Episode 818 transcript

FLOSS-818

Jonathan: Hey folks, this week, Doc and Jeff joined me and we talk about DeepSeek. That is the new AI out of China. That's really making waves and a couple of other open source things. It's a lot of fun. You don't want to miss it. This is Floss Weekly episode 818 recorded Tuesday, January 28th. I don't care about the Roman empire.

It's time for Floss Weekly. That's the show about free Libre and open source software. I'm your host, Jonathan Bennett. And today, well, we had a little bit of a scramble. We had a guest. scheduled who had to reschedule. And then we had another guest that started to schedule and then realized that it's about 4 a.

m. For him right now in Australia. And so that's not going to work either. We may have a special recording session later in the week to get that one. I'm pretty excited about that. So I sent an email out, a hurried email out last night that said, ah, help guys, there's news we can talk about news. And so we had a couple of brave souls that answered the call.

We've got Mr. Jeff Massey, and then. Very good friend of the show, Mr. Doc Searles. Welcome to both of you. Hey guys. Hey, really glad to be here.

Doc: Yeah, this is great. It's great to be back.

Jonathan: Yeah. I'm, I'm particularly excited to have Doc back. It seems like it's been too long since we've had you back. Every time I would, I would text you or send you an email and ask like, Hey, do you think you could make it for a show?

And Doc would constantly tell me, no, it's, it's traveling season. It's convention time. I'm out. It's always something. It's always something. Yeah. So. We stay busy. This, this is kind of an interesting opportunity. What, what conventions, what, what, what has been keeping you busy, doc? Is there anything particularly interesting for our show that they would love to hear about?

Doc: There's, there's a lot. They're probably the, the biggest thing that just takes a lot of prep is the twice yearly. Internet, Internet Identity Workshop. And we're coming up on our, I don't know, our 30, our 40th. We've had them two a year since 2005. There's always about 300 people. It's a lot of stuff's come out of there.

Um, OAuth and just all kinds of things having to do with identity, but lots of other topics. And that's the kind of a self sustaining thing that's at the Computer History Museum in Mountain View, California. But we also organize some events here at Indiana University. So, where Joyce and I are both visiting scholars, which is Half right, in my case, I am visiting but we like Bloomington enough that we moved here and so we've been very busy building a house in which I'm now in and and this is if the room sounds reverberant, it's because there's no furniture in it other than being a computer and a microphone and not much else.

So yeah, I mean, and, and, and I'm writing a lot. We've actually I've been following, um, the fires in California because Joyce is from Los Angeles and we know, we either know personally or are no more than one or two handshakes away from many dozens of people who've lost their homes. So that's been And I've been writing about that on my blog and especially covering the coverage because I'm still a journalist and I don't get paid, but journalists don't get paid anymore for the most part.

So unless they sell out and I haven't, I haven't been smart enough to do that in a while. So that's been fun.

Jonathan: So are you guys no longer doing the two houses in two different states sort of deal?

Doc: Oh no. It's like we have two houses in Indiana, one in California, and we still have. We partly pay for our old apartment in New York, which my son, our son occupies.

And he's getting ready to go to law school next year, so somewhere. It'll probably be in New York, but if not, somewhere else. We'll see how that goes, and we may lose that.

Jonathan: So you simplified things. Now you only have four houses in three states.

Doc: We only have four houses in three states. That's right. Which sounds so much fancier than it really is.

It would be great if one of them was like an estate or something, but Nope, they're just four houses.

Jonathan: So out of the out of the identity workshops, the internet identity workshops, is there anything particularly interesting that, that came out of this last one that was, what was the hot topic?

Doc: Actually, I write, so there are two things.

One, one is that I wrote this, I wrote this book called the, an earlier book called the Cloutier Manifesto was very popular and hot. It's actually 25 years since that came out. That was a bestseller. Right. It's been wrong now for 25 years. It's very optimistic about the future. That, that future is one in which we're all trapped now.

It hasn't quite happened, but people love it. The other one was called The Intention Economy, which turned out to be The inspiration, it was a worst seller. It's like number two million on Amazon, but Tim Bergers Lee based a solid project on it. Kawhi, which is this open source personal AI project is based on it.

And And more recently, Consumer Reports, the, the, the organization that rates products and all that kind of stuff has decided that they, or at least their Skunk Works inside of it, but I think it's an important Skunk Works that they want to base the future of the magazine, I think, to some degree on what.

Not just the services they'll do on what I wrote about in that book. So that, that's, and they were there at IW this last time. Taking input and engaging in dialogue and planning things out. And very optimistic about where that's going to go. And another thing, which is this workshop that I have to jump to on the hour, is a new thing called IRA, A Y R A.

And I think it's at ira. forum, A Y R A. Yeah, Aira. Forum, A Y R A, and a trusted relationship network, but basically it's, if it works out and there's open source stuff in it, I don't, I won't know until this workshop exactly what's going on with it, but but it's, it's, it's the new, it's, it, it will do for identity kind of what credit cards did for, money, another, a better way to do it.

So, and I, and as I'm looking for things, I lost the zoom. So there you go. So that, that,

Jonathan: that might not be the best analogy doc I hear. No, it might not do for money. I think, Oh no, it's going to enslave all of us and make everybody poor.

Doc: I mean. If you look at, if you look at what, what D Hock, the guy who invented Visa did, basically got a lot of competing banks to get along and gave us this very convenient thing we can carry around in our wallet.

Now, all the bad things that happened since then are beside the point that it made that a lot easier. So. IRA is a, is a cabal of a lot of companies that none of them are like big household names, but there are some pretty good sized ones and some banks is based in Switzerland. But included in that is going to be IEEE P 7012, which is.

Machine readable personal privacy terms. And this is something that we started in 2013 or 14. The standard is finally being finished. It's, it's finished on an asymptote. I mean, it's never going to quite arrive, I think, but it, but we're, we're really, really, really close. But it will completely, if this works out, it'll completely reverse the way we do we do privacy online.

They agree to our terms rather than we always agree to theirs. We blow up the entire cookie bullshit. We, you go to a website, you go to a service, you have your terms. It's just like creative comments. Where I have this, I have a term, I've chosen it. It's that one there. And you can accept that or not, but if you accept it, we both have a copy of it and we agree and we can adjudicate it later if we have a dispute, but we probably won't and because we're not trying to screw each other and this is friendly and and that's what we've been working on with customer comments for some time.

But I think it's quite simply the most important standard in development today. And I, and I Pretty sure we'll have it done real soon.

Jonathan: So

that's, that's where

we are with that. Yeah. So this is an interesting thought. The idea is that I, I have, I've picked out, so I'm assuming that there's going to be like an onboarding wizard for this.

At some point I pick out, these are the things that I think are appropriate for a site to do. These are things that are not appropriate for a site to do with my. data, our business relationship, whatever. And then I go to a new website and they have done essentially the same thing. And so then there's going to be this algorithm that compares the two and we'll say, well, you're trying to access this site, but they're asking it.

So it's going to be like, it's on a web area, a phone application, right? This site is asking for these additional permissions. Like that's inevitably what's going to happen.

Doc: It's actually not that complicated. It's there's no. There's no centrality to it. You have in your browser or your app, whatever it's going to be, your agent, all these things have been called agents now, by the way but you have an agent and it says, it's, in a way it's probably just going to be just like.

Do Not Track, only it has teeth. And Do Not Track was way the hell ahead of its time. It just was, it was the wrong time to build that sandcastle. You know, before the tsunami. You know. Anyway but I think privacy matters a whole lot more now. Everybody wants it. And the. The GDPR is that it's thing, but it's kind of failing.

And so now it's, it's a good time to have something where, Hey, the user is not just the user. This is an independent human being and it has this terms and, and it's a simple term. Like one of them is, is called no stock music. Just don't track me out of here. Okay. And don't sell my data to anybody. Give my data away to anybody else.

But you and I, you could track me while I'm here. That's one of them. And and the, and, and it's, and they either accept it or they don't. And, and their row, their, their agent on their side says yes or no. And they'll notice that this is you know, term number one on the list of 10 that are on customer commons.

And that's a known thing. And it'll have a, you know, a unique ID for that one. It's just like, just like creative commons has, you know CC by, or whatever, whatever that they're like, like there's six. at Creative Commons. It'll be that simple. You know, it, you, it's a contract. It's, it's not more complicated than that.

Contracts are like the oldest legal ceremony in the world. You know, it's just, you know, I rent a wheelbarrow from you, I give you forty bucks in deposit, I sign something that says I'm gonna bring it back in good shape, you keep, keep the deposit and I bring it back and we've got a deal. It's over. It is contracts are laws that two parties make.

for themselves with each other. And contract law really only applies when there's a dispute. And that'll apply here too. But it doesn't have to be complicated. And we're optimistic that, for example, that WordPress will pick up on this. WordPress runs like 40 or 50 percent of the world's websites and most of them are not based on, on tracking you.

Don't make money on advertising that way. I think this will free up sites and services to make money a better way. You know, than just advertising or get better advertising because you're asking for exactly what you want. I mean, there's just lots of possibilities here that don't start with surveillance.

That's basically it. Like, hey, hey, world, stop starting with surveillance and then let's work this out. That's basically what it is. But we don't even have room. We have room for negotiation, but the assumption is There's not going to be negotiation involved. This is simple. I don't want to be tracked off your place.

Good, you got that? Great. I

Jonathan: can't help but think that the devil in this case, the devil in the details is going to be exceptions. Right? Like there's going to be some valid use case someone has for, I, we need to be able to do tracking. And so then there's going to have to be an extension added to this where it says, this website is asking permission to track you.

And it'll be, you know, an addendum to the contract. Yeah, and then, yeah,

Doc: and that, that will happen. And I think, I mean, in the beginning with this, I mean, All kinds of that stuff is possible. If we start there though, we're going to be exactly where we are now. You know, um, if, if we start with the assumption that privacy is the default, okay, I have privacy here.

It's, it's basically, that's what we have in the regular world, in the everyday world. It's true. You know, we don't walk around wearing name tags and we don't walk around with. with the fuzzy side of Velcro all over ourselves so that, so that, so that the prickly side of Velcro can be stuck to us everywhere we go.

Right. You know, that's, that's not what we do in the everyday world and, and we want to replicate online something that resembles the everyday world, but we're, there are tacit understandings. There's, if you look at two kinds of knowledge, there's explicit knowledge, there's tacit knowledge. Tacit knowledge is You know, I know how to drive a hammer with a, a nail with a hammer, but I can't explain it.

That's tacit knowledge. We have, the world is full of tacit agreements. We have tacit agreements, we're not going to track everybody everywhere they go in the everyday world. We can't have them in the digital world because everything needs to be explicit. So this is a way of making some things explicit.

Like, we're going to start with privacy. We have privacy. What a layer of stuff on to that, we'll work that out, but we'll start here. We'll start with We're all private here and then we can work out other stuff. So and, and our, and our like 10 terms that maybe they're 12 now to worked out by some people who are in our little cabal that are in, in Scotland and Ireland right now.

And and I think they have 10, but they have graduated ones where, okay, some of my data could be used, for example, to train AI. Some of it doesn't have to, you know, you can add that on like, I was going to ask. And we can transition from that to, you know, something else. So,

Jonathan: yeah, that, that that sounds good because there is an AI story that has just exploded upon the news scene in these past couple of days.

Oh, it's huge. And that is it's Deep Seek. Right? It's the new AI out of China from a Chinese company. And I think the reason that it's so huge right now is not necessarily because it's the best AI that's ever been made. I think the reason it's so huge is because it's an AI that was made for like 6.

5 million dollars. As opposed to the billions of dollars that places like OpenAI and all of those are throwing at this particular problem.

Jeff: Well, do we know for sure it really cost that? I mean, they just said that, right? Right. Yeah.

Jonathan: So that's, that is one of the things, right? Like, so we don't know for sure exactly how much was put into this.

This is a claim that is coming from a, you know, a relatively, Unknown corporation out of China. They are, they are claiming that is an open source AI and I've been chatting back and forth with Simon who, you know, had a hand in creating the definition of what it means for an AI to be open source.

And his take on it was it is more open than less, less closed is what he said. It is less closed than Llama from Meta. But, yeah. So far, they have not been able to find the training data. So it does fail that test of what OSI considers an open source AI. Right, so it's not a completely open source AI by that definition, because the training data is not published.

But you can. Git clone it, and run it locally. Lots of people have been doing that, and have had good success with it. I have been playing with that a little bit, last night and today. So far I've not been able to make it work on my AMD video card. Although, there was, there was part of the story was that it didn't necessarily rely on CUDA.

Which is interesting, so maybe it will work on AMD cards, I don't know. But, yeah, the, the big thing is that it came out of nowhere, and apparently Was so cheap to make. And so NVIDIA stock is cratered as a result. I think Meta stock is cratered.

Doc: It bounced down, but I, I'm certain it'll go back. Oh, I'm sure.

Jeff: I mean, yeah, it's just, you know, a lot of semiconductor stocks went down, but the, one of the official statements from. NVIDIA was well, it doesn't matter what the engine is. You still have to have the hardware to run it on cheap cheap or not It's still gonna be a lot of purchase of hardware. That's what they're banking to get the confidence back into the market, right?

Jonathan: So that's it. That's actually an interesting thought though because NVIDIA right now They're making their money by selling their hardware to meta and open AI and Twitter X Because it takes so much horsepower to do the back end work of actually training these things. Well, if we're going to see a swing back where it becomes easier to train them on the front end, And then you, you do your, your, your inference work at the user facing side.

That kind of suggests that NVIDIA is going to start selling more cards to consumers, more consumer level cards, perhaps. Does that, does that check out?

Jeff: Yeah, I mean, it, it really does, I think, which is might even tie into their. Recent 5090 launch where it's kind of geared less to the gamer and more towards the prosumer or home AI Enthusiast where they comes with 32 gigs of RAM in it to help train and you know, there's a lot of AI Specialized chips in there to just help do those computations that the specialized Neuronet computations and so I could totally see that You know, and less, less of the monster racks, just full of GPUs that take, you know, ungodly amounts of power and cooling and,

Doc: And, and have to know the whole world, you know, where I think, I think when, when this stuff actually does get personal, which is you have a box.

You know, it's either your laptop that you have already, or it's a second box that's a specialty box that, that's going to do your AI for you, but, but helps you run your life. You know, what, what's my data that I need to know? What, what are, what are my travel, my calendar, my contacts, my finances, my all the stuff that's out of control in our lives right now.

You know, especially when you start gathering it up, like, I mean, I have no idea what TV series I've watched. I don't remember them. But you know what? Our Roku TV's been narcing on me to, for, to God knows who for some time. They have that data. If you have a late model car, it is busy reporting on you to insurance companies at all times.

Your car comes with a cell phone for that matter, the cell phone you carry in your pocket has all kinds of travel data that you don't have access to, but Google does and Apple does. What about when we have that data and, and, you know, who, you know, who have I had lunch with the most? When, who, where was I last Tuesday?

What, you know, what is all my health data? How can I pull all that together? That's, to me, that's the most interesting stuff in the longer run. It's not the, you know mean, it was wonderful that we can have Claude give us a better recipe and, and I can do all kinds of research, right? Which I do. I mean, I spend.

you know, probably two hours a day on an AI of some kind or another doing research. But but I think the interesting stuff for us is going to be what we do on our desks, you know, in our phones.

Jeff: Well, and I could totally see that because I, I think, you know, a good case in point is Leo Laporte, where he, he ran his own AI and he fed his programming books into it.

Was it Go? I, I forget what the exact language was, but it, he said, okay, I wanna program. And I'm going to use this language. He had open source books that he fed in and said, okay, only use the. The source material from these books. So it was a very specialized AI. And if you start thinking, okay, I'm going to have an AI at home.

I'm going to do some programming. I want to know what's going to be on the next TV series I want to watch. You can feed in what you like and don't like. You wouldn't, wouldn't have to You wouldn't need so much data when you, when you start getting these very specialized niches to you know, have it, have it work on, you don't, you don't need, you know, if I go, Oh, I don't care about the Roman empire and I don't care about, you know, the rainfall in Uruguay or something, Oh, well, there's a lot of data that can.

You can start just filtering out to make the data sets more manageable. And thank you, mashed potato. Lisp was the language I was looking for and Emacs.

Jonathan: I can't help but think of what George Hotz, AKA GeoHotz is doing with tiny corp and the tiny box, which tiny is maybe not the right term for this, but he is putting together AI boxes and he's got a couple of them that are sort of.

Aimed at the consumer. Now, the cheapest one is 15, 000 and it's got six, you know, 7, 900 XTX cards in it, right? So these are not cheap and they're not tiny, but compared to a full data center, they are, they're, they're affordable.

Doc: Wow. The more expensive ones. I'm looking at this now that the red one, which is the cheaper one, 15 grand, The green one is 25 and the pro one is 40 and the latter two are sold out.

Jonathan: Yeah. Yeah. Yeah. Wow. But his, you know, his Focus with that is the same thing. Let, let people run their own little AI engines. Do their own machine learning project. You don't have to have something that big. So I've, I've, I mentioned I was spent yesterday and a little bit this morning trying to get The deep seek engine running on my machine.

I did not get that going, at least not yet. I will continue to try to do that, but what I have gotten on my, my poor little single AMD card, which is multiple generations old now I've got a text to image working fairly well. Now they're not creating great images yet, but it's, it's doing it. It's working.

In fact, this is the first time that I've, I've gotten that working on this machine. Which is, it's cool. Like, it's really neat to be able to do that on a local machine. And, you know, not using somebody else's service that does who knows what with the images you create and the text that you create and all of that.

It, it is interesting to see this. Sort of step further and further towards the consumer, towards the user. It's becoming a little bit more accessible for us, for everyday people.

Doc: And I think what happens is when this reaches critical mass, when this thing is 15, 000 and, or the box that you've put together, you know, your AMD box You know, becomes the price of an ATM withdrawal and somebody makes a cheap way to get a zillion of them out there.

And we're going to do more, so much more with this stuff than, than the bigs can even imagine at this point. I think there's going to be so much invention that happens at the edge. It's like, I mean. I, I watched this happen with PCs, you know, it's like when PCs came along in the late 70s it was like, Oh, you could put your recipes on there and you could do some art and you know, but then VisiCalc happened and then, you know, and all of a sudden it's like, I mean, if you told a, an MIS director, which is what an IT director was back then, an MIS director.

You know, you're you're gonna let PCs into your company and they'll say, no, no, no, we have a mainframe here. We don't need that. Right? We have guys in lab coats. And, and two years later, the, the buildings are full of PCs because people could do more with that, with these things than the, than the bigs could.

I think we're, now the bigs here could do so much more than mainframes ever could way back then. But the personal computing revolution was mass, was massive. And I think something like that can happen with AI as well. Once we have independence and power and it's ours personally, and not just corporately.

Was there, it was personal, not personalized.

Jonathan: Was there a feeling doc back in the day that PCs were a bubble? You know, we talk about the crypto bubble. Oh yeah.

Doc: No, not only that, it was a bubble in a sense there. The the the IBM pc, which really came out of a, a skunk works in Boca Raton, Florida. Was, you know, was a big hit in business, but it wasn't a big hit personally because it was too expensive for ordinary people.

It's still like three grand. You know, you, oh, you buy that. Then you get the peripherals. You put it in the back plane, right? And that wasn't, oh, you need an ethernet card. You need a storage card. You need a bunch of other things. Right. And. You need the micro domain frame card. I worked for the company that sold those.

So you could pretend to be a VT 100 or a 32 70 display terminal. But, and then the Macintosh came out Apple had a failure with the Lisa, which was 10, 000 and. But the Macintosh was a kind of a hit, but still kind of expensive for people. And there was a huge crash in like 84 in Silicon Valley.

And Atari was in the business. Commodore, a bunch of other companies kind of Came and went but, but they pulled out of it because the damn things proved too useful. I mean, that's really what happened. Yeah. They became too useful.

Jonathan: So I have, I have pointed out several times you hear on this show and on others that we are, we are obviously in an AI bubble, but the AI bubble reminds me of the.

com bubble. And I suppose it reminds me now and hearing this, it reminds me of the, the, the PC bubble back in the seventies. And it reminds me of that in this way, things are, Hotter than they should be, right? And there will eventually be a crash. Maybe, maybe this is the crash that we've been waiting for. I don't know, with DeepSea coming out, maybe this is the crash.

And the, the dumb businesses are going to go away. You know, the whole, let's add AI to your blender. Now with AI, you're AI powered coffee pots. Like hopefully at some point that ridiculousness is going to go away. Very similar to the. com bubble and the PC bubble. It's still going to change everything.

It's just not going to change everything in necessarily the way that we thought it would.

Doc: Right,

Jeff: right. I think

Jonathan: that's

Jeff: absolutely true. Maybe I'm making a bad analogy here, but I wonder if we could almost. Have this, the, be analogous to, remember when PC started getting cheaper and then laptops and note little notepads came out and people are like, Oh, this is the death of the PC.

It's going to go, nobody will have a PC in a few years. Well, I'm wondering if AI is going to be somewhat similar, only the PC is going to be a really large language models, you know, your chat GPTs and so on, but then you're going to have these smaller, more personalized units that took over You know, it would be the same as like your little notepads first, because just like when they said the PC is going to die, well, it didn't because it was just people, not everybody needed a PC, you know, grandma's just writing some email to the kids or somebody who's just kind of surfing the web.

They're not creating, they're not really doing anything with a heavy lift. They don't need the whole big. PC, the expensive PC when it's like, Oh, this little tiny, you know, now it's phones can do a lot of this, but you still need, when you're content creating, doing a lot of bigger stuff, you have to have that horsepower.

Well, maybe, you know, the chat GPTs and the other large language models, there's going to be a space for them where you really need the power, but a lot of stuff, you know, you're going to be able to specialize and it's like, well, I don't. I don't need all that power. I just can have my specialized topics on my home PC or my device and that will, that smaller model with limited data will, will do what I need most of the time for my specialized things I care about.

Jonathan: You know, I, I could, I could even see a future where you have, like, NVMe, I know there are some GPUs that have slots for NVMe's on them. Right? I could see a future where you've got a 1TB NVMe that's on your compute module, like it's connected to your GPU, and that's where your bit and most of the power for your LLM comes from.

And that just gets maybe updated every once in a while, because once these things are trained it can, it can take new information in and sort of work with the new information. I, I think, I think part of this becoming useful for everyday people. Not having to deal with those huge data sets is going to be a big part of it.

But I mean, let's, let's talk, let's think about like all of the other things, like why, for example, why did the cloud take off? Why is that a thing that everybody wants to do? Well, it's because nobody wants to run their own servers. Fundamentally. And I, I kind of wonder whether there's a similar problem, nobody wants to run their own AI.

Because it's still fiddly right now to get it set up.

Doc: Oh, we're not even close to that yet, I think, in terms of the convenience. But, you know, it's kind of, you put the resources where you need them. Obviously, I mean, like we're getting fiber here shortly. And so I'm going to have two gigabytes, symmetrical.

Back when I had You know, a few kilobits, I could get 16 IP addresses and I could set up a server and I could, I ran my own email and I ran my own website and it was all in my house. You know, I mean, I had 16 IP address, here's 16 fixed, there's not even thinkable now, right? You're going to have to use the service.

And it makes total sense. I mean. These guys have set up giant server farms and they're up 100 percent uptime and you're, you know, you pay rent for that and, and they could assume that privacy has worked out at least to some degree. And but there are going to be things that you need locally. I mean, you need the phone in your hand, you need the laptop in your, in your, in your, in your briefcase, you know, it's or on your desk or on your lap.

There's, there's, there. We're going to need compute resources on site and in our pockets and things like that and and we'll work it out But we need the invention that mothers the necessity. We don't have those yet You know, that's that's the missing piece and I have to get off. I hate to say this. Yep understood, but you weren't it's fine I'll leave you with one thing, for the next time you talk to Leo, have you noticed that um, Google's Notebook LM, which will take anything and turn it into a podcast, that the male voice in the podcast has to be trained on, on Leo.

It sounds like a young Leo of some sort. It's sort of interesting. Listen to one of those, one of these times where they're, they're really, they're really weird that they put this podcast together, but it's, you know, it's good, but anyway. They knocked off Leo for that, for the sound of that. Huh?

Jonathan: That's funny.

All right. Thank you, Doc, for being here. Appreciate it, man. Thank you. Talk to you. Good to see you both. Okay.

Doc: Bye bye.

Jonathan: All right. So, you know, talking about that idea of, of nobody wants to run their own servers, I wonder if you could sort of make an observation that we all run our own servers. They're just kind of become invisible, you know, because there, there are daemons that are built into the phones and there are daemons that are built into our, our desktops and all of that.

It would be, it would be interesting to kind of do, try to step through all of the different things running on a device and figure out, like, how many of these are invisible servers?

Jeff: Well, I was even thinking of another kind of analogy when you look at, so I run my own server, like a Plex server, and I have all my videos, you know, my own movies and music on it, so it takes care of what I want.

But it doesn't fully replace Netflix, Amazon Prime, Hulu, anything like that. So, I guess that's another way to look at it of, it's, it's gonna be smaller targets, I think, where the next real innovation in AI is gonna be. Which, if you look at the hardware, there's already, almost every CPU now is coming out with, they have a little neural processor in there for AI, I mean, even the pretty small, low power ones have that specialized chip in there, and we're not really utilizing it right now, but it's there, you know,

is the data going to become the real commodity? You know, okay, I've got AI everywhere, I've got all these little AIs. Oh, you want access to the data? What, you know, what are you going to train it on? Are you going to be limited?

Jonathan: Yeah, there's definitely some interesting questions in there about that. And it kind of, we're kind of at the point now where all of that, and it's a quirk of the way copyright law works, that like once you take all that data and put it through an artificial intelligence training it, the legal theory is that it strips all of the copyright away from that data.

And we, we had a lawyer on the show back a few weeks ago, and he sort of looked at that and said the genie is out of the bottle. There's no putting it back where that's just kind of because that's what we've done. That's where we're just going to have to live from now on. I thought that was really interesting.

So, you know, it depends, it depends on which side of the data you're talking about. One of the other things that I find really interesting with AI right now is They, well, you know, they, they use terms like trustworthy AI. But one of the things that you really, you have to dive, kind of dive into is, is this AI going to give, well, one, everybody, is it going to give me the right answers?

Is it going to give everybody the same answer? Is it going to give me the same answer every time? And I, I did some looking into this for a security article. And, you know, on AIs, on LLMs, there's a setting called temperature. And if you set temperature to zero, it is supposed to give you the same, you know, a given input gives you the same output every time.

If you turn the temperature up, it randomizes the output so that it gives you something different, because that can be useful to get a little bit different output every time. And there is at least one of the AI models that even in, I think it's I think it's OpenAI, one of their, one of their leading ones if you turn the temperature all the way to zero so that it should be you know, a completely deterministic model.

It will still give you different answers every time. Because, as it's loading those tokens from the disk and putting them into the GPU, it does it in batches. And which tokens end up in which batch, from what I could tell, depend upon the load speed off the disk. And sometimes the load speed off the disc is slightly randomized because physics.

And I found that really, really fascinating that some of these AIs are not deterministic just by not necessarily intended to be that way, but like a quirk of the way they're designed. They're non deterministic.

Jeff: Huh. That's really interesting. I, now, you know, I, I will put this caveat. I've only used a couple different AI is basically Google's Gemini and Microsoft's Copilot.

So I haven't played with ChatGPT, but I've put in documents that I've written into each of them and just say, okay, clean this up, you know, make sure there's no grammatical errors and spelling and, you know, sometimes maybe I use a little too much slang or whatever, you know, clean it up and make it, make it more.

You know, generally appropriate so that a larger audience can understand it. And I found it interesting that in my experience, co pilot does a pretty good job of just, you know, Oh, we've changed the phrasing here and you had mixed some tenses and, you know, maybe changed the vernacular slightly, but, you know, and improved the document.

Whenever I'd use, try that with Gemini, I would come up with a document that was different meaning. You know, I'd, I'd read through it and go, wait a minute, this, this changed the meaning or, you know, it, it, this phrase was credited to somebody that didn't actually say it. And it's like, You know, it's temperature was really high and, you know, kind of really, really my document out was not my document in at all.

Jonathan: And that kind of gets to like the danger that we've seen with people trying to overuse LLMs for different business things. So some of the some of the most hilarious cases were like lawyers. And they would go to the LLM and say, you know, here's the thing I want to prove, find me case law, and the LLM would hallucinate case law, and then it was discovered, and I think at least one lawyer, probably several, got disbarred over that because that's a definite no no.

So, you know, don't, don't trust your LLM. Who, who is it? Is it Simon? It may be Simon that makes the statement that LLMs are not telling you the truth. They're just trying to convince you. Yeah, I

Jeff: like that. Yeah, that's, that's I know in business, whenever you use them, they tell you, check the output because it's not always right.

And I always double check, you know, whenever I say, hey, clean this up, or, you know, it's, it's nice for some business documents to, Okay, but Here's what I'm trying to say. Put this in a nice business format and it'll format it nicely for you, but you have to read it because sometimes it's like, where did that come from?

You know, it goes off into left field and

Jonathan: yeah, well, there were a couple of other open source stories from this week that I thought were pretty interesting that we might touch on briefly. One of the big ones is that Microsoft has open sourced a new database format. Did you see this? What are they calling it?

DocumentDB or something? FerretDB? I did.

Jeff: Yeah. I'm kind of wondering what Microsoft is doing. They're open sourcing a lot of stuff lately. And I don't know What's the long game? That's, that's kind of what I want to understand because there has to be a business reason for it. Whatever that is. And

right now I'm just not sure what it is. It's working for them. Yeah. Other than I have said on the other podcast is I think Microsoft is trying to basically Make its operating system the similar to what Apple did. It's going to be open source underneath. Eventually be the Linux kernel, and they'll have their own custom interface to it.

And that way they can take a lot of the resources that are working on the operating system and have them working on cloud and AI and other other things. And the under under the hood just runs. through the open source.

Jonathan: I, ironically, I think you're absolutely right. I don't know about windows, but what you, what you just described pretty much encapsulates what Azure is these days, their big cloud system.

And that's where a lot of this stuff gets run. So, you know, Microsoft is sort of becoming a cloud company where you just you, you rent the resources from them, and you use Microsoft because they're the one that wrote all of these tools and therefore it's going to work well on their cloud. They're, they're moving into An open source friendly cloud company, which is very interesting to see.

So they're sort of trying to become Amazon.

Jeff: Oh yeah, and that's, that's a lot of money, you know, because Amazon doesn't make very much on their web store. When you buy a, you know, bag of cookies from them, it's all the web services and the cloud. And that's where, you know, they're making hand over fist. And I think, you know, Windows is not making them.

Well, I should say it makes them a lot of money, but it's kind of a stagnant area. I mean, they're not going to double their market share. They have basically, what, 95 percent of the market or, you know, somewhere thereabouts. And it's kind of where it's going to sit for a while.

Jonathan: Well, and you

Jeff: mentioned

Jonathan: it earlier, Microsoft saw the writing on the wall that maybe the desktop, the PC is going to go away in some ways.

Right? I think that was part of their their, the push to get them to look into new markets. And they tried, they tried the Windows phone, and that obviously did not work out very well for them. And so, you know, Microsoft sort of looked in the future and saw, well, what do we do when nobody is running desktops anymore, when the desktop market is tiny?

What is our company going to look like then?

Jeff: Well, and they're back to innovating again. I mean, they're trying a lot of things. I mean, you know, and I've said this many times, I worry when a non technology minded person runs a technology company. You know, that we The bomber years, or you can look at Intel when they had I can't, can't think of the person before yeah, against look good before guns.

I don't know who it was either. Yeah, but they weren't technology minded. And, you know, there's several other examples where can cause problems. Bob Swan was directly before Gelsinger. Yeah. Yes. So, I mean, I think now that Microsoft is back into technology and into innovation, that's, you know, there's a lot of stories where they're, they're opening things up and they're becoming more compatible with Linux.

They're open sourcing, they're, and I, and I think it probably eases the transition to The future because people say, Oh, Linux. Yeah. Ha ha. You know, is it? But people don't realize the world runs on Linux. You know, the desktop is like the one place that Linux isn't dominant. You know, you can argue that the handheld devices, the big servers, the big infrastructure stuff.

Pretty much all runs on Linux and desktop is the only thing that isn't. And well, if they're going to go cloud, they're going to make money. They better be compatible with the operating system. That's running it all.

Jonathan: Yeah. The most fun thing I ever, ever heard somebody do with that was if I said a Linux conference, the guy's like, how many of you run Linux on the desktop and a few hands went up.

It's like, how many of you used Google? In the past 24 hours, everybody's hand went up. It's like, well, you know, you were running then a service based on Linux on your desktop. It was just sort of this, this sort of a, you know, interestingly different way to look at it. Yeah, you know, when, when Microsoft bought GitHub, that, that was kind of the one that really, yeah, that was the one that really scared me the most.

Microsoft bought GitHub and a lot of people were already at that point using GitHub. That was kind of scary. That worried a lot of us. There was a bunch of GitLab installs that that popped up right after that. But I must say, you know, so far they have shepherded GitHub very well. And it's, it's grown and become more useful as a service, not less.

Jeff: Well, and it wasn't that, but, you know, when they bought it, it wasn't that far off of the, was it Embrace, Enhance, Extinguish era? Mm hmm. So, I mean, that was pretty close in the rear view mirror. So there's a lot of people that I, I totally get the. You know, when I heard about it, I thought, oh, you know, this can't be good.

But so far, you know, knock on wood, it's going great. And it doesn't seem like it's the Microsoft of old anymore. It's, they've, they've kind of evolved to change with the markets, change with, you know, what's, what's important and where their next revenue stream is going to come from. And, and, and realistically, you have to, you know, otherwise you're going to wind up The next Kodak or Xerox or, you know, you're

Jonathan: just not.

So there is a there's another story that sort of is the inverse of that. And I'm not sure if you've seen this one SimGrep and the OpenGrep fork. Did you get to look at this one at all? No, I haven't. Okay, so Semgrep is it's an open source security tool. It does, I believe it's offline code scanning, right?

Like it's it looks at your source code and it tries to find common errors in it. It's like a static analysis tool, I believe, is what you would call it. And they Made changes to their license back a few months ago You know, they they essentially added to their open source offering. They added a non compete clause is essentially what they did which is the same this is the same direction that we've seen with other source available licenses like the the Open business license.

I think it's one of the other ones and you know, we, we've seen multiple, multiple companies go in that direction. And what's interesting is with Semgrep, the exact same thing happened. The, the outside of the company group of people that were using it went to the last release that was under that. that truly open source license and said, okay, we're going to start at this point.

We're going to fork it and you know, going, going forwards, it's, it's no longer to be SimGrep and we're not going to use your new silly license. And you know, OpenGrep is the is the name that they chose. I'm trying to find the list of who all is involved in it. Oh, interesting. So in 2020, the the, the company behind Simgrip took a, a round of VC funding.

And then in 2023, they took a 53 million series C round of venture capital funding. Yeah, it's probably not a good sign when an open source company takes. Venture capital.

Jeff: No, I, I, in general, I'm not a fan of that at all. I I've seen a lot of companies kind of turn because they, they want their money back.

It's an investment.

Jonathan: Well, yeah, you've been, and so what, what ends up happening is you've got these outside investors that in most cases don't understand the company, right? They don't, they don't understand open source. They don't understand why the company was successful to start with. And it's just like, well, okay, we've given you a year now.

And you're not making the millions of dollars that you promised us. We now have this big lever in the form of stock ownership. It's time to make changes. Here's the changes we want you to make. And that sort of sometimes gets rammed through and does not always work well. In fact, I would say it usually does not work

Jeff: well.

Well, you know, I, I would argue though, if you, if you have an open source company, okay, I'm, I'm running the next, you know, super grip we'll say whatever. And it's becoming really popular. And someone said, Oh, we want to invest all this money. I think one of the things I would have to say is, look, this is not going to probably make, you know, we want to give you 50 million.

Well, it's probably not ever going to pay that out or not in any kind of reasonable time frame that you would ever want to see. You know, there would be better investments to make. Right. Because, you know, I, but I mean, maybe people get lured in by the dollar signs and want to, oh, we want a big investment.

But that's not what. Open source, you know, I mean, it can make money, but it's, it's, I don't see open source ever being, you know, a, you know, a Microsoft, a, anything like that, you know, you have a, you have a few, you have a

Jonathan: few unicorns that can do that. Right? Like, I mean, Red Hat. Red Hat was is still a completely open source company.

Well, almost, almost completely. I think they have a few closed source products. But they got, they got huge, and that acquisition by IBM was quite large. There are a few unicorns that really become big like that, but in general, you're right. I don't know if that is a failure with the open source funding model.

Like the, the ways that people make money with it, or if that's just sort of designed in, maybe it's, it's that way by design and that's a good thing. But you, you don't tend to see open source companies become huge like some of the others do.

Jeff: Well, and I guess, I mean, Red Hat's big, but it's still not private sector big.

I mean, there's, it's, it's one of the biggest examples. But you know, it's near the, near the top of the list, if not the top of the list, but it's not going to be, I actually, I'd have to, I don't know what their revenue was, you know, say even last year.

Jonathan: I mean, IBM, IBM acquired red hat for 34 billion with a B that's, that's chump change compared to somebody like Exxon mobile or, you know, where a company like say Tesla was at the top of their valuation, but that's, that's still quite a bit.

Yeah. Well, I mean,

I'm, I'm curious now. I guess Apple is also one of the ones at the top.

Jeff: Oh my goodness. It says, well, this is, this is old 2019. They said 3. 4 billion. So I guess, I mean, it was, it was bigger than I thought it was. I didn't realize it was in the billions.

Now I don't know what that is after. I don't know what the net income is.

Jonathan: You would imagine quite a bit because they don't have like manufacturing expenses and they have a lot of people expenses and paying a lot of salaries for engineers to work on stuff. You know, it's interesting with, with some of the legal changes happening, like in the European Union, there, there is sort of this new funding model that's, is beginning to suggest itself for open source.

And I, I don't know that this is going to create more unicorns, but I think it'll be a, a good source of funding. And that is essentially, you know, companies use open source products, open source license, libraries, and it's like, oh, we need your help. To get a software bill of materials because the law requires it.

We need your help to make sure that we close vulnerabilities because the law requires it. And you know, the, the appropriate answer for an open source library and open source developer to give is sure. I will be glad to help you. Here's my hourly rate, or let's negotiate a contract where you support the project and we support you as a business.

And you know, there is this opportunity that, that we're starting to see because of things like the Cybersecurity, the Resilience Act in Europe and some similar things in the United States that is kind of moving us, I think, I hope to a little bit more, I don't love the word sustainable, but it's, it's fairly accurate in this case, a more sustainable place for open source.

Jeff: Well, I mean, it, I, I, I would like to see it because, I mean, sustainable, I'm okay with it, I, I'm okay. You know, even, even if I had to pay for the, the Linux kernel or Linux operating system, you know, if it was a reasonable amount say, and it's going to vary based on, you know, your economic place in life, you know, but if it was pretty cheap.

You know, based on what you, whatever you make, you go, you know, it's pretty cheap. I'd be okay. Cause I, I at least understand there's people that are putting a lot of time in this and they've got to eat too. They've got to, you know, as long as we stay away from some of the outrageous licensing, you know, if they just go, Hey.

How about, you know, and, and you kind of have that with the donations you can do through various open source organizations, but, you know, and I try to give a little money to here and there to various, you know, I've donated to KDE, I've donated to Code Weavers. Well, I've kind of, to wine, I've, I've purchased a license through Code Weavers who was in the major backers of wine.

I've, you know, things like that where I. I try to support open source and, you know, give, give a few, few dollars, you know, I'm not breaking the bank, but you know, I'll, I'll throw them, I'll throw them a few bucks, you know, just to help out.

Jonathan: There is one project that immediately comes to mind that has made that work and that is Ardour, the the, the audio editor, multi track audio recording and editing suite.

And they've got a system where you can become a subscriber and they offer they offer different tiers you can either do you can chip in one dollar a month and I believe they they label that tier as the sort of the You know the low income from a from a country where this is all you can do I forget exactly how they put it, but they have that tier They have a four dollar tier, which is the one that i'm on a ten dollar tier or a fifty dollar a month tier and when you pay, that unlocks the ability to download the binaries that they compile for you, as opposed to, you know, if you're running on Linux, it's reasonably easy because most most of the links just shows also package it.

But particularly for people running on Windows, I think Mac as well. It's sort of a challenge to get our door installed if you're not a subscriber, and they make that work. They pay a couple of developers on that. And it's, it's really interesting to see that they have managed to turn this open source project into, into a business and able to pay a couple of people to work on it.

Jeff: Well, and I think even on Linux, the, the packaged versions are older versions. I thought, I thought you were always behind.

Jonathan: Well, they're going to be a little bit behind just because it takes time for Ardor to push. Ardor will push a release out, and then you have your packager will come along and go, Oh, hey, there's a new version of Ardor.

Go through the steps to package it, and then most distros, once it gets packaged, it gets pushed to some testing repository. And then it tests there for a couple of weeks. And if nobody finds any bugs, then it gets pushed to stable. So you do, you end up running, you know, probably two weeks behind the full release.

That's. Honestly, that's one of the reasons why I went ahead and paid for it. When there is a new release of Ardour, I can just go grab the installer script and run it and back up and up and running with the new version. I also, I, I had some feature requests and it felt kind of weird to go and ask for features and not as a paying subscriber.

So I figure I'd give them 4 a month. That gives me the right to complain about things that are broken.

Jeff: There you go. Yeah, I could see that. I, I guess I was mistaken a little bit. I thought, I thought part of their licensing or whatever was the, the free version was. Or maybe it's if you download it from their site or something.

I, it's, I, I want to say it was. You

Jonathan: can, there, so there is, there is a there's something like that. You can download. I'm trying to remember the details you may be able to download the old version but if you get the most up to date version, it'll do something like only let you render out 30 seconds at a time, or after every 5 minutes it does 30 seconds of silence, like it's got something like that built into it, I can't remember the details.

Oh, okay. But yeah, there is, there is something in there kind of a, I guess you would call it annoy aware, but yeah, that makes it kind of, kind of encourages people to pay the money.

Jeff: Yeah. And you know, and I'm okay with that too, because for me, I don't pay anything to them if I'm running an old version, you know, I use it so infrequently.

You know, it's like the, the newest feature bug fix probably doesn't even affect me cause I'm not a power user. I'm not, you know, Oh, I'm just recording 30 seconds of audio for something. And then I'm done for six months or whatever.

Jonathan: Yeah. I'm, I'm looking for their their, so they call it their demo version, the demo copy.

I guess it just has a timer. It, it it'll, it'll work for, and they don't, they don't say on this page how long, but you know, a certain number of minutes that it'll work. And then you just get a message, the timer's expired. And, I'm sure people don't like that sometimes, but they're, they're they have an FAQ on the Ardor site.

I thought this was free software. Ardor is free in the following ways. You're free to do anything that you want with it, including You know, make changes, put it on as many machines as you want. You can get the source code without charge and build it yourself. That's funny. I think it's worth it for, for

Jeff: Ardor.

Well, I was gonna say, and they have a single payment. Okay. So you can just say, okay. If you choose to pay less than 45, you'll get the current version and updates to version, all the point versions, but not eight. If you go over that, then you also get the next major version and a bunch more. So, okay.

Yeah, that's,

Jonathan: yeah, it works for them. All right. Well, we have hit the bottom of the hour again. We've managed to fill a full hour talking about deep seek and some other open source news things going on these days. Let's go ahead and see about. Wrapping this thing up and letting the folks go. We didn't ask Doc.

I should have asked Doc before he had to run. Because we are required to ask people about their favorite their favorite scripting language and text editor. I'm not sure what, I'm not sure what his would be these days. I doubt that he does a whole lot of scripting anymore. But what about, what about you, Jeff?

If you have to, if you have to smack together a script to make something work, what, what language and editor do you use?

Jeff: Language would be Python. Mm hmm. So I That's the main language I'm used to be fluent in, I guess. I don't do a lot of programming and scripting these days. But if I do, Python is my main tool in the toolbox.

Editor? I use Kate. I'm not a, you know, unless you need command line, you know, then it would be Nano. I'm not a VI or EMAX person. I never learned it.

Jonathan: I tried to use Emacs once. I'm like, Oh, this will be cool. I I'll, I'll get Emacs installed. For whatever reason, when I went to install Emacs, it gave me the GUI version of it and it was just terrible.

And I've never gone back and really tried to work with it ever since then. I can, I can get by in VI, but man, I, I prefer Nano myself.

Jeff: Yeah, I, I, I can like change something and then save it and that's about it. I don't have all the keyboard shortcuts and, but, but you know, I've never really had to learn because I've either been on the system directly or I redirect the display.

Or I'm running in a remote desktop, so I don't, I'm usually not limited to command line only. I'm a hardware person, so I, you know, I, I help build semiconductors. I don't do a lot of coding, I'm not running big banks of servers.

Jonathan: Makes sense, makes sense. Alright man, is there anything you want to plug before we let the folks

Jeff: go?

If you enjoyed us talking, you can catch more of us over on the Untitled Linux Show on the Twit. tv network and thank you for having me as a guest. I always enjoy being here. A lot of times I can't make it just because of scheduling, but. Always a pleasure when I can, and thank you for listening.

Jonathan: I don't suppose you have any poetry for us?

Jeff does the poetry corner at the end of the Untitled Linux show, and it's always a lot of fun.

Jeff: I don't, I wasn't prepared. Ha ha ha! Nobody's ever asked me on this show for poetry, so.

Jonathan: I know! If I remember next time I have you on, I'll make sure and bug you to see if I can get you to have something, because it's fun.

Jeff: Technological poetry. Yes,

Jonathan: yes. They're all, they're all fun. Our, our audience would appreciate them. I'm sure. All right. Thank you for being here, man. I appreciate it very much. All right. If you want to follow me, of course, there is Hackaday. We appreciate Hackaday being the home of Floss Weekly these days.

My security column goes there live on. And yeah, we've got the Untitled Linux show over on Twit that we've talked about. I am going to be taking a break from the security column for one week. It looks like that is in February. Because I am going to be traveling. So February the 21st, we will probably not have a column.

We are going to try to go ahead and have a Floss Weekly that week. And we're going to get a little bit ahead, I believe, on doing the recordings. But anyway, yeah, and if anyone happens to be at the Zero Trust World 2025, make sure to stop by and say hi. You can find me there. But yeah, we appreciate everybody that watches and listens, those that get us both live and on the download, and we will see you next week on Floss Weekly.