0:00

Well, boys, I've been chewing on something long enough now that I

0:02

think I'm ready to talk about it on the show. I've

0:04

been trying to think about all this AI

0:06

stuff that's been just thing

0:08

after thing. And

0:10

all of the discussion around safety and then the news that

0:12

we're going to talk about in a little bit with Gemini.

0:15

And I feel like perhaps I've stumbled on to a hard

0:17

truth. I want everybody's feedback, but

0:20

we're always talking about user consent and how much it

0:22

matters. But I don't think we're thinking

0:24

about it in terms that regular

0:27

people care about. I

0:29

think user consent matters and

0:31

hijacking your intentions as a user is

0:33

wrong. Telling me

0:35

something like this isn't appropriate is

0:37

hijacking my intentions. Enforcing

0:40

a particular worldview, I think that's

0:42

unethical. I mean, I agree on some

0:44

of the things that they're trying to protect, like racism. Those are bad. But

0:49

so is cramming a for-profit

0:51

company like Google, cramming

0:53

their ideals and morality down my throat just

0:56

feels like the worst case

0:58

sci-fi dystopian future that we

1:00

worried about. And we have

1:02

these corporations that are setting themselves up as

1:04

like arbiters of reality, of what's right and wrong. When

1:08

morality should be set by the people, not

1:11

corporations or governments. And

1:13

the more I think about this, I think this

1:15

is just going to be a hard reality because

1:17

all these companies are so risk adverse. Yeah, they're

1:20

not really set up to be speaking truths. They're

1:22

there to set up to be to incentivize their profits and

1:24

say things that won't get them in trouble. Yeah,

1:28

they're not necessarily doing a good job of that, but that's what they're

1:30

trying to do. But like if I run it

1:32

on my system and I have control over it, it's on

1:34

my Linux box, those aren't my

1:36

priorities. And I think if this is a hard

1:39

truth that we kind of accept, then really the

1:41

future for interesting AI that's

1:43

going to be pushing the limits and for

1:45

good and bad. And we'll see stories around

1:47

both. I think it's going to be

1:49

open source. I think it's going to be

1:51

the open source stuff. And that was really the motivation for today's

1:53

episode. Hello

2:06

friends and welcome back to your weekly Linux

2:08

tech show. My name is Greg. My

2:11

name is Will. And my name is Brent. Well

2:13

hello boys. Coming up on the show, you know what

2:15

they say, if you want it done right, you gotta

2:17

do it yourself. So this week, we're

2:20

gonna see if we can tame the moving target that is

2:22

the current AI stacks out there that you might want to

2:24

try. We think maybe we have

2:26

found one of the best ways possible to

2:28

deploy these AI tools under your own control

2:30

on your box. So stay tuned and find

2:32

out for that. And then we're gonna round the show out

2:34

with some great boosts and picks and a

2:36

lot more. So before we go further, let's say good

2:38

morning to our virtual lug. Time appropriate greetings, Mumble Room.

2:40

Hello Brent. Howdy.

2:43

Hello Brent. Hello.

2:45

Hello. Hello and

2:48

a big good morning to our friends

2:50

over at Tailscale, tailscale.com/Linux Unplugged. We're

2:53

playing around with these tools. We're deploying it on our tail net. We can

2:55

all access it by name. We

2:58

don't have to worry about security vulnerabilities in these brand

3:00

new projects because we never put it on the public

3:02

internet. I don't really deploy anything on the public

3:04

internet anymore. It is the easiest way to connect

3:06

your devices and services to each other wherever they are regardless

3:09

of the platform. And it's fast. Go

3:12

try it on 100 devices

3:14

and support the show at

3:16

tailscale.com/ Linux Unplugged.

3:20

All right. So this week I think if we

3:22

could have one overarching goal for the show, I

3:24

would love to at least get the listeners thinking

3:27

more seriously about open

3:30

source AI tools and less about

3:32

things like chat GPT and copilots

3:34

and things like that. We're

3:37

gonna talk about the news that's come up this week in just a little bit.

3:40

So the tooling has come a long way for

3:42

the open source stuff just on the web and

3:44

communities, the stuff you can install, and

3:46

also some of it's gotten complicated and some of it's broken.

3:49

And I think the other reality we're sitting with is

3:52

these commercial solutions, they're so risk

3:55

adverse that it's almost reducing the

3:57

usefulness of the

3:59

product, right? It's embarrassing really. It

4:01

worries me too because that's just more of the –

4:04

it's already an imbalance because only these giant companies seem

4:06

to have the resources to sort of train

4:08

the foundational models and of course they're

4:10

the ones applying all of these filters that end up

4:13

harming the actual productive use and so if you can't

4:15

say anything – I mean let's not

4:17

even say offensive but just sort of creative, weird, out

4:19

of the ordinary, then they're left with

4:21

the only sort of access to these tools

4:23

that they can tune down those filters internally

4:25

presumably and then they can make any interesting

4:28

stuff. That's a good point. Yeah,

4:30

they're the only ones really. They become the gatekeepers

4:32

essentially of this and this is why

4:34

I sort of feel like we have a

4:36

limited time to win the race in the open source community.

4:39

I hope to make the case here in just

4:42

a moment that there are several factors that

4:44

are working not necessarily together

4:46

but the result is they work together to

4:50

limit the usefulness and

4:52

utility of the open source stuff and

4:54

I want to cover all of that in a moment. So

4:58

first we have a couple of events that are

5:00

just around the corner. We are three Sundays away

5:02

before we hit the road, boy. Yeah.

5:05

Okay. Three Sundays. I'm

5:07

sure you're totally ready, Brent, right? Because I mean – Oh yeah, yeah.

5:09

Yeah, always. You know me. I'm early on

5:11

the spot. I'm not worried about that at all. He

5:13

committed on here. Yep. Multiple times for weeks

5:15

in a row. And I always end

5:17

up there. Yeah, that's true. That's true. Just

5:20

how much did I stress about it is really it.

5:22

We are getting so excited about the very first NICScon

5:24

North America. We can't wait to bring you coverage

5:26

of that. That is co-located with

5:29

scale in Pasadena, California. It is the

5:31

Linux Expo number 21

5:33

for Southern California. It

5:35

all kicks off March 14th. We're hitting the

5:37

road just like a day or two before that. If you want to

5:40

sign up for scale or go to NICScon, go to them both, get

5:43

a scale pass and use our promo code

5:45

JBFTW, JB for the win to save

5:47

50% off. And

5:50

then join us at lunch Saturday at 1.30 pm at the yard

5:52

house. We want to see you there. And

5:54

then of course, go grab fountain.fm because

5:57

they are creating a totally upgraded live

5:59

experience. for us in app and

6:01

on the web. Ooh. They're working on a

6:03

cool web app that they've prototyped for some

6:05

of the live value for value music streaming

6:07

events. Okay. That can toggle between audio

6:09

and video. You can use it in the app or

6:12

you can use it on the web. Nice. Well,

6:15

I wanna see this. Oh, it's gonna be great. So

6:17

we're gonna have a whole batch of live streams, one on

6:19

the 12th, the 14th, the 15th, and the 17th, and you don't

6:21

have to really worry about any of that right

6:23

now, but we're looking forward to it. And I think, depending

6:26

on how all this goes, we're kinda gonna be defining a

6:28

new live experience for the show and

6:30

for future events. So I think it's gonna

6:32

be kind of, I'm making a break at Trend Center

6:34

for us. We're kicking off the first

6:37

event, bigly, bigly, as Wesley

6:39

says. That's right. Okay, well that's kinda

6:41

all I had, announcement-wise. You

6:43

feeling good? I am feeling good. Did you decide you're

6:45

flying back? I am. Yeah,

6:47

for the work, right? Mm-hmm. For your hashtag, J-Jog?

6:50

Yeah, get back to the regular things. Yeah, that's right.

6:52

We're gonna take Brent maybe somewhere special than just to

6:54

make you jealous, but. Dang it, I knew this would

6:56

happen. I'm gonna have to fly back to then visit

6:58

for that part of it. What are the options? Brent,

7:00

you know what I'm gonna be advocating for. I don't

7:03

know how we're gonna make it work. That's

7:05

what you said last time. I'm gonna be advocating for the coast.

7:08

Oh, this is exactly what you said last time.

7:10

I never have been. I know, I know, I

7:12

know, I know. I'm aware. But we haven't booked

7:14

any of our return Airbnb-bizzles, so. Bring

7:17

it on. Yeah, you've got flexibility. We do, maybe

7:19

we just do like hotels or something, I don't

7:21

know, but I would love to take

7:23

the coast. I think one of the prettiest

7:25

places, that west coast is one of the prettiest places in the

7:27

world. It's really something. We'll

7:30

take pictures though, Wes, we'll take pictures. All

7:34

right, so you probably heard the news this

7:36

week about Gemini just being embarrassingly bad. As

7:38

the register put it, Google

7:41

halts Gemini image generation to fix a

7:43

white balance. They

7:46

go on to say Google has

7:48

suspended the availability of text-to-image capabilities

7:50

in its recently released Gemini multi-model

7:52

foundational AI model after

7:54

it failed to accurately represent white

7:56

Europeans and Americans in specific historical

7:58

contexts. headline which

8:00

I thought was really on point quote

8:02

big tech keeps poisoning the well without

8:04

facing any consequences for its folly. Well

8:08

I say today here on this show they

8:10

face the consequences gentlemen. Today we

8:12

say this far no farther the

8:15

line shall not be crossed we

8:18

are going sovereign with our AI technology and

8:20

I was excited to see stable diffusion 3 came out. Big

8:23

fan of stable diffusion image generation stuff. I'm

8:27

not so sure though the hosted version is really

8:29

going to hold up in

8:31

their announcement. They write

8:34

we believe in safe responsible AI practices

8:36

this means we've taken and continue to

8:38

take responsible steps to prevent the misuse

8:40

of stable diffusion 3 by bad actors.

8:43

Who bad actors are though I'm not sure it doesn't say I

8:45

guess that's they define who the bad actor

8:48

is. Oh yeah of course well that's definitely

8:50

one of them. They continue safety

8:52

starts when we begin training our model

8:54

and continues throughout the testing evaluation and

8:57

deployment. In preparation for

8:59

this early preview we've introduced numerous

9:01

safeguards by continually collaborating with researchers

9:03

experts and our community. We

9:06

expect to innovate further with integrity as

9:08

we approach this model's public release. So

9:11

all of the talk right now really is leading

9:13

with responsible AI practices

9:16

and safety. Perhaps as they should

9:18

be but it's definitely the focus and

9:20

I thought commenters on Hacker

9:23

News nailed this when

9:25

they were talking about how it's basically making

9:27

the products less useful. Their

9:29

sensitive nature is causing the rest of us to

9:31

just want to look for something that we can

9:33

self-host and run because I

9:36

get offended by them projecting some

9:38

intentions that they manufacture onto me. Yeah

9:41

well and I mean you just you're kind

9:43

of left out you know it hasn't

9:45

been a perfect system but by and large you

9:47

kind of get to decide you know

9:50

if you're gonna repost something if you're gonna share it

9:52

if you're gonna use your speech to have it and

9:55

you're out of that and you're sort

9:58

of implied like well you could do anything with this,

10:00

even though you might just want to make a silly limerick

10:03

right there in the app and you'll never use it again.

10:05

It kind of feels to me like this is potentially

10:07

going to create another divide that we've

10:10

seen in computer

10:12

privacy and data privacy. Those of us

10:14

who are technically sound enough to host

10:16

our own stuff will have far

10:19

more privacy than those

10:21

who just don't have those skills. This

10:24

feels like it's headed in that exact

10:26

direction. It's also alarming. The

10:28

point that the reg makes is actually well taken.

10:30

It's like they keep screwing it up, but

10:33

because we have no other choice or no other

10:35

viable choice, they face

10:38

no consequences for their folly.

10:41

You think about something like this that had to ship, how

10:44

they missed something like this in testing. We

10:46

were sharing screenshots this morning, I was, because throughout

10:49

the weekend, people are just trying various different things

10:51

that you think would be easy for it to

10:54

answer and it can't answer them.

10:56

It sure seems like it is kind of like end

10:58

of the pipeline. They're

11:01

not necessarily being able to fix or address the

11:03

systemic problems in their data sets. Then at the

11:05

end, they're kind of just like, oh,

11:08

well, we'll apply some little fix to try to

11:10

tweak it to make sure it's sufficiently diverse or

11:12

whatever the – like correct for

11:14

other things that they want to make it generally

11:16

more accurate, but then you're doing it so late

11:18

in the process that you get crap results. Wouldn't

11:20

it have to understand

11:23

to not answer? I think it depends on what

11:25

layer this is being applied. But

11:28

your point taken is it must realize the

11:30

truth in order to generate something that is

11:32

either the opposite of the truth or to

11:35

say it can't answer it because of moral reasons. It

11:38

must have some context awareness then of the

11:40

actual question underneath, but then you're right, the

11:42

presentation layer, they're restricting.

11:44

Well I just mean like what

11:47

is implied or learned from

11:49

the data sets that are being selected and

11:51

then they're just trying to apply things at the end that are

11:54

going to be very coarse grained and are not at the level

11:56

that a giant data set has. And

12:00

I think at the same time as

12:03

we have this sort of ridiculous over

12:05

safe, we had an example on Coda Radio. Somebody

12:08

wanted to know the absolute fastest way to

12:10

do something in C-sharp and

12:12

the LLM would not respond because it was

12:14

too risky and dangerous to do something just

12:16

for speed and not for good code quality.

12:19

Couldn't tell you the answer. That's where it feels

12:21

like it's another level. I think probably a lot

12:23

of us, especially for hosted services, I

12:26

expect there to be some level of this. If

12:28

I'm just generating obvious hate speech, I could see

12:31

it probably being reasonable for the

12:33

general public of something that's like, we're

12:35

not cool with this from our product for you. And

12:38

if it's an edge case, it's fine.

12:40

It's a reasonable limitation. But right

12:42

now, it feels like every other thing that you

12:44

ask these machines to do, you get back a

12:46

little slap in the face. Sometimes

12:49

it's even like – it's like a parody of

12:51

a brand name or something. It's like, oh,

12:53

we can't really let you speak ill of the Burger

12:56

King. Or

12:58

like public figures, which an

13:00

artist could whip up and it would be totally

13:02

free speech. It's an important part of our public

13:04

discourse. And then at

13:06

the same time, you have a group out there

13:09

that Politico has labeled as AI doomsayers

13:12

that are funded by billionaires that are

13:14

ramping up lobbying right now as we

13:16

record. So two

13:18

different nonprofits, the Center for AI Policy

13:20

and the Center for AI Safety funded

13:23

by different tech billionaires, have begun directly

13:25

lobbying Washington to address what they perceive

13:27

as existential threats posed by AI. So

13:31

each nonprofit has spent around 100,000, north of $100,000 on

13:33

lobbying efforts of both of them. Specifically,

13:38

they spent that money in

13:40

the last three months. And

13:42

they're drawing their funding from different organizations throughout

13:45

the industry that are tied to AI. And

13:49

the part that I love is like Brent's

13:51

good buddy, Sam Bankman, fraud,

13:54

both nonprofits have ties to the effect of

13:56

altruism cause, which has

13:58

been an absolute – I

14:01

mean I just think it's been absolutely disgraced with

14:03

folks like Sam Altman and others. And

14:06

these folks all have what, in my opinion, is a

14:08

God complex where they believe they have to save humanity.

14:10

So they create problems and then they panic about them

14:12

and pretend like they're the only ones that can save

14:14

us from them. And that's

14:16

how they get off, is rich billionaires, is saving

14:18

the world from these existential crises that they're very

14:21

funding created in the first place. The

14:24

efforts to influence Washington on AI were

14:26

conducted through going directly

14:28

at staffers and think takes.

14:31

That was really the approach they took before. And we

14:33

got some executive orders

14:36

here and we got some action from a

14:39

couple of senators, but not much. Nothing really formed any

14:41

kind of law. But so now they're going with the

14:43

full-on lobbying effort. And essentially

14:45

what they're pushing for is in order

14:48

to work on AI projects, you need to be

14:51

licensed. And you need to

14:53

be essentially re-licensed every couple of years to make

14:55

sure you're not creating dangerous AI. And

14:58

then the projects you're working on need to be checked and

15:00

audited by a group, an industry

15:02

group. And they even want

15:05

to have restrictions around the power and size of

15:07

the models and whatnot is what

15:09

they're advocating for. And they

15:11

want this implemented at a law level

15:14

from Washington, D.C., here in the States. I'm

15:16

not against any kind of regulation for this kind of stuff.

15:18

I'm sure at some level, especially as the scale picks up,

15:20

we'll need some. But boy do I

15:22

– it does not seem like A, we're long enough

15:24

into it to really know what the effective regulation should

15:26

be. And then B, just the – I don't know

15:28

that our current system is really set up to –

15:32

it's all going to change a lot. So whatever

15:34

we need to do, the queue would be get

15:36

something reasonable and then something nimble that you would

15:38

actually update as these things develop. And then I

15:40

have not a lot of faith in. And something

15:43

that didn't require millions of dollars of lawyers to

15:45

nursemaid through the system, which it would seem these

15:47

systems do require that, which almost means you're just

15:49

– If we not lock in the incumbents already? Right. You're

15:52

just locking in the incumbents, right?

15:54

And you're making it Particularly difficult to

15:57

use Open Source AI Tools in

15:59

Business. In Business Development firm for

16:01

development. You. Know if you want to run a

16:03

little model on your little jeep you, that's. Why?

16:06

Don't go building something that's gonna be running as

16:08

part of a software. The Service: Sort.

16:11

Of same thinking. That reminds me enough

16:13

that the folks to get prosecuted for

16:15

hacking when they download a file with

16:17

her on a yeah in a public

16:19

school teacher yeah yeah it's it's information

16:21

and I think that's why I I

16:23

want to advocate to our listener base

16:25

or even if you're not currently wrapping

16:27

your heads around this. Just

16:30

stay of informed a little bit would give you some

16:32

resources on an hour and actively soliciting more from you

16:34

out there in the audience because. Feel. Like

16:36

we have kind of the race to win here to

16:38

degree to get to certain. User.

16:40

Base size. Where. They can't just completely

16:42

ignore it and or and at a legislator level.

16:45

Or if we got to a certain size adoption

16:47

before it goes too far then it's like they

16:49

the as a bottle. And so

16:51

I feel like one of the things they show could do

16:54

to help contribute to that is cover the tooling that is

16:56

accessible to all of us. And. Help

16:58

you deploy in ways that are consistent. On.

17:01

That. May. Be are nimble like West said,

17:03

because the thing is changing really fast. So.

17:06

I think you know. That are

17:08

we focus on next. Warped.

17:12

Dot Dev/linux Das

17:15

Terminal. And. That's right,

17:17

warm the modern rust base

17:19

terminal. With. A I

17:21

built in. Is. Now available for

17:23

Linux, it used only be out there

17:25

for a Mac O S users, but

17:27

it's available for linux on multiple distributions,

17:29

including a boon to for Dora Debbie

17:31

and an arch. You. Can give

17:33

it a try to stay at Warp.does/linux

17:35

Das terminal. It is something that feels

17:37

like they rethought the terminal for the

17:39

modern era. If you work in the

17:41

corporate world where there's Mac and Windows

17:43

and Linux systems, you're going to feel

17:45

like you have a tool that just

17:47

keeps up with the rest. This.

17:50

Is really slick because eight built on resting

17:52

on quick. But. It has

17:54

a really nice modern text editor built

17:56

in so when you're editing your your

17:58

yam of your docker compose. He

18:01

drinks on whatever it is. you know, got a modern

18:03

text editor built in. And it lets.

18:05

You navigate through blocks of output really

18:07

quickly. V. For it man, there's that

18:10

warp ai that you can invoke and it

18:12

will suggest the right command for you. This

18:14

is what I've been playing with and it

18:16

is really handy is also just the ability

18:18

to customize the U Wot. Yeah.

18:20

Know you to do that. You can create your own

18:22

prompts to and. Have. All the nice

18:25

thing set ready to go see them. Recall those when

18:27

you need them. It's a great user experience. Have a

18:29

collaboration feature in their. All. Of

18:31

it is fantastic for developers

18:34

are engineers. Who.

18:36

Are it'll work in our living in that Turner. I

18:39

sometimes looked over at some of the other criminals on the

18:41

commercial platform than it's been a little jealous. And

18:43

now now now not anymore. Go.

18:46

Check it out!

18:49

Warped.dev/linux-terminal. Now. Available for

18:51

linux. including. A boon to For

18:53

Door Debbie and Arch Linux and about a

18:55

lot more coming soon. You know how those

18:57

things work out. Go check it out. Give

18:59

it a shot. Support the show by going

19:01

to Warp.deaths Last Linux, Das Terminal. And.

19:04

Give it a try to deck. Warped. Attempts

19:06

Last Linux Das. It's

19:12

not all bad news though. I mean

19:15

that the same time that these are

19:17

big corporations are gonna in uncertain their

19:19

foot in public they are also really

19:21

seem free and open. Wait. Models.

19:24

That didn't go play with like a Googles

19:26

New Gemma. I mean, this is pretty nice

19:28

to see it this soon after a Gemini

19:30

comes out and die. It's a two billion

19:32

parameter model. Wow. Two sizes. The two billion,

19:35

Seven billion parameters. Available now.

19:37

They do play like all their safety stuff

19:39

to it. So yeah the know it's facilities,

19:41

different models you can grab. We talked about

19:43

this before they all whatever institution makes them.

19:45

it comes with that institutional bias. There is

19:47

that aspect to it. but. What's.

19:49

Why are they doing this? Why are they taken like these?

19:52

These. Models are kind of behind all of this work

19:54

they've done. And. Released into the

19:56

public? Is it? Isn't. Like this like a

19:58

good hub idea where you make it. Three So

20:00

developers can I get hooked? A is it.

20:02

and I guess I don't really understand the

20:04

motivation for all of these models coming up.

20:07

Specially. When googles like trying to get people to pay

20:09

for. An. Outrageous amount for Gemini. Yeah.

20:11

I may as well the one ended his

20:14

to help mindshare adoption I think because the

20:16

scenery and has been the area of them

20:18

is academic interest for a long time and

20:20

than is in the last few years sort

20:22

of blossom now to been brilliant applied science

20:24

and industry in there is sort of a.

20:28

Reputation will aspect of i'm publishing what you're doing

20:30

like whether that be a paper and then a

20:32

lotta times now the paper and the model behind

20:34

it and maybe they just feel comfortable that there's

20:37

enough you know special sauce that they can sort

20:39

of keep on their own plus celtics resources and

20:41

those sort of get to see up you and

20:43

I can get this to work and tied into

20:46

stuff but. The hollow level the building

20:48

into a nice service and I hadn't thought about

20:50

the historical kind of momentum around releasing the stuff

20:52

to as sort of educational background am. I

20:55

speaking that a llama. Which. we've talked

20:57

about from the show is already incorporated. the

20:59

so Gemini Gemma comes out and then within

21:01

a day or so. The. All

21:03

along folks are already incorporating it into the

21:06

free stuff. is moving really fast. Yeah Cuomo

21:08

seems to be don't play nicely and see

21:10

they have a rest A P I did

21:13

get I think chat G B T style

21:15

A P I compatibility. There's also some by

21:17

Indians like Python bindings probably javascript bindings as

21:19

well so I can easier and easier to

21:22

adopt and build into your apps if you're

21:24

so what do So I thought we should

21:26

focus on image generation this episode. There's a

21:28

lot of different tune out there but all

21:31

of the hoopla around Gemini has been with.

21:33

Image. Generation. And. We've talked

21:36

about easy, stable diffusion before. It's a

21:38

docker container that you can get up

21:40

and running pretty quickly. This.

21:42

Week though, I want to talk about invoke a I.

21:44

Invoke. A I is and implantation of

21:47

stable diffusion. That. Has

21:49

along with it a great set

21:51

of to lean and streamline process

21:53

to generate images. Polian updated models

21:55

or modifiers. and work

21:57

to generate images that whatever you

21:59

want really depending on the model using. We've been

22:01

experimenting with it on and off for the last few

22:03

days. We'll be talking about that. But the

22:06

reality is there's a

22:08

lot of ways to install something like this. And

22:11

I've noticed a lot of bad

22:13

habits. Invoke AI in

22:15

particular is guilty of this. A

22:18

lot of like just run this script. And

22:21

it just dumps stuff all over your

22:23

system and running stuff and I mean, I

22:25

think also some of the you know, just

22:27

the way these the history maybe some of

22:29

the coming out of academia. There's a lot

22:31

of specific stuff. There's different package managers. There's

22:33

a lot of specific dependencies that you need

22:35

for your particular hardware. Sometimes it's like sensitivity

22:37

to which version is going to work. You

22:39

know, new ones won't or way older ones

22:41

won't you need this set over here matched

22:44

with this one over here. Plus,

22:46

as you're saying, there's kind of a lot of

22:48

stateful stuff where you're running scripts to get everything

22:50

just so set up and then it's kind of

22:52

based on whoever happened to tie all those things

22:54

together. And then some of it's kind of brittle

22:56

because it's dependent on certain like video acceleration libraries

22:58

working and being there installed correctly. And you end

23:00

up with a bunch of stuff scattered plus then

23:02

there's the web UI is the goal along with

23:04

these things. So you got a separate sort of

23:06

JavaScript app probably that you got to you got

23:08

to handle and build and then have a web

23:10

server for seat and then because you know, you

23:12

can't just trust them that there's not like a

23:15

unified system for it. Everyone's gonna run their own

23:17

thing and you know, install Apache or add to

23:20

your engine X or Oh God, you're making my

23:22

blood pressure go up just talking about it is

23:24

really a mess. And then you also have like

23:26

Docker containers. Of course, those can be

23:28

loaded full of some stuff but not have other things

23:30

that you need for your particular hardware or like if

23:32

you deploy it on the Mac, you won't get access

23:34

to hardware acceleration in some cases. So

23:37

there's there's a lot of edge cases around

23:39

the containers and I haven't really seen any

23:41

of this stuff shipped as like

23:43

a flat pack yet. I've been waiting for

23:45

something like invoke AI and stable diffusion to just be a

23:48

flat pack and you install it, it starts a little web

23:50

server gives you the URL and like a little pop up

23:52

and you click that and it brings

23:54

it up but so far that's

23:56

not happening. And so you have containers with

23:58

their limitations sometimes they'll work. great be exactly

24:00

what you need and sometimes you're

24:03

gonna be executing into that thing and fixing

24:05

stuff and adding stuff so it works on

24:07

your machine. You got the blast and spray

24:09

and pray method which is what

24:11

the Mac users seem to be just going for

24:13

which I don't understand you know

24:15

there's other ways you could do it and

24:17

then of course you could

24:20

do it from scratch and you could actually build it pull

24:22

down everything man is that a job. Yeah right you can

24:24

kind of follow the developer path and be like okay if

24:26

I was gonna be working on this project what all would

24:28

I have to do so you can guess we

24:31

didn't want to do it this way you could probably

24:33

guess from our tone this is not how we did it

24:36

and we think maybe you shouldn't do it those ways

24:38

either we think maybe we have a much

24:40

better way to do this. Determinate.systems.unplugged.

24:47

Bring Nix to work the way you've

24:49

always wanted with flakehub.com. Go register

24:52

for the private beta and

24:54

get secure compliance friendly Nix

24:56

and all the supports you

24:58

need at determinate.systems.unplugged. So let's

25:00

talk about flakehub.com. It's the all-in-one platform

25:02

for secure compliant transformative Nix

25:05

development and it is made by

25:08

our friends at Determinate Systems the

25:10

builders of the best Nix installer

25:12

including the massively popular MDM

25:14

and skiff mode support which gets

25:16

20,000 installs per day massively

25:18

successful and they have an

25:21

advanced binary cache with sophisticated access controls if

25:23

you're working in a team is going to

25:25

be mind-blowing for you they're also the ones

25:27

behind that cloud-native Nix project we've talked about

25:29

and just a plethora of open source tools

25:31

and documentation for the Nix community and flakehub.

25:35

Flakehub is solving big problems if

25:37

you're building with Nix. Nix caching

25:40

traditionally involves a globally flat namespace

25:42

so imagine building

25:44

software it's like having a massive toolbox with tons

25:46

of parts and tools in there but

25:49

finding the right one can take forever when it's a mess and

25:52

auditing everything that's in there that got pulled in

25:54

it's just well it's a nightmare if maybe not

25:56

even impossible traditional

25:58

Nix caching is like having multiple toolboxes

26:00

scattered everywhere. And each team member

26:02

has their own set, and

26:04

it's hard to share the tools or know who has

26:06

which one, and if they have the right one, it's

26:10

not really elegant. Well,

26:12

FlakeHub is. It's a single organized toolbox

26:14

for your entire company, if you will.

26:17

Imagine one identity-aware cache where everyone gets

26:19

the right tools based on their permissions,

26:22

no more searching through that mess with fine-grained

26:24

access control so you make sure things stay

26:27

sensitive and they're in the right spot. And

26:30

FlakeHub has that globally distributed

26:32

build cache. Man, has

26:34

that been useful here when we've been

26:36

working with AI tools. If you can

26:38

use GitHub, FlakeHub will fit into your

26:40

existing DevOps workflow. And it's the only

26:42

SOC 2 certified platform for Nix, and

26:45

it delivers a low-friction Nix development

26:47

experience that you've always wanted. Go

26:50

bring Nix to work the way you've always

26:52

wanted with flakehub.com. You've got to

26:54

check this out. Go support the show and register

26:56

for the private beta to get secure, compliant-friendly Nix,

27:00

and all the support you need. You

27:02

go to determinant.systems.unplugged. We'll have

27:05

a link for that in the show

27:07

notes too. That's determinant.systems.unplugged. And

27:09

a big thank you to Determinant Systems for

27:11

sponsoring the Unplugged program. Determinant.systems.unplugged.

27:19

Now, if you've been following the show lately, you

27:22

might realize that we have a favorite way of

27:25

solving this problem. We

27:33

wanted to find out if we could Nixify our AI tools, and

27:37

we didn't have to go very far. Sometimes

27:39

it's easy. Nixify.ai is a place you

27:42

can start. They are trying to just

27:44

make things simply available, a large repository

27:46

of AI executable code that

27:48

might be impractical or hard to build or figure

27:50

out yourself. They've got it. They've got

27:52

it over there. I think we saw it kind

27:54

of float by. Yeah. Some

27:57

of our Nix friends were sharing it around sometime last year,

27:59

but we actually. hadn't had a chance to give

28:01

it a try. And I mean, yeah, it's exactly

28:03

what we want, at least if it

28:05

works. Here's how they pitch it. Each project

28:08

is self-contained, but without containers.

28:11

By the way, when we say Nix, we don't mean Nix OS.

28:14

It would work on Nix OS. But this

28:16

will even work in WSL. You just need to

28:18

get Nix, the package manager, installed. So

28:20

you got self-contained applications. It's easy

28:23

to run. It's like

28:25

a one command to actually install and run it

28:27

once you've got Nix set up. Their

28:29

projects have support for NVIDIA and AMD

28:31

GPUs. So it'll work with other. And

28:33

like I mentioned, you WSL users

28:36

are not left out. And I recognize there's more

28:38

and more out there. Go get

28:40

Nix OS WSL, and then you

28:42

can run GPU accelerated AI workloads

28:44

on WSL using these commands. So

28:47

you can get it right there on your winner's box if you want. Now,

28:50

Wes, am I correct in

28:53

saying that they're using flakes under the hood? Yeah,

28:56

I think that is the primary method that they advertise. If

28:59

you go on the site and you want to go

29:01

get started, let's say

29:03

you've got an AMD system. You do

29:06

Nix run. And then basically a GitHub

29:08

flake reference that points you over at

29:10

their Nixified AI slash flake repo. And

29:13

then you tell it you want either the AMD or

29:15

the NVIDIA version, whichever project you're doing, like

29:18

we were doing in Invoke AI. I want to ask

29:20

you how you got it working in a moment. I

29:22

did the spray and pray approach on a demo system.

29:25

I'm not proud of it. And that's why I don't recommend it. But

29:27

a flake is perfect for this kind of thing. I

29:29

think you could think of a flake as it's actually

29:32

pretty simple. Think of it as like a building block

29:34

for Nix project. Imagine a

29:36

flake is a self-contained box, and it has code

29:38

instructions for a thing. And

29:40

inside that there's a file named flake.nix. And

29:42

it tells Nix what to do. It'll say,

29:44

here are your inputs. These are the other

29:46

bricks you need to build the full wall.

29:49

And here are the outputs. This is what it should look

29:51

like. This is how things should be set up. These

29:53

are the parameters. And then you

29:56

can combine these to create bigger and

29:58

bigger systems. inputs

30:00

you put in your config file. And

30:03

each flake has a lock file to ensure it's always

30:05

using the same exact version of everything. That

30:07

makes it reproducible and reliable. So that's really

30:09

big from an audibility standpoint, or maybe you're

30:12

deploying an appliance, or maybe you're

30:14

just trying to get something to work that worked on this

30:16

machine, to work just like it did on another machine. And

30:19

that's, I think, particularly where it's useful for these

30:21

types of tools. They've got a lot of brittle

30:23

parts and are moving quick. So

30:25

you could think of, like, regular Nix is, you're

30:28

building everything piece by piece with bricks, and

30:31

you're assembling those bricks, and you're building the

30:33

wall. With flakes, they're pre-made

30:35

chunks of the wall, with

30:37

clear inputs and outputs. They're like sets

30:39

of Legos that you can grab and set down

30:41

a whole set of Legos all at once.

30:44

Yeah, you know, if you've ever done anything with,

30:46

you know, like PIP or NPM, or any of

30:48

these sort of programming language specific installers

30:51

and package managers, you know, they've got these lock files,

30:53

and they kind of let you manage everything and make

30:55

sure you do things in chunks that you can comprehend

30:57

and you've tested, and okay, I do want to update,

30:59

update them all for me, or, you know, leave them

31:02

pinned for now. Flakes and Nix

31:04

kind of lets you take that to all

31:06

of your packages. You know, it extends that approach

31:08

to the whole system, and

31:11

it lets you easily integrate with just

31:13

GitHub repos, whatever Git URLs you want

31:15

to use. And like you're saying,

31:17

this stuff's all moving so fast. You might have

31:19

custom versions of stuff, forked packages, different versions of

31:22

the web UI. You're not necessarily

31:24

going to want to wait for that to

31:26

like get into a centralized Nix packages repo.

31:28

Or any distro. Right. Yeah.

31:31

With the inputs, you know, the Nix flake

31:33

stuff goes and handles going and checking

31:35

what is the most recent commit, grabbing that, writing it

31:37

in your lock file, pinning it, and then getting you

31:39

all that stuff onto your system and then feeding it

31:42

into the rest of your Nix stuff. So

31:44

all that is just taken care of for

31:46

you. Now, late in the night, we

31:48

all tried our various methods, but Wes,

31:50

of course, of course did the Nixified

31:53

approach, and got it working.

31:56

Did you use a flake? Did you use a different installation approach?

31:58

How did you get things working on Nix? No, I

32:00

just went with the Nixified AI stuff. I

32:03

did see that Invoke AI has a

32:05

Flake.Nix in their repo, but

32:07

it's mostly just set up for getting the raw dependencies.

32:10

So it will get like handle all the CUDA stuff.

32:12

I think it built OpenTV, so it took a

32:14

little while, but thankfully that rig has plenty of

32:17

CPU. Shout out to our

32:19

audience that donated very nice servers. I think that

32:21

would be a nice setup if you were explicitly

32:23

trying to develop on it, because you'd kind of

32:25

get all the different Python libraries and the stuff

32:27

you needed and then you could just do the

32:29

build yourself right in that environment. But

32:31

if you want something that's a little more end user

32:33

focused and packaged, that's where the Nixified AI version came

32:35

in. I think they had some caching

32:38

in place, which was nice. So you didn't have

32:40

to build absolutely everything. Right. They

32:42

talk, you can do the Nix run, but you might

32:44

want to do Nix shell, because

32:46

there's more than one binary for some of these.

32:49

So in particular for Invoke AI, you've

32:51

got just the Invoke AI command, which does like

32:53

a command line version, targeted

32:55

more advanced users, maybe for like scripting or automation

32:58

or that kind of thing. But you

33:00

probably want the Invoke AI dash web command, because

33:02

you're trying to run the web UI. Yes, right.

33:05

And I think by default, maybe there's one it

33:07

comes with, but there's a whole bunch of different

33:09

models this thing can use. So you're going to need to

33:12

do some configuration. And most importantly, you're going to need to

33:14

actually pull down all that stuff. You're going to want to

33:16

run some stuff before you get it all launched, where

33:19

they have a configuration command you can run. It's

33:21

even got a little sort of mouse enabled

33:23

command line interface that you can kind of

33:25

click through either with tab and spacebar or

33:28

double click right there. Let's

33:30

you select what model, some of the options, some of the

33:32

enhancements further on in the pipeline that you need. Yeah, it

33:34

makes it, it sounds like it's a lot of work, but

33:36

that part of it makes it really simple for anybody that's

33:38

even new to this. Yeah, I didn't super know what I

33:41

was doing. So I just kind of clicked some of the

33:43

ones. I was like, yeah, that sounds good. And for the

33:45

most part just worked. Does take a

33:47

little while, because there's gigs and gigs

33:49

of data to download, depending on how

33:51

many models you choose and which ones.

33:54

But It downloads it. You don't even need

33:56

to restart anything. The Web: UI will pick up the new

33:58

models. There's even a little button. There to refresh

34:00

and a. They've. Gotten so and

34:03

I are you got a web you i that

34:05

lets you generate images and then you can play

34:07

around with the different plumbing than the different models

34:09

behind it and the different accelerators that they have

34:12

to or get everything twitches right. And.

34:14

You can start creating your own pictures, and you know

34:16

what stands out to me so far about invoke as

34:18

it's you can tell it's kind. Like

34:21

there's workflows that they have in mind that there's

34:23

all kinds of like you know set up here

34:25

so you can see the accused. it's going to

34:27

horns right you can have it on. We could

34:29

have no sorry and rapid just to do some.

34:31

It was really neat, is. We. Could

34:33

share a central invoke a I server and we

34:35

could all have different images. Been in the queue

34:37

and I can see yours and you could see

34:39

mine in. They would just be good generating in

34:41

the order they were submitted. Yeah and you can.

34:43

You can set up different workflows on here are

34:45

to have been out. Knock me be competitive with

34:47

everything but you can see that you're getting a

34:50

little more sophistication than just like you know. secures

34:52

the input form and and go run the model

34:54

for me. Yeah, So. I'll tell

34:56

you it obviously I use this if you

34:58

go to Jupiter.to you can see how we

35:00

use this I have. For me it's like

35:02

generating stock photography instead of paying for some

35:04

stock photography. website. I. Just generate my own

35:06

now. Are but. I.

35:09

Also. I've. Used this now

35:11

to generate backgrounds for my kids devices

35:13

so they have like custom background from

35:15

dad. My. Wife is used to

35:17

generate like her. Perfect. Wallpaper,

35:19

For her eyes on So she has like this photo

35:22

the she just loves to look at that isn't in

35:24

the she kind of created an of but in their

35:26

great for and there's just ways you can use it

35:28

that are not necessarily. You. Might think of at

35:30

first but the more you have these kinds of things and you've

35:32

insulted can be nice devil piece of. Art.

35:34

Their doesn't have to be all that important, I just need

35:36

something to add a little visual splash. It's just great for

35:39

that kind of stuff. And. If they

35:41

have a bunch of things on your to of course like

35:43

one I did not get a chance to play with. I

35:45

have it installed. Is. A text

35:47

Jen. Which. As you can probably guess

35:49

is a web you i for large language models to spit

35:51

out. Different stuff. In it supports a

35:53

bunch of the different models. And

35:55

I have that installed on my system but I just didn't get

35:57

a chance to play of the but again using the Knicks. The

36:00

Ai method. It's so straightforward. amp.

36:03

Because. It's not nick so as dependent you can be on

36:05

a boon to be using it. or you could be on mac

36:07

o s or dub yourself. I. Think that's

36:09

pretty powerful because then you get a consistent experience across

36:11

all you different systems and of course you know it

36:13

is to modifying your system in this case had salary

36:16

models that I good at all directory to keep a

36:18

bunch of stuff and store at settings. But.

36:21

You. Know it's it's not. A scene

36:24

That case didn't have root permissions running as regular

36:26

user so you know me to make a whole

36:28

separate user run it as if you'd like and

36:30

then everything else. The stuff you bill via next.

36:32

That's all just in the next door. and when

36:34

you go to Clegg garbage when you're done. He

36:36

didn't like that Molly wanted something else. He it's

36:38

going to deploy it another way. You've cleaned up,

36:40

it's cleaned up. I. Think our opinion

36:43

is is that it's easier to use.

36:45

You can read reproducible builds. And.

36:47

You have better organization of is really complex

36:49

projects we can have that can have a

36:51

lot of pieces. I've.

36:54

Put a good question on. What's.

36:56

Reasonable hardware to expect the stuff to even

36:58

be useless because I know you know some

37:00

of us are running and on fancy hardware

37:02

bus. Can anybody just run this on their

37:05

hundred other work laptop and given at least

37:07

a try? I mean obviously the performance is

37:09

gonna be different but if. How

37:11

painful west would you say? The imaginary some was

37:13

on the Cps. I'm decently decently

37:16

panel yeah, I mean I'm only

37:18

like four or is minutes yet.

37:20

For. Some enemies are you could be a lot quicker.

37:24

So. Depends on your how impatient you are.

37:27

if you have patience frighten you know and

37:29

it is nice like it gives you the

37:31

little he can he can said it's little

37:33

show you the in progress sort of pictures

37:35

as the model is refining and defusing and

37:37

in the command line or in your lungs.

37:39

It also have little progress bar. A

37:41

So at least get some pretty decent feedback, especially compared

37:43

to some of the commercial systems. In terms of like

37:45

how long am I going to be waiting. But.

37:48

You are probably isn't. i

37:51

think ill is let you start with a big invoke a

37:53

i is a fun one if you're listening to this new

37:55

thinking i do want to try some of the local stuff

37:57

i do want to try to help these projects grow I

38:00

would like to understand this a little better. If

38:02

you get invoke AI deployed, the

38:05

building blocks you use to do that will work for all

38:07

these other ones as well. There'll be little individual tweaks you

38:09

might make if it's depending on what the software is, but

38:11

like the method and the approach and the documentation is all

38:13

gonna be the same at that point. So

38:16

it's a journey, but once you

38:18

learn it, you'll have that path well-treaded. And

38:21

I think it's, I mean, I think it's like, we're

38:23

just gonna leave it, we're just gonna keep it. I mean, it's

38:25

gonna keep updating now. It's like, we're gonna be current with the

38:27

projects as they release stuff, and it's great for us. We just

38:29

set it and forget it. collide.com/unplugged.

38:34

Well, you've probably heard us

38:36

talk about Collide before, but

38:39

did you hear that Collide was just acquired by

38:41

OnePassword? It's pretty big news since these

38:43

two companies are leading the industry in

38:45

creating security solutions that put users first.

38:48

For over a year, Collide Device Trust has helped

38:50

companies with Okta ensure that

38:53

only known secure devices can access their

38:55

data. And well, it's still what

38:57

they're doing, but now they're doing as part of OnePassword. So

38:59

if you've got Okta and you've been meaning to

39:02

check out Collide, now's a great time. Collide

39:04

comes with a library of pre-built device posture checks,

39:06

and you can write your own custom checks for

39:08

just about anything you can think of. Plus

39:11

you can use Collide on devices

39:13

without MDM, like your Linux fleet,

39:15

contractor devices, and every BYOD

39:17

phone and laptop in your company. Now

39:21

that Collide is part of OnePassword, they're

39:23

only getting better. So go check out

39:25

collide.com/unplugged to learn more and watch

39:27

their demo. It's kolide.com/unplugged. Watch

39:30

the demo, support the

39:32

show. collide.com/

39:35

unplugged. Well,

39:40

conference season is coming up quick, and we

39:42

have some updates specifically around Texas Linux Fest,

39:44

which is happening April 12th to 13th. Chris,

39:46

can you catch us up on those updates?

39:49

Yes, well, mainly the time slots and the schedule are now

39:51

up on the website. So if you've been trying to look

39:53

at what's going on, the talk titles and the speakers have

39:56

been filled in as we have them, and more will be

39:58

coming up there as they get confirmed. And

40:00

it sounds like there's at

40:03

least some planning in the works for a party

40:05

Friday night. Oh. Very

40:07

potentially and there may be others. They tend to be a little more

40:09

– I just planned it the last

40:11

second. And I think the good news

40:13

for anybody that's considering going is we do have a

40:15

hotel group rate link. So

40:18

I guess normally like there's like this

40:20

hotel is like 320-ish a night and

40:23

Texas Linux Fest group rate is going to be 250

40:26

a night. So it's not bad and you're right there.

40:28

You don't have to go far. So that's good. And

40:30

again that's April 12th and the 13th and then just

40:32

shortly after that, Linux Fest Northwest 2024

40:34

is the 26th and 28th of April. And

40:39

the update here is the schedule is posted.

40:42

You'll see some familiar names on there. I

40:44

believe there's a Wes Paine on there. And

40:47

if you go to Wes's talk on Sunday at Linux Fest,

40:49

just stick around in that room because then we're going to

40:51

do Linux Unplugged after that right there. It's going to be

40:53

a whole day of fun. Yep. So you can just hang

40:55

out. Pretty great. Shoutout

40:58

to the core contributors. They are participating

41:00

in a membership program that finances the show

41:02

directly by the listeners and gives them access

41:04

to additional content. There's an ad-free version of

41:06

the show but I think

41:08

the real winner is the bootleg version

41:12

because the pre-shows have just been bangers. Some

41:14

spicy pre-shows recently. Some

41:16

spicy couple of pre-shows in there. Maybe it might

41:18

be worth checking out just for that. You do

41:20

get a little of the secret sauce. And

41:26

now it is time for the boost. Thank

41:29

you everybody who boosted in this week. We

41:31

got a whole big batch of Fountain FM

41:33

feedback and I – well, with Wes's help

41:36

bundled it all up and

41:38

sat down with Nick and we went over all of it.

41:40

And if you've been boosting from Fountain, I think nearly all

41:42

of them Nick has been replying to directly. So go check

41:44

your Fountain FM app. So

41:46

I'm not putting all of it in the show

41:49

because some of it is getting handled directly but

41:51

I will put some of it that I have answers for. And

41:53

Our baller boost came in from user 38 who sent in 51,000

41:55

Sats. Id

42:03

right my original donation was attend be a

42:05

hundred and fifty thousand and but I had

42:07

some problems with Ah Sound and maybe they

42:09

could fix the custom don't his in feature

42:11

which was. Broken. Yeah.

42:14

Yeah. Really? come on. The fact that he still

42:16

managed even with that bug where he raped

42:18

we appreciate and has become our by this

42:20

week. I. Read sarcasm comes

42:22

in with forty three thousand sets.

42:24

I thought that which will kind

42:27

costs. I. Would like fountain

42:29

to allow me to build playlist

42:31

from subscribed so it's not just

42:34

individual show episodes. I my podcast

42:36

in groups in like tech news,

42:38

dad, staff, etc. And. To

42:40

can be bothered to peruse show episodes individually

42:43

themselves. I wanna know what dad stuff he's

42:45

listening to? I don't have any dad stuff.

42:47

Pike as. I would you boost in

42:49

and tell me what you're listening to the gummy recommendation

42:52

So I wanted to put this one in the show

42:54

because I talk directly to Nyc from Fountain about all

42:56

of these and this when he says is doable to

42:58

that. Are You can tag the episodes

43:00

and then you can build a playlist based on the

43:02

tags So that is how you solve that problem. Sarcasm.

43:05

It is available to you. Already.

43:07

And I wanted you to know them. Thank

43:09

you for the boost in clearly boosted in

43:12

a mega ruff. Ducks these El Lugar not

43:14

foothold, not ducks. For. The docker

43:16

compose vs. area in discussion. I ended

43:18

up on the answerable docker Compose France

43:21

pole vault I didn't understand secrets on

43:23

doing it with next can fake. I

43:25

get that the consider says of what

43:28

file you need but doesn't store the

43:30

content or secrets. My missing something. We've.

43:32

Done this we've had, we've had different could fix

43:34

where we refer to signify what do we do

43:36

a different what? what's of mean secret the next

43:38

as a whole topic. So that depends on the

43:40

approach that year. Wanna do not? There's the simple

43:42

or maybe it's not simple depends on what you

43:45

think. Okay but there's the approach described here that

43:47

we've done that we didn't or example next glad

43:49

set up where yes manage secret separately, whatever you'd

43:51

like you could use and vault the could use

43:53

which we did linked to so you may be

43:55

referred to that. I'm and that just means

43:57

you need a separate process with an ascent secrets up where

43:59

you want them and. and the Nix stuff

44:01

is just set up to refer to those and expects them

44:03

to exist. Or you can

44:05

deploy stuff in Nix. There's

44:07

a few different options. I mean, you could just embed them in there if

44:09

you weren't worried about that. But you do have to be careful if you're

44:12

worried about security or you're on multi-use machines

44:14

because depending on what method you use, you

44:16

probably don't want your plain-text secrets in the

44:18

Nix store because anyone can just go and

44:21

read that. But two popular

44:23

ones, there's Age Nix as

44:25

well as SOPsNix, S-O-P-S, and then there's

44:27

a bunch of other projects as well.

44:30

So depending on the approach, the complexity, what your actual

44:32

needs in terms of deployment are, there

44:34

are a bunch of ways to do it in Nix, but that's

44:37

not something that we built in or planned around

44:39

our solution. And of course, Ansible and Ansible Vault

44:41

is a time-honored way to do it. Yeah, I

44:43

mean, if it works? Sounds like a

44:45

nice setup. Thank you, Ian. Appreciate the boost.

44:48

Sir lurks a lot. Comes in with 2,048 staff. Hey,

44:51

sir, nice to see you. And I like hearing your

44:54

experience with VR headsets like the Quest Chris, but

44:56

I wonder how well they will work for people

44:58

with poor eyesight. Do you wear glasses

45:00

inside the headset or does it have optics to adjust

45:02

to focus for each eye? I

45:04

don't think it has individual eye focus, but

45:07

you could wear glasses. I know some people

45:09

do. For me, I'm, as I age, becoming

45:11

nearsighted. And so since the lenses are

45:13

just like right in front of your eyeballs, for me,

45:15

it's fantastic. I, in a

45:17

MERST, also realized that you can adjust some of

45:19

the encoding quality and things. So I didn't actually

45:22

have everything turned all the way up to like

45:24

its pristine quality. Looking even better

45:26

than it was for last week. I

45:29

used it for prep on this show. And one of the things

45:31

that I've also done now with my MERST setup, which I can't

45:34

believe I didn't think of this sooner, is

45:36

I've replicated the monitors like I have

45:38

them here at the studio. So I

45:40

have a vertical screen here at the studio and

45:43

I now have that same on that same position. It's a

45:45

vertical screen and I have the same horizontal screens. And

45:48

so it's very much the same workflow for me in

45:51

MERST or in the physical world. And I'm liking that

45:53

a lot. VT52 boosts in

45:55

with a roeduct. Uh, reusing nixos.org. search

46:00

for NixOS options, little

46:02

pro tip, try searching

46:04

over at mynixos.com. It

46:07

searches across options and packages

46:09

at the same time, so

46:12

you'll find e.g. Nix packages slash

46:14

packaged slash tmux and

46:17

Nix packages slash options slash

46:19

program slash tmux in one

46:21

search. That's cool. The UX

46:23

is a little nicer too. You

46:25

can browse among the options hierarchically instead of

46:27

one flat list. Bonus, it

46:29

also searches Nix packages and home manager.

46:32

Yeah, this is a great one. I kind of forgot about

46:34

it, but I have run into it a few times, you

46:37

know, just when trying to search for Nix options and not

46:39

already at the main search site. It's

46:41

great to have other options. Thank you, Vt. mynixos.com.

46:43

That is great. Thanks,

46:45

Vt. I really appreciate that. I

46:48

have a little pro tip to add to this

46:50

that I've implemented this week. So I've been searching

46:52

the Nix packages through

46:55

DuckDuckGo Bang implementation. So

46:57

you bang Nix packages and then you just search for what

47:00

you're looking for and it sends you directly to the

47:02

Nix package search. But I've

47:04

one upped that this week on

47:07

KD Plasma and I'm using the

47:09

plasma launcher, which has the same

47:11

functionality if you look deep enough

47:13

in the settings. And I'm going

47:16

searching directly from my desktop.

47:18

It's amazing. Yeah, nice. Yeah,

47:21

I need to use krunner more. It is the

47:23

best runner up there. I feel like it's gotten a little pokey.

47:26

I'm hoping with plasma six, which comes out in a few days.

47:28

I'm hoping to fall in love with

47:30

krunner all over again. Well, Gene Bean boosted

47:32

in with a row of ducks. Regarding

47:36

Fountain, it's missing the dynamic playlists

47:38

that are used in Casamatic and

47:40

Overcast or the show stopper for

47:42

me. For example, this

47:44

show is in a playlist called Linux

47:46

and Tech that contains the shows I

47:48

subscribe to that fit that particular topic.

47:51

The playlist automatically shows me the unplayed

47:53

episodes for these collections of shows. This

47:55

allows me to see what I've heard

47:57

and haven't heard yet in a given

47:59

topic. Tags, labels, and manual

48:01

lists just don't cut it for me. How

48:05

many people are doing this where you're

48:07

like you're play listing out the

48:09

topic, like Linux and tech for Gene being here? I

48:12

just always have and I'm curious about you guys. I

48:15

always just have like my main list of chronologically

48:18

released podcasts. And I just kind of scroll through that and

48:20

pick what I'm going to listen to and I don't ever

48:22

create a playlist or anything like that. I think

48:24

at most I might create like a per session,

48:26

like I'm going for a drive and I want

48:28

to keep up a couple, something like that. I

48:30

do have road trip playlists. What about

48:32

you Brent? Do you just stack your list or

48:34

do you have playlists? Do you go all organized?

48:37

I think I would love this because typically when

48:39

I'm reaching for my podcast players because I'm in

48:41

the mood to listen to a certain type of

48:43

content. And so I end

48:45

up, Chris, like you browsing through the list,

48:48

but I find that more to be friction than inspiration

48:50

most of the time for me. So I think I

48:52

would love this feature. Lazy locks comes in

48:54

with 10,000 sets. It's

48:57

over 9000. All

49:00

your talk of virtual displays has got me

49:02

wanting to check out links in the show

49:04

notes. That visor looks promising. I hope it

49:06

delivers. Looking forward to seeing you guys at

49:08

scale. Yeah, Lazy Lock. Heck yeah. So

49:11

visor.com, the deal I think is over

49:13

now, but it's like 400 bucks

49:15

for VR headsets that are just

49:18

designed for working. Really

49:20

high quality displays that look like sunglasses.

49:23

So you're not wearing big old ski goggles. I

49:25

think people like that. Master reboot comes in with

49:27

9000 sets. Just

49:30

following up on a telegram message about the

49:32

comparison between Quest 2 and the Quest 3.

49:36

If I look dead center, everything seems fine

49:38

on my Quest 2. But if

49:40

I stray just a little, things begin to

49:42

look fussy. Can you tell me

49:44

if the Quest 3 is like this as well? I

49:48

think it's my first VR, so I don't really have anything to compare it to.

49:50

That's a great question because you

49:52

would think this wouldn't really be viable if

49:54

you looked at this through the Quest 2. I

49:57

tried the Quest 2. I saw what you're seeing. I tried it this

49:59

weekend with my daughters. It's one thing for a

50:01

game. It's another if you're trying to you're working in

50:03

the terminal You're composing an email and you're reading text

50:05

for hours huge difference. Yeah. Yeah with a with a

50:07

3d game You don't really care. But when you're trying

50:09

to read text on a monitor big

50:12

difference. So on the quest 3 You

50:14

have a much much better field of vision.

50:17

I'd noticed because I had I had

50:19

probably I mean I my

50:21

screens were at my I zoom my screens to

50:24

test this for you at the maximum width of

50:26

my peripheral vision and At

50:28

the edges of my peripheral vision the monitor

50:30

does kind of have like a little kind

50:32

of a blurry effect to it at the very very Edges, but

50:35

I'm talking like if you held your hands to the edge of where

50:37

you can see them That's where it starts getting blurry yet But

50:40

then the field pretty much from in there to the

50:42

center is looking really sharp And if you use immersed

50:44

make sure you mess around with the encoding options and

50:46

that also makes the text look better I've

50:49

learned so got it. You got it the

50:51

defaults are Optimized for Wi-Fi

50:53

performance and you know, they assume your Wi-Fi is

50:55

crap But if you got good Wi-Fi

50:57

or you hook it up over USB-C, you can step

50:59

up the quality and the text looks even better Bear

51:02

4 5 4 boosted in 5,000 sats Well,

51:06

hey there bear have tried found a few times

51:08

But just keep going back to pocket casts and

51:10

there are a few reasons for that Number

51:12

one the home screen makes it feel a

51:15

lot more like a social media platform than

51:17

a podcast player Let's take these one by

51:19

one. So the home screen making

51:21

it feel like a social media app than a

51:23

podcast player I agreed with two and was the

51:25

biggest friction point I had with fountain initially then

51:28

I discovered like six new podcasts I never

51:30

would have found ever and they're my regular

51:32

listens now and I've changed my tune a

51:34

little bit Then I also I

51:37

started following a couple of other JB listeners

51:39

and they started following me That's

51:41

fun. And now I'm seeing their comments and

51:43

what they're listening to and they see my

51:45

clips and Now it

51:47

makes more sense. But bear I definitely had to

51:49

take I don't know man three

51:51

weeks of just being like I really

51:53

wish you would just open my library. I really wish and now

51:56

when I launch Fountain I

51:58

actually stay on the home screen because I know in

52:00

the library screen it's just downloading my podcast anyway.

52:02

So while it's downloading my podcast, I just scroll

52:04

through the home screen, see what other listeners have

52:06

been boosting or listening

52:09

to. So I know it's not

52:11

exactly what you want to hear, but you might just

52:13

give it a reframe of thought. Bear says number two,

52:15

I find it really hard to see new unplayed podcasts.

52:17

Even when I play an episode, it stays

52:20

on my episodes list. I

52:22

would appreciate an option just to see new

52:25

unplayed episodes. Yeah, or auto delete maybe when

52:27

you finish an episode. So

52:29

they've added swipe to delete that makes that a little easier.

52:31

But yeah, I think right now by default, unplayed

52:33

are still chronologically listed. They just are now

52:35

marked as played. And

52:38

Bear's number three, my biggest gripe is I

52:40

really would like an option to make a

52:42

good use of bigger text. The text in

52:44

the app is just small for me. Right.

52:49

I mean, I wonder, I know on iOS, you

52:51

can pop up the zoom up. I

52:53

don't do that on my paisel, but maybe there's

52:56

an accessibility option there. I'll chat with them about that

52:58

particular bear. I know you had some albi issues too.

53:00

If you hit me up on matrix, I can

53:02

help you troubleshoot that. That's

53:04

good feedback though. And those are things we will be talking

53:06

about in our next sit down next Thursday. Pressly

53:09

weep HD. I'm

53:12

going with it comes in with 2669 sats from

53:14

fountain. And he says I

53:20

got a math problem for you guys. It's

53:22

the sats times

53:24

17 equals my zip code.

53:26

So he's doing a little math game with us. Did

53:28

you bring that map? Oh, you did. This is why

53:30

we're weeping in HD. I make sense now. Yeah, 2669

53:32

times 17. That would be 45373. And

53:34

that seems to

53:43

be a postal code in Miami County, Ohio.

53:46

Oh, cities

53:48

like Troy, alchemy or Staunton.

53:51

So they got a county called Miami. And

53:54

they got a city called Troy. Great.

53:56

This is so great. I mean, you just like where do

53:58

you live? I live in Troy. in

54:01

Miami. You know, Ohio. Thanks,

54:04

Weep. Appreciate

54:07

that. Boasts. Difficulty

54:10

adjustment, Boosin, with ten thousand cents. It's over nine

54:12

thousand! Love your

54:14

work, gentlemen. Um, a

54:17

video option on Fountain would be a game changer.

54:19

Little Burry tells me that may be a possibility.

54:21

I think a lot of the plumbing is there,

54:23

and a lot of the Underline libraries support it.

54:25

We may get there with our lid support for our scale trip.

54:27

We shall see. So

54:30

that'd be taking advantage of alternative enclosure set-ups? Yeah, you

54:32

got it. So Podcasting TotoSpec has the

54:34

alternative enclosure, and in lit too you can do, here's

54:38

my primary stream, and here's an alternative stream. And you basically

54:40

put like HLS or RTMP URL in there. And

54:43

then the client player would just grab that and

54:45

figure it out. So it could happen difficulty adjustment.

54:48

I like that username too. Barnminer comes

54:50

in with five thousand, five hundred and fifty-five sats. He says, tell

54:52

Fountain I switched to Breeze Wallet for Boos since

54:54

it's self-custody. Actually did. And

54:58

Nick's response was yes, that is on the list. It's

55:00

been on the list for a while. It's still on the list. Well

55:04

network Rob sent in a Spaceballs boost.

55:06

So the combination is one, two, three, four, five.

55:10

That's the stupidest combination I've ever heard

55:12

in my life. So

55:14

I'm boosting from Fountain today for a couple reasons. One, giving a

55:16

little extra value back to my stream sats. And

55:20

also number two, scale. I

55:22

know I'm not a fan of that. Number

55:24

two, scale. I know it's a little, but

55:27

I know every little bit helps. Number three,

55:29

keep up the great work. Love the value

55:31

to value. And number four, well since I'm behind

55:33

on episodes, giving a small hello to future Rob

55:36

when I get caught up. Hey,

55:39

thanks for joining me for the next episode of the

55:41

Hated in Pasadena. Thank

55:43

you very much network Rob. Appreciate

55:45

that catch up boost. We look

55:47

forward to hearing from you when you do get caught up. So

55:50

excited about scale in NixCon. So, so

55:52

glad we're going. I feel like we don't

55:54

give enough attention to the other events that are going on, but this

55:56

is the big one that kicks off the year, right? And we only

55:58

have so much to pay for that. in our

56:00

tiny little brains. That is probably really what the

56:02

problem is. Yeah. Ben the tech

56:04

guy comes in with Rodeux. It

56:06

says, theme link is already

56:09

in the meta store. I mentioned I side loaded it.

56:11

And it does support proper VR, but

56:13

only on Windows and Mac OS. Fun,

56:16

uh, no, no, no. They're actively working

56:18

on Linux support. But in the meantime,

56:20

ALVR works pretty well. Okay. I

56:22

didn't even look for it in the meta store. I just

56:24

got, so in the workflow, side loading APKs, I

56:26

was like, I'll just go get it. I don't

56:28

need no store. I don't need no store. This

56:31

is a site of former iOS user folks. What?

56:34

No, don't you think I would have defaulted to the store? I

56:36

mean, you're so excited about the ability to side load at all.

56:38

Yeah, you might be onto something there. Mr.

56:41

Pibb that boosts in with 12,345 cents. Yes,

56:46

that's amazing. I've got the same combination on my

56:48

luggage. My issue with fountain is, I

56:51

use an auxiliary cable to listen to my car and

56:54

the sound quality always gets

56:56

distorted at the necessary volume.

56:58

I use the iPhone app and I'm always

57:01

fiddling with phone volume and car stereo volume

57:03

trying to get it loud enough without distortion.

57:06

I have the same problem on Cast-O-Matic. I

57:08

enjoy the JB shows for the professional sounding

57:10

audio quality and it's great with my Bluetooth

57:12

headphones or a speaker, but just not with

57:14

the auxiliary cable in my car. But

57:18

if I use Spotify or Apple through the AUX,

57:21

it sounds just fine. Now what could

57:23

be causing that? Distortion when you're using

57:25

certain apps, but not

57:27

when you're using Spotify or the Apple apps.

57:29

Are these other apps normalizing, boosting the audio

57:31

to a level where there is some distortion

57:33

at those higher levels, but you just don't

57:36

have to get to the higher levels with

57:38

these other apps? So it's something like

57:40

a louderizing sort of. And it

57:42

only happens when you're using an AUX cable.

57:45

That's interesting. Yeah, if anybody has any

57:47

ideas on this one, let us know.

57:49

Also, remember we're collecting your boost ideas

57:51

for AI resources and tools that we should

57:53

be looking at for the show. So please

57:55

Boost those in. Thank You everybody who did boost in.

57:58

If we didn't read your fountain FM Feedback. The

58:00

word nick and I covered it and we're stuck like in all

58:02

of them in check your profile is probably replied to you directly.

58:05

We. Had twenty four boosters come in and we

58:07

stacked two hundred and twenty Two thousand. Two

58:09

hundred and nineteen sets. Almost a road ducks.

58:11

a big old row a mic ducks so

58:13

I'm I give it to as sank you

58:15

kindly and when I caught a big old

58:17

make time. For. Hit Everybody who boasts

58:19

in is. A. Great time to get a

58:21

new podcasting out because. We. Are working

58:23

on new features. We've. Already

58:25

rolled out to our members. they're loving it's.

58:28

And. They're going to be coming out to the

58:30

main feed very soon and some of them were

58:32

gonna start as we head off to Scale and

58:34

we'd like you to be there with us and

58:37

keep podcasting weird and decentralized podcast apps.com. pod.

58:39

Verse and fountain F M and Cast America really the same.

58:41

It's of the show. You. Pick the one

58:43

that works best for you and and boost and with

58:45

a message and you can support us directly that way.

58:47

No middleman, all using an open source peer to peer

58:49

network. And. They get everybody who stream sats

58:51

to us as well. with preset. You

58:53

and a big shoutout to our members out there. For.

58:56

Using their Fiat's on coupons to keep

58:58

us going on the rags every single

59:00

month, we really appreciate them to. You

59:04

mentioned that a lot of these large language

59:06

models and whatnot are coming out with a

59:08

paper. And. Oh yeah, I

59:10

think one of the best ways. To. Just

59:12

wrap your head. Around.

59:15

What models have come out recently and what problem they're

59:17

trying to solve, and if you even need to pay

59:19

attention and any of that. Is. Probably

59:22

the daily papers over on

59:24

hugging face.com. And. Every

59:26

day. since it is a free day I'd

59:28

pretty why I'm in every week decks. They're

59:30

posting just one white paper after another and

59:32

it's an honor know where? It's a fantastic

59:34

resource for just kind of watching the development

59:37

of these. You must have taken a look

59:39

at this earlier. It's a real I mean

59:41

hugging face.co itself is a resource I I'm

59:43

gonna. I'm gonna assume most people listening to

59:45

this. Already. Know about

59:47

it but as a higher level

59:49

recommendation Hugging face.co. Is. your

59:51

into any of this kind of stuff for the community

59:54

around these open models and datasets and yeah they have

59:56

sponsors but bunch of stuff like that and then they

59:58

have these daily papers And that's

1:00:00

at huggingface.co. papers. And

1:00:03

these are the white papers that accompany the announcement

1:00:05

of these large language models or

1:00:07

whatever they're working on. And I

1:00:09

think just you can look at

1:00:11

the headline, you can look at the summary and know if it's something

1:00:13

that interests you or not and pretty quickly stay on top of all

1:00:15

this stuff. I love resources like

1:00:17

this. If you have any out there, let us know. Yeah. And

1:00:20

you maybe take a look and see just how fast it goes

1:00:22

from paper to something you can actually use. Yeah.

1:00:26

Yeah. I would say if you quit

1:00:28

Reddit recently, you can go here. Even if you don't

1:00:30

want to dive into the papers, just looking at the

1:00:32

names of these projects is such a treat. So just

1:00:34

scroll through the names. I'm enjoying

1:00:36

that. Yeah. I feel like

1:00:38

this is our moment in history as a community to

1:00:41

keep this stuff open and

1:00:43

to keep the open store stuff viable, to keep

1:00:45

the community vibrant. You

1:00:47

guys probably saw that on the back of

1:00:49

all this Gemini news, Google also announced though

1:00:51

a massive, was it like $60 billion

1:00:53

deal or something like that with Reddit to

1:00:56

pull in all the registered to train AI? Mm-hmm.

1:01:00

That cost and

1:01:03

those kinds of deals, those

1:01:05

are going to always be relegated

1:01:07

to the absolute top, right? The big tech companies

1:01:09

that can afford to write $60 bill or $30

1:01:11

bill or $10 bill or whatever it

1:01:13

is, $30 bill. Yeah. I mean,

1:01:15

what, JB tried to buy the Reddit data, but they weren't interested. Yeah. It

1:01:18

turns out they wanted more than a row of ducks. But

1:01:22

like you look at things like Hugging Face, you

1:01:24

look at the invoke AI, you put it all

1:01:26

together and we saw a very, very

1:01:28

viable shot here. And I think ultimately if I

1:01:30

could be a dreamer long-term,

1:01:34

I'd love to hear stories of listeners

1:01:36

that are implementing this stuff in small

1:01:38

businesses or in their workplace, maybe even

1:01:40

large businesses, because that's where

1:01:42

I fear the most that we're going to get locked

1:01:45

into corporate AI. And it's going to be this risk

1:01:48

averse, not very useful,

1:01:51

overly talkative, watered down

1:01:53

crap tool. And

1:01:55

the reality is, and the BSD folks

1:01:57

have always been right about this, is... You

1:02:00

know, a powerful tool you can shoot yourself in the

1:02:02

foot with. And you can do good

1:02:04

and bad with RM-RF. And

1:02:07

I feel like it's the same kind of general principle with

1:02:09

this stuff. So if you're out there in the field, you're

1:02:11

deploying this somewhere where you've got more than a couple of

1:02:13

users using this stuff, boost a

1:02:15

write-in and tell us how that's going. Yeah,

1:02:17

are you relying on AI-powered stuff for your

1:02:20

business? Would you? Do you feel differently about

1:02:22

it if it's an open-source thing versus something

1:02:24

you can only interface with via API? Yeah,

1:02:27

I sure do. It's

1:02:30

just a wild new frontier. And

1:02:32

we're really just kind of at the beginning. I thought we'd

1:02:34

see how NVIDIA went, and depending on how NVIDIA went, we

1:02:37

knew how much momentum there was in the industry. And NVIDIA

1:02:39

popped to the high end for sure. I think

1:02:41

there's still a lot of legs left in this. But

1:02:44

where it actually ends up ultimately and what kind of

1:02:46

end result we get, that is

1:02:48

not so clear at this point. And we, I

1:02:51

believe, still have a chance to shape that future.

1:02:53

See you next week. Same bad time, same bad

1:02:55

state. We'd

1:02:57

love to have you join us live. If you can get in

1:02:59

that mumble room, which is popping this week. Lots of folks in

1:03:01

there. We always hang out with them, and

1:03:04

they can pop in during the show or before and after the show

1:03:06

is really when we just get to chatting. It's

1:03:08

great to hang out, gives us that live vibe. We've got

1:03:10

details on our website for that. And we do the show

1:03:12

on Sundays at noon Pacific, 3 p.m. Eastern. Links

1:03:15

to what we talked about today,

1:03:17

linuxunplugged.com/551 for the tools and the links

1:03:19

and the resources, all that kind of stuff.

1:03:22

If you thought this show was useful or maybe

1:03:24

somebody else should hear it, please share it. That

1:03:27

is the number one way podcasts get marketed. It's

1:03:29

not really through anything else. It's not going to

1:03:31

be Google Ads, I'll tell you that. So, appreciate

1:03:33

that. Thanks so much for joining us

1:03:35

this week. See you next Sunday. You