , but basically there were some

24:21

issues on the REPL that made it nicer

24:23

to work with PDB ShootTail

24:25

, which is something to work with your file systems . There

24:28

was also some fixes there , small concurrencies

24:30

, uh , what I wanted to bring here

24:32

? The new annotation syntax allows

24:34

comprehension , comprehensions and

24:37

lambda , so this is type annotations , okay

24:41

, um , now the

24:44

annotation change that nobody asked for . So if

24:46

, if you go here , you see class name . And

24:49

now in classes you can also add type hints , right

24:52

, so not just functions . And you add it with this brackets

24:55

syntax . Here you

24:59

have the star operator and

25:02

then on the function you actually have a walrus operator thingy . Yeah , actually , I'm not

25:04

sure actually if this is the type hint or is this the type hint

25:06

. Anyways , for

25:09

the people just listening , basically have two classes definition

25:11

nested and then you have a whole bunch of names and a whole

25:14

bunch of stuff with if statements , lambdas

25:16

, walrus operators and all that

25:18

, and apparently this is valid

25:20

python code now in python 313 .

25:23

So and what is the hint that

25:25

it gives me ?

25:26

I have no idea

25:28

, because again , so the

25:32

type hints , they've been relaxed

25:34

to allow comprehensions on Lemtos . Basically

25:36

, and actually the bug ticket is

25:39

exactly this example . They're like , oh , this doesn't

25:41

work . They're like , oh , this is a problem , we

25:43

should fix it . And then they fix it

25:45

. So I thought it was a bit yeah

25:47

.

25:51

I think that's when the but you can basically say if , let's let , because you're just showing an example

25:54

of a class , but if I understand you correctly

25:56

, you can have a function , and that the

25:58

, the output of the function is a , is

26:01

a lambda , I think

26:03

so .

26:03

So a generator , I guess ? Okay

26:06

, I guess , or maybe just to say like , or

26:09

maybe just to say if this , if you give this

26:11

, then I'll give you that , or I'm

26:13

not sure exactly what's the use case , sure , when I would use

26:15

it because I don't know , that's not really

26:17

a type right like , it's something that generates

26:19

yeah , to be honest , I'm not

26:22

sure either . Um , I also came across

26:24

this because of this reddit post as well , and I think they

26:26

also mentioned I don't think I have any news

26:28

for this except typefire , which is pretty sweet

26:30

, but it's a pity that a person read blah . But

26:33

I think it's also a good example of , okay

26:38

, it starts to add more things to your brain , right , like

26:40

to understand this . You're

26:43

probably gonna have more of a headache than if you just didn't

26:45

add types at all , right , but

26:47

uh , yeah , they're making it pretty , pretty

26:49

flexible , so hope it doesn't get as

26:51

far . I don't see this in any code base , but

26:54

it's . It's something that is there today

26:56

, all righty

26:58

, um , maybe

27:00

something quickly as well . That I also saw

27:03

talking about python 3.13

27:05

. We talked about uv a

27:07

lot in the past . Um

27:09

, I saw this

27:11

on linkedin . So

27:14

sebastian ramirez , the guy from fast api

27:16

, um , he mentioned now that uv

27:18

support dependency groups . But what I wanted

27:20

to just highlight is that the , the pep , was accepted

27:23

on the 10th of october and 16

27:25

days afterwards it was available on uv

27:27

. So I

27:31

guess it's like . My

27:33

first thought is like uv is really trying to be

27:35

almost a

27:38

synonym to python standards , right

27:40

? So everything that gets accepted , they

27:43

will make a push to to

27:45

add it there , um , which

27:47

actually I think is a it's . I think it's

27:49

a nice way to go , in a way , you know , like

27:51

if you become the Python standard , then

27:53

I think you're pretty like

27:56

, you're not , like no one's going to accuse you for being too

27:58

opinionated , right , because you're just following the

28:00

community guidelines . Really .

28:02

And what are dependency groups ?

28:04

There's a definition . So I think it's like if

28:06

you have , normally you have the dependencies

28:08

and you have the dev dependencies , but then you

28:10

can have something like test dependencies . You

28:12

can have something like it

28:14

was already there right , but for uv I

28:17

don't think so . And there

28:19

was no standard . So , for example , poetry implemented

28:21

it . Okay , there was no standard there was no

28:23

standard exactly , no , exactly

28:26

finalized .

28:26

So the pep Okay , because

28:29

this

28:31

already existed , then If

28:34

this library with the Postgres

28:37

backend , for example , and instead

28:39

of the Postgres I want to use the DB backend , exactly

28:41

. And now there is a formal

28:44

definition of DB .

28:45

So I think it existed already

28:47

. Indeed , many tools implemented because there

28:49

was a need , but the community didn't agree

28:51

on what to do , right , maybe

28:54

something on that line , and I feel like I'm hijacking

28:56

the whole pod . We'll

28:58

get back to your topics

29:04

. For example , are these log files right

29:06

, requirements of txt and all these things , um

29:09

? But turns out that actually there's no convention

29:11

about this right . Even requirements of txt

29:13

is something that someone just did and everyone just

29:15

kind of did it as well , but it's not a pep

29:18

. It's not like something quote-unquote accepted

29:20

by python , right , um

29:23

? One thing that it was shared in our slack

29:25

, it was . It was a while ago as well . That was was shared

29:28

. There was another PEP and actually let's see if this

29:30

is . It

29:32

was already rejected . File

29:35

format to list dependencies for reproducibility

29:37

for an application . So basically a log file , right

29:40

? Is

29:42

this the one actually superseded

29:45

by no ? Maybe this one yeah

29:47

it's a draft . Basically

29:49

, there's a lot of attempts to make this

29:51

log file well

29:55

, a standard for log file , and

29:57

I think that's the biggest beef I have with UV

29:59

, because they do have a log file but

30:02

that if you use the UV log file

30:05

, you're stuck with UV . If you read the poetry

30:07

, then you're stuck with poetry . If you're stuck with Rhino because read the poetry , then you're stuck

30:09

with poetry . If you're stuck with rhino because it's requirements

30:11

of txt standard , right . But I

30:13

do feel like if you've , if there is a log

30:15

file format and uv adopts it and

30:18

other tools start to adopt it , then I would have no

30:20

reason to not go for uv , for example . That's , that's

30:23

the only , but it's not even uv's fault in

30:25

a way , right , um

30:27

, so yeah , there's a lot of tries for this and

30:30

one thing I thought it was funny in a way is

30:32

that I went through discussions

30:34

, the Python discussions , and they

30:36

actually reached out to Poetry to talk

30:38

about this proposal and Poetry already

30:40

said from the beginning like we're not going to support this because

30:42

the way we've set up our

30:45

tool is too different and we cannot make any changes now

30:47

. So

30:49

, poetry , I think it got really popular because it was one of the first ones , but at the same time

30:52

I think it made it very hard to change because

30:54

they built on top of that and

30:56

I do feel like Poetry is being left

30:59

behind more and more because they don't comply

31:01

to the standards , like the PyProjectoml from

31:03

Poetry . It's not standard , uh

31:05

lock file . They wouldn't be able to adopt it right

31:07

.

31:08

so well

31:11

, it's a , it's a design choice , right , like they want

31:13

to be to remain some somewhat

31:16

backwards compatible . Yes

31:18

, I mean they can say fuck everything . Like from

31:20

now on we do it a different way , right , like they could say that's

31:23

true , like there's also some value

31:25

in the maturity , that there is

31:27

stable stability , right , yeah

31:29

. But I also , most likely

31:31

, if you use uv today for a production project

31:34

and you look and you want to upgrade

31:36

it , uh , next year there's

31:38

probably not going to be an easy upgrade part yeah

31:41

that's true poetry probably will have .

31:44

Yes , that's true , but at the same time . So I

31:46

mean , I fully agree .

31:48

The only thing I have a bit of , and the thing is like

31:50

none of these things matter when it's your personal

31:52

pet project , right ? Yeah , you don't care

31:54

about these things . Yeah , from the moment

31:56

that you build something for a production

31:58

environment , you want to have some stability

32:01

in terms of years , not

32:03

weeks yeah , I think the the

32:05

issue for me is when you take that a bit too far

32:08

.

32:08

So I do think there's some changes . I do think poetry

32:10

should have had a breaking change by now . The

32:13

reason why I say this uh , like

32:15

poetry , they used to have like dependencies and dev dependencies

32:18

and then they also implemented group dependencies , but

32:20

then dev dependencies became a group and

32:23

then I see a lot of pyprojecttomo

32:26

that has two poetry sections one for dash

32:28

dev , dev , dash

32:31

dependencies , and then one dot groups

32:33

, dot dev . Yeah

32:35

, and that's because , like , people update the tool in

32:37

the middle , but like the actual like , I

32:40

guess for me the thing is like they buy , they

32:42

don't want to make it a breaking change , so

32:44

they just kind of keep adding stuff to it .

32:46

But I'm like , I'm not even debating that poetry is good

32:48

or bad . He had

32:51

another packaging tool . But

32:53

there are two sides to this coin and to being fast

32:56

and agile A Pepper's release

32:58

and 16 days later it's

33:00

a new thing . You can question

33:02

how good were these reviews ? How sure

33:04

are we of this implementation ? How clear

33:07

was this implementation ? Is this something that you can do one-to-one

33:09

, or do you need to have some discussions about

33:11

how the standard is actually implemented

33:14

, like there's pros

33:16

and cons to being very fast versus very

33:18

stable ?

33:19

No , that's true . That's true , I guess . For me

33:21

, my main point was I

33:24

think it's good to have backwards compatibility

33:26

, but I also think you

33:28

shouldn't . I think it's good to have backwards compatibility

33:30

, but I also think you shouldn't . I think there should be a limit to it , right ? Like , if you want to change the API , there

33:33

should be a breaking change , and you shouldn't just support

33:35

two different versions of the same API , just

33:37

so you don't make things breaking changes , right

33:40

. So ? But I agree , it's

33:42

always a trade-off . It's

33:45

always a trade-off . What

33:49

else ? What else is new ? Maybe back to the

33:51

ai stuff , bart ?

33:54

what the ice stuff ?

33:54

metal yama training to

33:56

be bigger than ever

33:59

. Is that what you said ?

34:01

oh yeah , there was an article on uh on

34:03

wired the last week

34:05

yes that um said

34:07

. Actually Meta today has the biggest

34:10

, biggest server

34:14

park not sure if that actually directly translates

34:16

to a server park , but at

34:19

least the most amount of resources to

34:21

do training of their Lama model versus

34:23

their competitors , like the most notable , of course

34:26

, being OpenAI . So the resource that Meta today has available is the most notable , of course

34:28

, being open the eye . But so the resource that , uh , the method today as available

34:30

is , as is the most

34:32

um of all the big ones , which

34:36

um questions

34:39

a bit like what ? How does the ecosystem look like

34:42

? I think a year ago everybody

34:45

thought llama was very cool but

34:47

no one really took it really seriously as a competitor

34:50

. Today the performance has become

34:52

really good . You

34:55

see it being used in

34:57

the industry , in the community , a lot because it's

34:59

way more open , you

35:02

can build upon on top of it . So

35:07

I'm wondering what Meta's position

35:09

will be like two years from now I'm also wondering

35:11

what's the end game here ?

35:13

because meta , are

35:15

they profiting from these models , and how

35:17

much ? Because

35:19

the , the models , are open

35:22

source . Well , maybe we can talk a bit about that

35:24

in a bit . But , like the , the weights are available

35:26

and I think , uh , I

35:29

think the only restraint is that , as long as you're

35:31

not competing with meta , you can use them

35:33

for whatever you want , kind of uh

35:37

, well , there are some limitations on there , uh

35:39

, in terms of number of users and stuff like

35:41

that okay , yeah , but it's it's very

35:43

permissive . It's very permissive , yeah

35:45

and um , so

35:48

I'm assuming they don't make a lot of money

35:50

from these models , but if they

35:53

have the largest cpu , gpu

35:55

, rack I'm there's

35:57

probably a lot of investment right , and

36:00

is it just to train the models or is it to also

36:03

use it in the meta products ? Probably

36:05

both right , yeah , probably both I

36:07

like .

36:08

It's . For me it's hard to like they . They

36:10

have a lot of features . I think there is now a model

36:12

. I don't think it's available in europe actually , but in whatsapp

36:14

as well there's . Uh , there is a

36:16

an ai model available . Um

36:18

they we don't know how it

36:21

is being used behind the scenes , for example , to optimize

36:23

ads , stuff like that we don't know . Yeah , meta

36:25

is very much driven on ads as well . Um

36:28

, with

36:30

all the resources they have available , I could see

36:32

them at some point providing a service yeah

36:36

, and a service for their lm . Um

36:39

, let's see , they're

36:41

still also very much on on the

36:44

uh augmented reality

36:46

side .

36:47

I would see something like this generative

36:50

ai also playing a role there yeah

36:53

true , so let's

36:55

um but I think it's like for

36:57

me , this , this

36:59

news , puts me a bit on the edge of my seat , in

37:01

a way , um , because

37:05

I feel like there's something I'm missing . You know , I

37:07

feel like , yeah , maybe they will release some gen

37:09

AI for , uh , augmented

37:12

reality , or , yeah

37:14

, maybe they'll release a proprietary model soon

37:16

, but there's something that I feel like it's , you

37:18

know , something's in motion that I don't see yet .

37:22

Let's see . Let's see . I think

37:24

it's always a bit difficult to a meta Like , like they . I think it's always a bit

37:26

difficult with meta . They did a lot of the VR stuff

37:29

, the AR

37:31

stuff and they invested more versus

37:34

what the result is today . I think that everybody

37:36

is thinking at the same time . If

37:39

anyone is going to pull off AR in the near

37:41

future , it's probably going to be meta .

37:43

Yeah , what do you think of that ?

37:47

Do

37:50

you think there is a future for AR ? I think so . Yeah

37:52

, I think so . I think the question is like how

37:55

will it look like and what kind of devices

37:57

, what kind of peripherals ? But I

37:59

think it's just waiting for

38:01

it to be available .

38:02

And do you think , and what kind of who's

38:05

the classical user of this ? Do you think it's more

38:07

for entertainment ? Do you think there is a

38:09

business

38:12

in terms of working people to use

38:14

? Because I also saw the Apple Vision . You know they had a

38:16

huge screen on the wall and they had this and they had that

38:18

, but where

38:20

do you see that they are fits there

38:23

? Maybe for you ? What are

38:25

the things that you say ? Ai would be a good use case

38:28

for this thing that I'm doing today um

38:32

AR , you mean yeah , uh .

38:37

I don't see myself using it okay today

38:39

. Yeah , especially not with Apple Vision

38:41

. I think Apple Vision is way too it's clunky

38:43

, you know it's too much in my way yeah

38:46

, yeah , yeah um , I think

38:48

it's very cool if you , if you like these type of gadgets

38:50

yeah , yeah , yeah , it's more . It has a very high gadget

38:52

factor . Um , I think with

38:54

the what's called again the glass

38:56

that were introduced very recently by meta the

38:59

, the . It's like the ray-ban thing , no yeah

39:01

, but also another one , um , which is it's

39:04

more or less comparable to apple

39:06

, uh , to apple's apple

39:08

vision pro orion maybe

39:10

yeah yeah , but much

39:13

more uh , accessible

39:15

, uh . At the same time , you can

39:17

. You can debate , like it's not much , much

39:20

, not very far off , on the google glasses

39:22

that there were 10 years ago yeah , yeah

39:25

, indeed , yeah

39:28

to me , I don't know . I think like yeah

39:30

, so maybe I'm showing here on the screen for people

39:32

just listening the orion , uh

39:35

, announcement right I

39:37

think what it wants you to do is to and today

39:39

, the only way that we really

39:41

have to do this is glasses is to add

39:44

information to real life

39:47

, right ? Yeah , like what you do already

39:49

with with if you have an alexa home

39:51

, if you have a smart home , these type of things , like you

39:53

can interact in a digital way

39:55

with your environment . What we , today

39:57

, do not really have yet is like

39:59

have these augmented things ? Like

40:02

we have notifications , we have

40:04

extra information popping up , yeah

40:06

, or glasses

40:08

the way ? I don't know , do I like them to be the way

40:11

? Probably not , like I think it's a privacy horror

40:13

, but I do think

40:15

we were going in that direction . I'm

40:17

not sure how it will look like . Yeah , I

40:19

think , and I think if there's a company that

40:21

today has a huge amount of knowledge invested

40:24

and capital invested in that , it's made yeah , that's

40:26

true .

40:26

I feel like , if the , if this becomes a big deal

40:28

tomorrow , they are the

40:30

leading one , right , right .

40:31

They even changed the name

40:33

company name to to reflect more a bit this

40:36

If tomorrow they can come up

40:38

with contact lenses that look normal , that you don't

40:40

see , and people wear them . I think

40:42

you will see yeah .

40:43

Yeah , yeah . But I also

40:45

think it's interesting because to me , I always associate

40:47

this with gaming . I

40:49

guess , like . So

40:54

when I think of augmented reality , I think , okay , some gaming stuff , like for

40:56

fun , right , that's also what we see today , like when you have the I mean , it's not

40:58

really augmented reality , but like when you have the vr

41:01

headsets , um

41:03

, it's more for like game-like environments

41:05

, right for vr , yeah , and uh

41:08

, yeah , I think they pretty much

41:10

are . They are advertising this very

41:12

much for the everyday user , for someone

41:15

like business . So in the video

41:17

here there's like someone having a call and like floating

41:20

in the middle of the living room , right .

41:21

But if you could have that with contact lenses .

41:24

But I feel like , even if it's like a sleek glasses , I

41:26

could buy that . I mean these glasses right now for

41:29

people that are just listening .

41:30

They're not sleek eh .

41:36

They're not sleek , they're like there's a google glass were better than this . Yeah , they look like uh

41:38

, there was a . I saw on the saturday night live . I think they were comparing this with the , the minions

41:40

. You know , the banana , you know , you saw that

41:43

. Yeah , it's really funny , um , but

41:45

yeah , they're a bit , they're a bit clunky , but this is much better

41:47

than uh , than what we would see three

41:51

years ago . Would you wear

41:53

something like this , alex , because you wear glasses normally

41:55

?

41:56

No , they're too clunky .

41:57

They're too clunky Okay . But

41:59

if you could have , like you're wearing glasses now this

42:04

functionality would be possible in your glasses , but then

42:06

maybe it's just a matter of time , right , but

42:09

that's what I'm saying . This

42:16

is not how it's going to look , look like , but I think , if anyone is ready to do this , it's meta

42:18

, yeah , yeah , but I do think , yeah , because , alex , glasses for people that are just well , it's not on the

42:20

screen , but like you're just uh like a thin

42:22

frame , right , and I think that's the the

42:24

main difference , right , like , uh , the

42:27

frame of the disc glasses are super , super thick . So

42:29

, but yeah , yeah

42:32

, true , true , I guess , uh , to

42:35

be seen what happens I think also with contact

42:38

lenses , would you see where your eyes

42:40

are looking .

42:40

Would it look ?

42:42

weird , it's

42:44

a good question .

42:45

Well , it doesn't exist in contact lens today , so no , but

42:47

I guess because imagine

42:50

the , the contact lenses .

42:51

I guess , like it follows your eyes right like you get stuck

42:53

to your . So I guess it's like because

42:56

right now on the image you saw like oh , the top left

42:58

there's this and the bottom there's this , and

43:00

you're looking here and you're doing there , but I guess if it just

43:02

follows your eyes it will always be fixed right to

43:04

in your view pane . I

43:06

guess it's hard to describe right like what

43:09

you look .

43:10

It would always be the top quarter , this will ever be possible

43:12

, a contact lens , but I think with the orion

43:14

, if I'm not mistaken . I think it projects it on your

43:16

eyes actually , so it's only virtual

43:19

that you're looking in a , and that's just

43:21

for where your eyes look to , how to , to how

43:23

it gets rejected is it ?

43:25

I'm wondering if this is okay for the eyesight if

43:28

you just have stuff projecting your eyes well

43:30

, that's what's happening all day long , I guess yeah

43:33

, but like photons , did you always hear like , oh

43:35

yeah , don't , don't watch tv in the dark because it's bad

43:37

for your eyesight and this and that ? But

43:39

is it that's what ? That's what I heard growing up ? I'm

43:41

not . Uh , that's what our mom stole this

43:43

yeah , right , yeah . So if your

43:46

mom , if you're listening , did you ? Did

43:48

you lie to me , mom ? Okay

43:52

, to be seen , to be seen , I think yeah

43:55

could be very exciting , though I

43:57

could buy into it if it

44:00

didn't look clunky indeed . Maybe

44:02

more on the AI tech , but

44:05

this is from OpenAI , actually not from Meta . I

44:08

saw this . This is experimental . It's called Swarm . Have

44:10

you ever did you come across this at all ? Part no

44:12

, so experimental , slash , educational

44:15

, so I guess not something for to

44:19

be picked up tomorrow , uh , but uh

44:22

, basically it's . Uh , how do you say orchestrating

44:25

agents , handoffs and routines ?

44:26

so basically , oh , it's the agents

44:28

talking against each other .

44:29

Yeah , yeah so this is from

44:32

OpenAI , so I think that's why I made a lot of noise . Openai

44:35

released a package which is not

44:37

just a wrapper around your models . Basically

44:40

, it's so you can create multiple agents

44:42

with different prompts

44:45

and then you can have them interacting

44:47

with each other . So this is supposed to help that . So

44:51

yeah , for example , the

44:54

example they have here , the , is like what's the weather , new york ? And then

44:56

there's a triage system and assistant

44:59

, and then there's

45:01

a function that says , well , transfer to weather assistant , and then

45:03

you'll go to the weather assistant and then the weather assistant

45:05

has a different set of prompts

45:07

and expertise and functions and all these things

45:09

, and it goes to 67 degrees . So

45:11

I guess you can also think if you had a bot in

45:14

the DataRoot's webpage , for example , and

45:16

then the first step is to know what

45:19

kind of question are you asking , so classifying what

45:21

these intents are , and then this is a

45:23

way to do it . So I think it's something

45:25

really cool . I think it's something that is

45:28

much needed , but

45:30

I wouldn't use the they just because they are

45:32

saying that it's experimental and educational , right , so

45:34

maybe just something for fun , but you did

45:36

see something like this .

45:39

I think I read this on Reddit . There

45:41

was a discussion on this .

45:43

Do you know if there are any other quote-unquote competitors

45:45

for this or no ?

45:52

I think the question is a bit about what is the definition of multi-agent

45:54

orchestration . There are definitely multiple frameworks

45:57

that do multi-agent yeah , that's true . I

46:00

think the interaction between those agents is a bit

46:02

the what the discussion is about

46:04

true

46:08

, true , true .

46:09

So just something I wanted to share , because one wise

46:11

man once said that a week keeps the

46:13

mind at peak , um

46:16

, and then , uh , maybe more

46:18

on the ai , or

46:21

maybe rag , to be more specific , pg

46:23

, p , pg rag , uh

46:26

, or pig rag , yeah

46:28

pg rag .

46:29

This this was released

46:31

by Neon , the

46:34

company behind the

46:36

managed Postgres database , with Serverless

46:39

right , that's the big Exactly , yeah , serverless

46:41

. So they have a way to separate out

46:43

storage and compute for Postgres . I

46:46

use it a lot actually , neon , you like it , I

46:48

like it , and

46:50

the reason I know of this is that they sent an update mail

46:52

, I think last week , where they said that

46:55

they launched PG Rack , the PG Rack extension

46:57

, and it's a Postgres

46:59

extension and it allows you to do big

47:03

parts of the Rack pipeline within

47:06

Postgres . So

47:10

you already have some of these things . I think the most notable

47:12

is pg vector . Yeah , but it's just

47:14

a vector database , it's just a way to

47:16

hold vectors , um with pg

47:18

rec . What you can actually do is that you can , uh

47:20

, you can , convert text

47:23

like html , like markdown

47:25

, like these type of things , to regular text . You

47:27

can chunk it , um

47:30

, and then you can . Also , there is a local

47:32

embedding model that they

47:34

use , so you can immediately from that embedded

47:37

in a vector space that

47:39

you can then query , and all

47:42

that in your postgres database so

47:44

it takes a step further from just a vector database

47:46

. It also process documents

47:48

and chunks them exactly and retrieves them based

47:51

on Exactly , yeah when , with just

47:53

the vector database or

47:55

just vector storage , if

47:59

you ? Simplify it like that it's just the storage of those vectors and the querying

48:01

of those vectors . Wow , this is also going to everything

48:05

up to that storage , all from Postgres

48:08

, which is interesting

48:11

.

48:11

Yeah , interesting . Also

48:14

, they're showing some examples on how to use it . You can literally

48:16

just put the prompt here , the query , right

48:18

, what is dot dot dot ? How does it work ? And then you

48:20

can actually , in your select statement , embed

48:24

it for query and then you put this in the query . That's how

48:26

Imagine works . So it

48:29

looks pretty cool . You haven't tried this . I'm assuming I

48:32

haven't tried this . I'm assuming I haven't tried it . No , no , and neon is it

48:34

like ? Is this anyone with the like who can

48:37

use this ? Only if you're using neon , only

48:39

if you're using postgres if you're using

48:41

using postgres . Yeah , this is like a regular postgres

48:43

extension okay , because it's on the neon

48:45

docs , but since open source , I guess it's

48:47

open source . Yeah , is neon open source as well ?

48:50

uh , it's a good question . I know that it's open source

48:52

. Yeah , is Neo open source as well ? That's a good question . I

48:54

know that they have open source parts of

48:56

their implementation . I'm not sure if everything

48:58

is open source , to be honest , or maybe

49:00

are they source open ? Which

49:02

is a bit , I think they're actually open source , but

49:04

I'm not really educated

49:07

on this . It's been a long time

49:09

. I looked into

49:12

it a bit when it was just released . That's why I'm quite sure that big parts

49:14

of it are open source . But really cool

49:16

.

49:16

I'm not sure of the latest status uh , and

49:18

maybe , uh , yeah , I mentioned open

49:20

source versus source open . Um

49:22

, I guess the main

49:24

difference , that way I see and maybe correct me if I'm wrong

49:26

here open source is something that is community

49:29

driven right , so people can contribute

49:31

with code and then it can

49:33

get accepted and can get incorporated in their

49:35

product . Source open is

49:37

more like the code is there but

49:39

you can see , but it's

49:41

not like you're not going to interact with it necessarily . You

49:43

can report bugs , but who

49:46

builds it and maintains it is more the

49:48

. Is that a good ? I think it's a bit of a simplification

49:50

. I think source available just means that you can maintains is more the . Is that is that a good ?

49:51

uh , I think it's a bit of a simplification , I think source available just means that

49:53

you can look at the source , yeah , and

49:56

maybe you can do other stuff with it . Open source

49:58

is a definition , um

50:01

, it really depends on the license

50:03

that you're using , and it's maybe a good

50:05

segue uh into the open

50:07

source initiative , which is , I think , the most

50:10

well-known organization that has has

50:13

built a number of these uh of

50:15

these open source licenses and

50:18

manages them , has discourse on

50:20

them to to let it evolve , and

50:22

last week they actually uh , after , I

50:24

think , more than a year of uh of

50:26

uh research , they uh

50:29

released version 1.0

50:31

of the open source ai license

50:33

so maybe , why

50:35

is this a big deal and why is this different

50:37

from just regular open source ?

50:39

what's the tricky thing ?

50:41

regular open source . What we're typically talking about

50:43

is source code , so just

50:45

written lines

50:47

of source code , which is different

50:50

from ai models , because an ai model typically

50:53

has it has source code , it

50:55

also has data , it has trained weights

50:57

. Um , it's

50:59

much more , let's

51:01

say , just with with just the source code , you

51:03

can't really talk and speak about a model because

51:06

it needs to be trained , etc . Yeah , and

51:08

there is a lot of chatter in the in the community

51:10

about what our open source models and whatnot . I

51:13

think Lama very much states that

51:15

it's open source .

51:17

But I think they play a bit because , again , this

51:19

is the first , as I know , the first

51:21

open source AI definition , exactly

51:24

, and I think people before , because there was no third

51:27

party that was saying what is open source was

51:29

not .

51:29

It was a bit of like anyone

51:37

can say it's open source because yeah , exactly , yeah , exactly . Well , and to be honest

51:39

, like it's not . That uh , osi , the open source need should have as really like a legal standing and

51:41

like , if you , if you use the definition wrong , you're

51:43

gonna , we're gonna sue you . They don't do this

51:45

right , but they , they try

51:47

to uh create a community-wide

51:51

consensus about what is open source and it has

51:53

been used .

51:55

These open source licenses have been used in legal

51:57

settings as well .

51:59

Yeah , yeah , that definitely has , so it

52:01

has like a legal .

52:02

it is a legal instrument as well .

52:03

It's like it's not just From the moment that you use a certain

52:05

license , there is a legal

52:08

meaning to it .

52:10

Yes .

52:10

But it's not because I say what

52:13

I do is open source yeah , yeah , yeah that

52:16

you can say I have it or not . And from the moment that I

52:18

say my code is GPL , yeah

52:20

, and I do something that is not GPL , yeah , that is

52:22

not okay , right .

52:23

Indeed so . But just to say , like the , like

52:25

these licenses they

52:36

show , it's like there is a very legal , practical , exactly . There's an implication , there's a consequences

52:38

. It has to be thought through like the , the , even the hashi corp that we talked , like the open tofu hashi corp , they , they changed licenses

52:41

in the way that the stores were still available , but they're restricting some things and there was like

52:43

very big legal implications from

52:45

the moment that you , let's say , implement the licenses

52:47

, which is what hashi corp did I think it was was g , it

52:49

, gpl . I'm not sure .

52:51

Yeah , I'm not sure , but from

52:54

the moment that you implement it , there is , there is an implication

52:57

, like it has implications , right , but

52:59

from the moment that you just like you don't care , but

53:05

, you just . So

53:08

this is uh , this is interesting in the sense that

53:10

, uh , I think , when just browsing

53:13

through this , I don't think llama is open source can

53:17

we have the hot , hot , hot hot ? I don't know if it's really

53:19

hot

53:24

, but , um , I think , uh

53:26

, the uh , the

53:31

interesting things about this is that it's

53:33

it says also

53:35

way on how to . So you

53:37

don't need to . You need to publish how

53:40

you trained it . I read that

53:42

as the source code . Okay

53:44

, you need to . You need to be

53:46

very transparent on what data

53:49

you used , how you acquired the

53:51

data . If there are ways

53:54

to acquire the data , either for free or

53:56

paid , okay , so

53:58

that , even if you don't have access to the data

54:00

set , that you could reproduce it

54:02

with the right amount of efforts and

54:09

that typically , like these open source models , they

54:11

have also their weights are included

54:14

.

54:14

Yeah , what they mentioned here is parameters , right .

54:16

Yeah , um , there

54:21

are some the definition and I'm don't have it exactly

54:23

fresh in my my my head , but the

54:25

it says that users can

54:28

freely use and reuse

54:30

this without specific , uh

54:32

, limitations . I think the

54:34

limitation that Lama defines

54:37

and again I don't know exactly how

54:39

it is anymore , but it says something about from

54:41

a certain amount of users , you can't

54:44

use this anymore or you need to inquire

54:46

. That is very

54:48

bit of arbitrary . That

54:50

is not something like . You need to show

54:53

that you use this as a base right .

54:54

Yeah .

54:55

Then still everybody can use this . Yeah , but this

54:57

is very arbitrary , like from that moment on what , you can't

54:59

use this anymore or you need to ask for permission

55:01

. Yeah , I see I think that falls out

55:03

of the scope of this open source definition

55:06

.

55:08

This is really cool . I

55:17

think it's much needed . I also think you mentioned that they've been for one year . You said they've

55:19

been uh , something like that , yeah , which I also think is good . I think if you have

55:21

something like this , like this is uh well , I think you need to take time

55:24

to to study and see how it is and

55:26

release before releasing something

55:28

right , this is kind of like it is

55:30

a one-way door in a way . Right , like , once

55:32

you release there , you create expectations

55:34

and you have this to kind of go back or modify . This

55:37

may not be as simple , because people are going to build

55:39

on top of this . So I also think it's it's

55:41

nice to hear that they it was a well

55:43

thought , yeah , you know , decision , and

55:46

I do think it's much needed , right . Like you said , we we

55:48

see a lot of stuff from models and all these things . I think

55:50

there are a lot of people that want to do as

55:52

well , but they don't know exactly what to do . Um

55:54

, how to say like , yeah , this is proper

55:56

open source and this and that kind of agree .

55:58

So , even if this is a proper , proper

56:00

way to educate the community on what

56:02

is indeed , and what not

56:04

?

56:04

and also why not , indeed , indeed

56:07

and I think , even if it's , even if people don't

56:09

fully agree , at least it gives

56:11

a common ground to discuss these things Exactly and

56:13

that you can evolve from this . Indeed

56:16

, indeed , really cool , really

56:18

cool . Maybe . I also saw here that we

56:21

can endorse the open source AI definition . So

56:25

look at that . We have quite a lot of companies

56:27

here . Probable . This

56:30

is the company now is behind Scikit-Learn

56:32

, ricardo Libre , big

56:35

retailer in South America , bloomberg

56:37

, mozilla . Really

56:41

cool , really cool . All

56:44

right , and now maybe to something a bit lighter

56:48

topic , maybe still

56:50

in the food for thought .

56:56

A lighter topic maybe um

56:59

, still in the food for thought . Someone used v0 by vercel and has some

57:01

thoughts to share . Uh , yeah , just think

57:03

it's cool . Okay , all right , thanks everyone . V0 by vercel is

57:05

uh , it's like their geni , like a

57:08

geni tool by vercel that allows you to

57:10

, with prompts , it generates

57:12

basically web sort

57:14

of web components for you . Yeah , I

57:18

think it leans very much towards React

57:20

components and

57:23

I actually , to be honest , I thought that you can only

57:25

do React components . Maybe I can share

57:27

my screen actually . Okay , there we go , I'll let

57:29

you do it if that's possible . I

57:32

thought you could only do React components and

57:34

I was playing with this over the weekend and you can actually

57:36

, but it's less performant . Like you can also say no

57:40

, this UI find the React

57:42

component looks cool , but please implement this with

57:45

HTMX .

57:48

And it does that to some extent as well , but it's probably because it probably uses

57:50

a base model that has some knowledge of HTMX , but it's

57:52

probably fine-tuned on the . Because Vercel

57:55

is the company behind Nextjs . Yeah

57:57

, which is also like . I

58:00

don't know if they're built on top of React . It's

58:03

built on top of React . Built on top of React .

58:05

So we're looking at a page now and it says

58:07

what can I help you ship and maybe

58:12

generate a

58:15

login page specifically

58:18

geared towards

58:21

a Brazilian

58:23

guy , you were going to say this . I was just

58:25

waiting , called Murilo , make

58:34

it super fancy and sleek . I

58:37

haven't tried this yet , so Is this free

58:40

or paid ? Well , what I'm using now is free . Okay

58:42

, I haven't like Brazilian

58:48

login , so it's generating . Bem-vindo

58:50

, faça login para continuar . Can we get an applause ? So it's generating

58:52

. You see , benvido Murilo Fasa login

58:55

para continuar .

58:55

Can we get an applause ? That was pretty good .

59:01

Yes , and then the email already

59:03

holds like a placeholder muriloexemplarcom

59:06

. There's a password place . There's an enter

59:08

button . How do you know it's a password ? It's

59:10

an enter button . How do you ?

59:12

know it's a password , it's just a senha . How

59:14

do you know what that means ?

59:16

It's I get it from the context .

59:20

It's very yellow with green .

59:22

Yeah , yeah , yeah .

59:24

It looks very Brazilian . Right , it looks very Brazilian

59:26

and the Portuguese is perfect . Bem-vindo

59:29

, faça login .

59:30

And if you look just as the aesthetics

59:32

, it looks really nice , right yeah ?

59:33

the aesthetics , but it does look very . I

59:36

mean , I guess it's very modern yeah yeah , yeah , that's what

59:38

. I was going to say . I feel like the style is very

59:40

much like Vercel , which is the modern style

59:42

. Right , like that's how you see the borders

59:44

are round , there's a bit of shadow

59:46

, like on the background . Right

59:49

Like the font , it looks sleek , it

59:51

looks really cool . I think it also it's TypeScript

59:53

.

59:54

It's TypeScript . Well , again , this

59:56

is a TSX file so it's TypeScript

59:58

, but if I ask it to just

1:00:00

do it in JSX , it will do it .

1:00:02

It will do it , I think , if you look

1:00:04

at the underlying code , you would say yeah .

1:00:07

If you look at the underlying code and the locations

1:00:10

of that stuff , it looks a bit else to

1:00:12

be trained on Shed's CDN , which

1:00:17

is a typical , very modern , sleek-looking component

1:00:20

library and it has the same bit of principles , right

1:00:22

Like the Shed's CDN .

1:00:24

from what I remember is a bit like that copy-paste

1:00:26

philosophy which is kind of what they're encouraging

1:00:28

you to do here .

1:00:29

I'll make it even better , make it even

1:00:31

more sleek

1:00:33

brazilian . Wow , it's gonna be

1:00:35

like samba football like

1:00:37

the brazilian times

1:00:40

10 , let's

1:00:43

see what it is and now it starts

1:00:45

making this very brazilian let's

1:00:48

see what it does first .

1:00:51

But uh , one thing I also like I think is

1:00:53

interesting like the , the web page is

1:00:55

in portuguese , yeah , but , um

1:00:57

, the logging is all in english as well . So

1:00:59

I also , because one time I was writing something for

1:01:02

like portuguese , and

1:01:04

hola morello , ah

1:01:11

, this is so funny . You know , jogadores , no

1:01:13

, it's like a football player , like it's

1:01:15

just player , right , but like e-mail , do jogador is

1:01:17

like . And

1:01:19

then murilo at canarinho canarinho

1:01:22

is like a canary , which is like a bird , which is the

1:01:24

mascot of the brazilian national football

1:01:26

team . So it's like and

1:01:28

then it says in the bottom novo , no samba

1:01:30

.

1:01:31

Like new at samba , join us

1:01:33

at blah blah blah , there's actually a palm tree , but you don't see

1:01:35

it . With the overlay , we can hold

1:01:37

on .

1:01:37

We can change this . Can we take

1:01:39

this ? Hold on

1:01:41

we can do this .

1:01:43

Let's remove this and

1:01:46

there's a small palm tree . Yeah , jumping .

1:01:49

There's a bit of sun on the top left , so if you do

1:01:51

something .

1:01:52

Brazilian time 10 times 10 , you

1:01:54

end up with football players . Yes it's

1:01:57

no yeah that's the conclusion of today it

1:01:59

just becomes more stereotypical .

1:02:00

Right , it talks about samba about

1:02:04

football just to dance , you know ? So

1:02:07

, yeah , so it becomes more stereotypical

1:02:09

, but yeah .

1:02:11

You want to one-up this ?

1:02:13

Let's see how racist you get .

1:02:14

Make it a thousand times

1:02:17

more Brazilian

1:02:19

. I

1:02:25

wonder if , at one point , you would just say but would

1:02:27

you be more comfortable filling this in versus a regular

1:02:30

login form ? For sure , Credit

1:02:32

card yeah , I'll do it all If you do A-B testing .

1:02:34

Yeah for sure Interesting . I

1:02:36

like how they said import coffee , sun , palm

1:02:38

tree , music , umbrella , flag feather , so

1:02:42

you can already tell what it's going to be . Oh

1:02:44

wow , it's

1:02:46

a lot of animations .

1:02:50

Yeah , they have a lot of animations also

1:02:52

I see they

1:02:55

went with carnival .

1:02:55

Now what's the feather thing does ? It doesn't mean

1:02:58

anything , I'm not sure actually . Ah

1:03:00

, email the sambista . So usually you dance samba

1:03:02

at carnival . So that's why , like within the

1:03:04

carnival , sambista folia

1:03:07

is all like kaina folia . Wow , this is

1:03:09

a novo , no bloco . Oh

1:03:11

, this is funny . So it means like new at the block , because

1:03:13

usually in carnival you go like in little

1:03:15

blocks , you know . So it's like this

1:03:18

is this is great .

1:03:19

There's a football , football

1:03:22

in the background

1:03:24

as well but aside from I mean aside

1:03:26

from the Brazilians , it looks good , right , it

1:03:28

looks good like with minimal effort it looks good yeah

1:03:30

that's what I really liked .

1:03:31

I didn't expect it to be so mature

1:03:34

in terms of uh this

1:03:36

almost makes me feel like

1:03:39

I could do something like this , right . But

1:03:41

I'm also wondering how , because I guess like

1:03:43

yeah , it's the same thing when you take a template and try

1:03:45

to modify it , right like as

1:03:47

soon as you try to get the two things to interact .

1:03:49

I don't know how easy it will be to

1:03:51

put everything together , to me the challenge with

1:03:53

Gen AI when it comes to visual stuff and

1:03:56

it would be a good test to do this here as well but

1:03:58

especially with images and videos and stuff , it's like

1:04:00

to generate something in the same style

1:04:02

, so you have this consistency . I think that

1:04:04

is very hard If you now say , do a login page

1:04:06

and then do a homepage and then do that page a

1:04:13

home page and then do that page .

1:04:14

What did ?

1:04:14

you pass in as a input . Can you like say well with images and maybe with with something

1:04:16

like v0 . It would be easier , because a lot of this visual aspect

1:04:18

is expressed in code exactly , so it's easier

1:04:20

to say like following this , this style

1:04:22

probably . Game

1:04:27

style make a calendar

1:04:30

application

1:04:35

. Let's try

1:04:37

it . I

1:04:39

hope I don't reach my limits , my

1:04:42

credit limit .

1:04:43

But it's good , it's also free . You said right .

1:04:45

Well , what you're seeing now is all free

1:04:48

and it's super easy to use . It's

1:04:50

like literally just copy this , this tsx

1:04:52

file , here , um and when , when

1:04:54

it it also generates . When I , when I was doing

1:04:56

it over the weekend with hd mix , it generates

1:04:58

all the files , so it also generates

1:05:00

multiple files that you can all copy

1:05:02

paste . This is really cool yeah

1:05:05

, um now we see the calendar

1:05:07

page . The same type of looks very similar

1:05:09

, same type of . Also a lot of animations

1:05:12

, the same background very brazilian

1:05:14

from

1:05:16

the yeah it's actually quite , uh , quite consistent

1:05:18

styling oh wow , look at that .

1:05:20

They have the , the . It

1:05:22

looks great , huh , cool . Yeah

1:05:24

. Well , I'm not sure I'll make the same stylistic

1:05:27

choices , but but like

1:05:29

it's very impressive as a as a product , huh

1:05:31

yeah so , yeah , I was

1:05:33

excited about it when I tried it . No , this is really

1:05:35

cool . This is really cool . Maybe I'll well

1:05:37

, I don't do a lot of things like this , but , uh

1:05:39

, I ever have a use . Maybe I'll give it a try , see how

1:05:41

far I can get I'll

1:05:44

stop the sharing yeah , do you think

1:05:46

there's a danger of doing things like this because , uh

1:05:48

, you don't understand the code necessarily .

1:05:50

And then I think not more than any other

1:05:52

, ai supported code generation

1:05:55

yeah to be honest okay no

1:05:58

, but this is uh , this is really

1:06:00

cool .

1:06:01

I gotta I'm gonna jump on that chip I think

1:06:03

the possibility even , but like you said

1:06:05

, it's not specifically with uis like

1:06:07

you can generate very

1:06:09

shitty functional code now without

1:06:11

knowing how you're doing it and without being opinionated

1:06:14

on how stuff needs to be done at all right , yeah , and

1:06:16

I think also like this , for ui is like

1:06:18

yeah , if it looks well on the ui , it's

1:06:20

fine , right , yeah , and then it's

1:06:22

a mess underneath , it's a bit uh , yeah

1:06:25

, I feel like there's always a bit of a danger , right , but yeah

1:06:28

, that's , that's , that's the truth , for I

1:06:30

think the challenge is very much like with very limited

1:06:32

effort , you can make something that appears

1:06:34

very functional , yeah

1:06:36

, which might not do what you're expecting

1:06:38

yeah , I think the moment

1:06:40

there's a bug or you need to change something before

1:06:42

if you didn't have the knowledge , making something that appears

1:06:45

to be very functional was very hard , yeah

1:06:47

, so that barrier was completely removed so

1:06:49

.

1:06:49

So I guess from yeah , yeah , but

1:06:52

I do think one use for chat GPT is

1:06:54

like when I say I don't know , write a recursive

1:06:56

function of this X and Y with this type of elements

1:06:58

and I kind of know what I want to do , but

1:07:00

it's just a matter of saving the time for me to type it . So

1:07:03

if you know what you want to do , I feel like for this maybe it's

1:07:05

too big if you know

1:07:07

what you kind of wanted to do and you can read the code and

1:07:10

you understand and you can make changes , I

1:07:12

think that's also valid . I think the issue is when you don't

1:07:15

really know what's happening

1:07:17

underneath and

1:07:19

then it's almost like a house of cards , right , like if

1:07:21

something , if you need to change one thing later

1:07:24

, if there is a bug , or if you there , I don't know then

1:07:26

kind of everything falls on you , right

1:07:28

. So , but there are there , it's a valid

1:07:30

, uh , there are valid use cases for

1:07:33

it . All righty

1:07:35

, but really cool . I feel

1:07:37

like uh covered quite a lot of stuff today

1:07:39

. We did have a few

1:07:41

more topics , but we can leave it for

1:07:43

next week as well . We

1:07:46

also have a long weekend ahead of us ahead of us , yeah , right monday is also

1:07:48

holiday , oh wow . And then you realized again a long weekend . We just came from long week , I know . That's

1:07:50

why I think a ahead of us , ahead of us .

1:07:51

Yeah right , monday is also holiday . Oh wow , I didn't even realize it . Again

1:07:53

a long weekend . We just came from long week I

1:07:55

know .

1:07:55

That's why I think a lot of people are taking this week as

1:07:57

a holiday , because then you have friday , saturday

1:07:59

, sunday , and then you take this week off , and then you have saturday

1:08:02

, sunday , monday off . Why do you tell ? Me this now

1:08:04

I'll

1:08:07

tell you earlier next time . I'm so sorry

1:08:09

, so

1:08:14

I'm assuming you don't have any plans .

1:08:16

No , not really no .

1:08:18

Okay , what about you , alex ? Anything

1:08:21

? No , I also

1:08:23

forgot . Okay , yeah , I didn't know

1:08:25

either . To be honest , I forgot I

1:08:28

was filling in my holidays and then I saw like , oh yeah

1:08:30

, a lot of people are taking holidays . Oh yeah , okay , there's a long

1:08:32

weekend , long weekend , that's smart . But

1:08:35

yeah , I also feel like in the end of the year , cause

1:08:37

you have to use your holidays in

1:08:40

the year cycle right In Belgium

1:08:42

. So I also feel like either

1:08:44

you plan well and then you allocate this time

1:08:46

, or you just didn't use holidays throughout the

1:08:48

year , and then you have to use it at some point . Oh yeah , I can

1:08:50

use it here , or you used it before and then you realize

1:08:52

you don't have enough . So it's a bit so

1:08:55

, okay , but cool , then

1:08:57

I guess I don't have anything planned either , but

1:09:00

I didn't . Well , I kind of knew about this , but

1:09:03

I think that's

1:09:05

it . I think we can call it a pod . Stay

1:09:07

warm everyone . Stay

1:09:09

warm everyone . Thanks , you have taste . See you next week In

1:09:11

a way that's meaningful to self-improvement Next week . We have to think while we're

1:09:13

recording . Hello , good morning

1:09:15

sir . I'm Bill Gates . I just

1:09:18

talked about it .

1:09:19

I would recommend TypeScript . Yeah

1:09:22

, it writes a lot of code

1:09:24

for me and usually it's slightly wrong

1:09:26

. I'm reminded , incidentally , of

1:09:28

Rust here , rust , rust , this , I'm reminded

1:09:31

it's a rust here Rust , rust .

1:09:32

This almost makes me happy that I didn't

1:09:34

become a supermodel .

1:09:35

Cooper and .

1:09:37

Netties .

1:09:38

Well , I'm sorry guys , I don't know

1:09:40

what's going on .

1:09:41

Thank you for the opportunity to speak to you today about

1:09:44

large neural networks . It's really an honor to be

1:09:46

here .

1:09:46

Rust Rust Data topics . Welcome to the

1:09:48

data . Welcome to the data topics podcast

1:09:51

.