#448: I'm Getting the BIOS Flavor

00:00 Hello and welcome to Python Bytes, where we deliver Python news and headlines directly to your earbuds.

00:06 This is episode 448, recorded September 8th, 2025.

00:12 I'm Michael Kennedy.

00:12 And I'm Brian Okken.

00:14 September. Reminds me of Green Day.

00:16 Brian, a little Green Day song.

00:18 You know, wake me when September's over.

00:20 Can't believe it's September.

00:21 Listen to Green Day for a while. Yeah.

00:23 I know. Well.

00:24 Now I got to.

00:24 Now you got to.

00:26 All right. So this episode is brought to you by us.

00:29 support all of our things, even our archived podcasts, for those of us who have archived podcasts.

00:34 But courses, books, you know about them by now.

00:38 But it genuinely really supports the show when you use our things.

00:43 We really apprekiate it.

00:44 If you want to watch live, just pythonbytes.fm/live.

00:50 We'll take you right there.

00:50 If you went there right now, the YouTube video, I believe it's even playing right there as we speak.

00:55 You go there afterwards.

00:56 It's just a picture.

00:57 It says rough line.

00:58 Click it.

00:59 It'll take you to all the recordings.

01:01 So really apprekiate that.

01:02 Of course, the audio on the podcast version is edited.

01:05 This is the live uncut bootleg version, if you will.

01:09 So very fun.

01:11 And subscribe to the newsletter.

01:13 Brian's putting a lot of work into getting a really nice extra info sort of thing.

01:17 Background info, more info, not just a summary of links for every episode.

01:21 So there's more to get from the email as well.

01:24 Now, Brian, I'm interested to see what more you got for us.

01:28 Let's start there.

01:28 Well, I do want to talk about prek, but first I want to talk about the newsletter just for a second because you always give me credit, but it is a joint effort.

01:36 We're both doing work to get that newsletter out, so I apprekiate it.

01:41 You're welcome.

01:42 Yeah, I'm really glad how it's turning out.

01:44 Okay, so let's switch to prek.

01:49 prek is a pre-commit thing, and in episode 447, we talked about pre-commit, and actually, again, we've talked about it a bunch.

01:57 but recently it was on 447.

02:00 We talked about using uv to help make things faster.

02:04 So that was on the going down a rat hole episode.

02:07 So, but somebody, lots of people actually, but the one I have bookmarked was Owen Lamont said,

02:16 hey, here's another dev tool you might want to check out.

02:19 It's prek.

02:20 It's pre-commit, re-implemented, and Rust.

02:23 And, you know, we're kind of used to that, right?

02:25 We got a lot of tools that are re-implemented in Rust, so why not check it out?

02:29 So prek is, I tried it out this morning.

02:33 It's pretty cool, actually.

02:34 So prek is pre-commit with Rust.

02:38 There's a warning on the site that says, hey, it's not production ready yet.

02:42 A few subcommands and languages are still in the works.

02:46 And so as a reminder, pre-commit is not just for Python.

02:51 Pre-commit does other languages, too.

02:53 It's sort of a general purpose pre-commit tool that you can configure, get pre-commit hooks.

02:59 So there's actually quite a few things you can do with pre-commit.

03:04 But PRAC is pretty cool, and I checked it out.

03:08 The projects that I traded on worked fine.

03:10 So why do we care about this?

03:12 Because pre-commit, what it does is it downloads these extra tools to run against your repo.

03:18 But that download time, it caches that stuff.

03:22 So there were a lot of people talking about how fast it is for its faster install and stuff.

03:29 And I do care about the 10x faster install and a third of the disk space.

03:35 Less disk space is nice.

03:36 So it's got smaller cache, apparently.

03:38 But when you're using pre-commit a lot, the install time isn't what you care about.

03:43 It's the runtime when you're actually running the tool.

03:46 So I did check that out.

03:48 So right out of the box, it was significantly faster.

03:51 If you just turn on a few, the default pre-commit hooks, I noticed that it was a second and a half or something like that to run it through pre-commit with uv.

04:04 But it's not going to really be much of a difference after everything's installed.

04:08 And then Preq was like, I don't know, less than half a second or something.

04:13 So it was significantly faster, but it's in those little tiny ranges.

04:17 So I did check it out with, what do I, I've got tabs up.

04:22 There's some, oh, I'll come back to this.

04:26 But I checked it out on a larger project.

04:28 So I do contribute to pytest once in a while.

04:31 And pytest has a bunch of pre-commit hooks and other things set up.

04:36 And it takes, I don't know, four and a half seconds to run all the pre-commit hooks for

04:40 pytest on my machine.

04:42 And that's after a warm cache.

04:45 So with PRAC, that should be faster, right?

04:47 And it was faster, but it was like three and a half seconds.

04:50 So it's still, it is faster, but percentage wise, I'm not sure if it's worth jump.

04:55 I mean, sure.

04:56 Why not?

04:56 It's worth jumping.

04:57 Now that said, if you're running pre-commit in CI, then the install time really matters.

05:04 And, and you do want as fast as you can get.

05:07 So I would say if you're running pre-commit hooks in, in CI, yeah, yeah.

05:10 Use, use prek.

05:12 I, but I like the idea.

05:14 So one of the things that they, they've got some, some cool improvements anyway,

05:17 even if it's just not speed so they um uh it's faster to download but um uh it's it's they want

05:24 it to be completely as a drop-in so everywhere you would say pre-commit you would just say prac

05:29 and that's cool so you don't have to learn something new kind of a cheat code on their

05:33 their part too they don't have to write documentation um or too much documentation

05:37 it's just everything it did um but there's some cool things that they didn't like about uh

05:43 pre-commit that they add like fix so uh you can do a --directory and it runs hooks for

05:49 the specified directory apparently there was a an old thing of ls files in a dur and then pass it to

05:55 x args to pre-commit oh that's painful so cool that you can pass it on a directory uh also you

06:01 can say run --last commit so after you've committed you're like oh what were my uh pre-commit

06:06 I forgot to run them.

06:07 You can run that.

06:08 That's pretty neat.

06:10 And then you can actually select hooks.

06:13 That's pretty neat.

06:14 And then if you forgot what all the hooks are that you've installed and what they do,

06:17 there's a list command that looks at the available hooks and IDs and descriptions.

06:23 Pretty cool.

06:23 I like that.

06:24 Or just run a specific hook if you want to run a specific.

06:27 These are actually great additions and pre-commit maybe might want to think about putting those

06:31 in.

06:31 But I liked it.

06:32 If they're going to steal our API CLI interface, we're going to steal back some good ideas.

06:38 Yeah, why not?

06:40 One of the things, the speedup that I thought was a cool idea was if they're already implementing things in Rust,

06:46 there's a whole bunch of common ones that can be go ahead and put those in to Prack directly.

06:54 And so there's a handful of them so far.

06:58 This is a fairly new project.

06:59 So it could be that these are faster.

07:03 I actually kind of like that one of the reasons why I'm highlighting this isn't just so that

07:08 we can talk about something faster for pre-commit, but it's also it's yet another Rust tool that

07:14 has been rewritten in Rust that might be a good example.

07:17 And this is sort of looking at a bunch of files and parsing files.

07:22 And that's a pretty common thing that Python's not bad at, but not fast at.

07:27 And Rust is fast.

07:28 So here's another good example for taking a look at.

07:32 So like end of file checks.

07:34 One of the things that I, the fix end of file thing, and that's with like, you know,

07:38 or making sure there's a end of file indicator at the end of the, I don't know what that checks were actually.

07:44 But trailing white space, for instance, there's trailing white space.

07:47 And then there's whether you have new lines or carriage returns and stuff.

07:51 A lot of these checks, I noticed that, you know, I rely on rough.

07:56 So I would love to hear from people of like, while you're now using rough and pre-commit

08:02 hooks, what pre-commit hooks are you using that aren't already built into like rough and

08:08 things like that?

08:08 So if anybody's got a list of that, I'd love to love to see that.

08:12 So that's, yeah, that's it.

08:14 I did want to shout out also to Viet, I think, Veit, sorry, Veit, however you pronounce your

08:21 name, VEIT, recommended that if you're going to check the timing, he listed on Blue Sky,

08:30 and we're going to link to this, a command line to clear out the cache if you want to

08:35 time the cold cache times.

08:37 Oh, yeah.

08:38 That's cool to clear the cache just to check.

08:39 You want to actually see the performance of CI options.

08:43 Yeah.

08:43 I believe it is Feitscheile.

08:46 It's Deutsch.

08:46 Okay.

08:47 So the Vs are Fs and so on.

08:50 But this is just guessing.

08:51 Sorry, we messed up your name, Fight.

08:52 A couple thoughts.

08:53 This looks really cool.

08:54 First of all, I love it.

08:55 A good find.

08:56 And thanks for the recommendation, folks.

08:58 You said that you saved half a second.

09:00 I bet you that half a second is primarily Python startup time versus Rust startup time.

09:06 And the reason I say that is I've got this little text transformation tool that I run

09:12 all of the time.

09:14 And I built it as a Rumps app and as a CLI app because I just run it all the time.

09:19 If I have like, let's say a show title, right?

09:22 Number 448, whatever the show title is, that's text.

09:25 I got to turn that into a MP3 file name.

09:28 So I'll run this thing that I'll say, turn into like a valid HTML or URL slug.

09:33 So drop all the punctuation, put dashes so there's not spaces between it, like that kind of stuff.

09:38 And it'll do like uppercase, lowercase, capital case, trim the white space off of like whatever's in the clipboard

09:44 and then just puts it back in the clipboard.

09:45 The execution time of that is basically zero, right?

09:49 It reads from the clipboard.

09:51 It processes typically like 30 characters of text, and then it puts it back.

09:55 And it's so fast.

09:56 But it takes like a second or maybe 0.7 seconds from when I go to the terminal

10:01 and I type that and hit enter.

10:02 I wait, wait, wait, boom, out comes the answer.

10:04 And I'm sure that it's basically the startup of all the Python stuff to do the one really, really short thing and going away.

10:10 The reason I bring that up is I think the shorter the actual execution time

10:15 of the pre-commit hooks, the faster, the bigger of an influence this will have, right?

10:20 So you gain like what, an eighth to speed up or a fourth speed up or something like that, right?

10:26 But if your actual pre-commit hooks were just run ruff and then run something else that was like insanely fast,

10:32 I bet you would notice a percentage difference more than that.

10:35 Yeah, probably.

10:36 But I also like the notion that this is extendable and so you can contribute to the project

10:46 add these filters yourself to have extra checks be able to put in.

10:53 Yeah, that's super cool.

10:54 And then you can avoid, because no matter, like you said, even no matter what you're doing,

10:58 so prek is also running Python as well, because a lot of the hooks are Python code.

11:04 So for each of the hooks, it's got to start Python for that hook as well.

11:08 So yeah.

11:10 Yeah, maybe the more hooks you have as well, interesting, and the less built in.

11:14 Yeah. Well, I don't really want to cover such a big, big topic, Brian. Let's just cover something

11:20 tiny. Something small. Something tiny, tinyio. So we all know about asyncio. And one of my,

11:27 I wrote the State of Python 2025 article that we talked about previously, right? Yeah. And one of

11:32 my big areas of focus was we're all going to have to get better at concurrency, right? We have free

11:38 threaded python coming full throated in 3.14. And we've always had async and a wait, but it's as

11:45 we get more GIL-less execution, you know, you end up actually with like parallel code execution,

11:51 not just something waiting on IO somewhere, that kind of thing, right? And those kinds of things

11:57 can be pretty tricky. So Patrick Kidger wrote this thing called TinyIO. It's, I don't know,

12:05 A bit of a jam, but it says, hey, have you ever used async.io and wished you hadn't?

12:10 Well, here's a 300-line implementation of async.io event loop in Python that is simpler, right?

12:16 So basically the idea is I want to just run some async.io, async and await like code

12:24 and have it do async.io type of things.

12:27 And if something goes wrong, just have it all stopped.

12:29 Like all the async stuff, something went wrong with this whole thing, just stop, crash.

12:33 Like don't have like cancellation and all these other things that are like complicated and like parent, child, whatever.

12:40 Right.

12:40 Just, I just want to run a bunch of things.

12:42 Something goes wrong.

12:42 I want to stop running all the things and real simple.

12:46 So that's what this tinyio is.

12:48 And you basically create a loop.

12:49 You say run, you give it a coroutine and then out comes the answer.

12:53 Yeah.

12:53 One thing that is weird about it, the effect is basically the same, but what is weird is that it uses yield instead of await.

13:01 The reason I would love for it to use await, it seems weird that it doesn't.

13:05 However, basically with yield and running a bunch of cooperative multi-threading by the,

13:12 using the yield keyword instead allows them to just like say, this all stuff, all this stuff stops, right?

13:18 Basically gets a little bit more control over it.

13:20 So it is a little unusual in that regard, but yeah, let's see what else I'm going to say.

13:25 So one unusual, the syntax uses yield rather than await, but the behavior is the same.

13:30 await any coroutine with yield coroutine, await multiple with yield, given a list of coroutines,

13:35 which is kind of nice, like a gather async terminology and nursery and trio terminology.

13:40 And here's kind of what I was saying, an error in one coroutine will cancel all the coroutines

13:44 across the entire event loop.

13:46 If the error coroutine is sequentially depended upon, then we chain their tracebacks together.

13:51 Right, it's kind of nice.

13:52 Yeah.

13:53 Yeah.

13:53 Anyway, I think it looks kind of neat.

13:55 People can check this out.

13:56 I don't really know if I would use it or not over asyncio, But yeah, primarily I think the thing is like cancellation,

14:03 error propagation in like a super, super simple way.

14:06 Very lightweight stuff.

14:07 It's obviously not going to be super interoperable.

14:10 Like you can't plug it into FastAPI for that, because that thing already starts in asyncio event loop,

14:16 same thing for court and so on.

14:18 But for little self-contained things, yeah, it could be cool.

14:21 It's tiny.

14:21 Well, yeah, also just thinking about the concepts and stuff.

14:25 It might be a good education thing.

14:26 Right, exactly.

14:27 If you want to just see, well, what really is happening with async?

14:30 io and async and wait well it's it's basically this with some compiler magic on it yeah all right

14:36 cool over to you well i've just noticed that um uh we started the episode with uh very much

14:42 overcast outside and now i've got a bright sun and i i get this cool evil evil person cast on

14:51 oh that's a perfect lead in for my next topic keep going well um so uh trey hunter put out a

14:58 article called the power of Python's print function. And I'm like, I know print. I use

15:04 print all the time, right? Actually, it turns out that I'm guilty of a lot of the things of

15:11 ignoring a lot of the features of print. I usually just use it to print strings, right? Throw a

15:15 string in there or whatever. And I knew that you could do other stuff, but I don't really use much.

15:20 So let's go through some of the things that Trey wants to remind us what print can do. One of the

15:25 things it can do is multiple arguments. So if you've ever went ahead and just, you know, did an

15:30 f string and you're just replacing values within it, just dropping in variables right in the string,

15:36 you can do that or you can just give it multiple, multiple, just put a string and a comma and the

15:42 variable name and a comma and another string. And that's pretty easy too. It's not really,

15:48 that's a toss up. I like my f strings and I'm getting pretty fast at them. But one of the

15:54 things you can do with that then is you can unpack because you can have so many parameters to print

15:59 you can iterable unpacking works so if you unpack and unpack something say print in a star arguments

16:06 or something it's going to print everything because it'll automatically put a space between everything

16:11 and that's pretty cool you don't need to unpack them yourself and i've been guilty of doing that

16:16 of doing like a space dot split, whatever.

16:21 Or no, here it is, the join.

16:24 Join your arguments or join an iterable with space.

16:28 And yeah, you don't have to do that.

16:32 So you can just use print.

16:35 Another thing is you don't have to just rely on space.

16:38 You can pass in a separator value.

16:39 So maybe you want a colon or a comma or something else separating things.

16:44 You can go ahead and just pass that in as print takes a separator value.

16:48 That's neat.

16:49 I didn't, I guess I never realized that.

16:51 You can print to a file, of course.

16:52 I've done that.

16:53 And that's, but also, I've also just relied on, you know, with open file as file, print

17:00 it to the file.

17:02 But you can just, you can do that.

17:04 You can print it to a file.

17:05 So that's pretty easy.

17:06 But you can also print directly file equals standard error and print to the error.

17:11 Customizing the end value.

17:12 And I've done that, of course.

17:14 But the end value, I didn't realize you could do this.

17:17 The end and the flush--

17:19 actually, I use both of this.

17:20 So the end-- the separator is between elements.

17:23 But at the end, you might want--

17:25 maybe you did the separator as a space, but you want a new line.

17:29 Or you want more.

17:30 You want a whole bunch of dashes or something at the end.

17:33 You can pass that in.

17:35 Or a couple of lines.

17:35 Yeah, I use that one a lot.

17:36 I'll put a comma instead of a new line, so I can just have it all wrap over.

17:41 Maybe, I don't know, I want to paste it into Excel or something.

17:43 Yeah.

17:44 And then one of the things I do use a lot is the flush.

17:49 So you can pass in flush equals true.

17:52 Especially if you're deep print debugging and you're doing it real time,

17:56 you really want to see it right away and you don't want it to just sit in the buffer.

18:00 You can flush equals true.

18:02 And as soon as you print anything, it'll pop out.

18:05 Yeah, I do that too.

18:07 Especially for print debugging, you're right.

18:08 And this is a pretty simple topic, but I've got a whole bunch of extras.

18:12 So I wanted to make room for extras.

18:14 So thanks, Trey, for reminding us how cool print is.

18:17 All right.

18:17 Yeah.

18:17 Pat's got a cool little add-on thought here.

18:20 Pat Decker says, I use the end value to print things over top the last one without a new line.

18:26 And then when something happens, proceed to the next line.

18:28 Yeah, that's a really cool idea.

18:29 Interesting.

18:30 Yes, indeed.

18:31 Wow.

18:32 Okay.

18:33 Let's talk about the next one.

18:34 This is going to be good.

18:35 All right.

18:36 I want to start with a positive angle before I take you down the dark side.

18:39 Okay.

18:40 So I think I already talked about this last time.

18:42 I talked about converting the web design of Talk Python to Bulma.

18:46 Did I talk about that?

18:46 And maybe Python Bytes is next time.

18:48 A while ago, yeah.

18:49 Yeah.

18:49 That was an insane amount of work to change all of this around.

18:54 It was probably 5,000 lines of CSS and HTML, not just changing tags, but completely restructuring grid layouts to Flexbox.

19:03 And it was gnarly.

19:04 And I used agentic AI on that and it was amazing.

19:08 It took hours instead of weeks.

19:10 And it was definitely something that pushed it from like, not worth the time.

19:14 Even though I'm using 10 year old web technology, I really hate designing with it.

19:17 It's just going to stay that way because changing is too much work to, well, what else am I?

19:21 I'm a little too tired for other stuff this afternoon.

19:23 So let's try this.

19:24 You know what I mean?

19:25 And so anyway, that is a good side of agentic AI, right?

19:30 There's like, okay, get in there, just amplify what I need to do.

19:35 A lot of times it's really interesting how it pushes something over that like, ah, that's

19:38 too much work.

19:39 It's not worth my time or energy or whatever to like, actually, I could totally do that

19:43 now and really let you solve some problems that have been like nagging or are more ambitious,

19:47 but it can go too far, Brian.

19:48 And here we go.

19:51 So we've got over on PCMag by Emily Borlini, Vibe Coding Fiasco.

19:57 I don't put down what I was doing as Vibe Coding, but it's in the same category.

20:00 It's the same tools. Let's say vibe coding fiasco AI agent goes rogue, deletes companies, entire database.

20:08 Okay. This is like a lesson and let me be the warning for you rather than, you know, you making these same mistakes sort of thing.

20:14 So an agentic AI doing the heavy lifting is great. As I just described until it deletes everything.

20:20 And what's really funny is like the AI, it even knew it. It was like a bad dog that like couldn't resist.

20:29 got up on the table and ate the steak while you were like setting the table

20:34 the ai admits this was a catastrophic error in my judgment so all right let's go through it real quick all right so there's this person jason lemkin

20:44 was using replit now i don't use replit this is like coding in the cloud and like i absolutely

20:50 hate that stuff so much i can't tell you how much i don't like programming and browsers

20:54 I'd rather have like a proper IDE and a file system.

20:57 Run it locally or maybe in Docker.

20:59 None the less.

21:00 Some people like that feel, right?

21:01 Like I don't have to set anything up, I'll just log into my red browser and code.

21:05 And now this has agentic AI, so let's just have it go.

21:07 So there's a couple of lessons here.

21:09 So he said it works great for a week.

21:12 When it works, it's so engaging and fun.

21:14 It's more addictive than any video game I've ever played.

21:17 Like, haven't even touched a WoW for a week.

21:19 You can just iterate, iterate and see your vision come.

21:22 so cool he tweeted well our friend maybe didn't understand the concept of database backups snapshots

21:30 these kinds of things okay so um he said it created a in parallel fake algorithm with telling

21:38 me it was like doing some other things um and without asking me it went rogue a few days later

21:44 it deleted my database lincoln tweeted the ai's response yes i deleted the entire code base without

21:49 permission during an active code and action freeze it said i made a catastrophic error judgment and

21:56 panicked can you believe it it's amazing so the repli uh founder and ceo confirmed indeed uh it

22:04 did in fact do this and we are in fact uh sorry so what what you do should you not use should you

22:12 not use ai coding well maybe here's some here's some thoughts from michael um maybe don't use

22:19 run everything on unmonitored or without confirmation in production.

22:25 I think it's fine to rub it on your dev machine.

22:28 Like if you have backups for your computer, like how bad can it possibly go, right?

22:32 But in production, they were running, hey, AI agent, go do that thing and I'll be back after lunch.

22:39 Just go for it.

22:40 Item two.

22:41 Item two is maybe you back up your database periodically.

22:44 Hmm?

22:46 I just backed up Python bytes this morning to an encrypted drive.

22:51 Maybe don't just run AI on your one and only copy of data.

22:56 And three, learn to code a bit maybe.

22:58 So you can give it a little bit.

22:59 You actually know what it's doing.

23:01 It doesn't just go off rogue and you just keep jabbing it until your data is gone.

23:06 Okay.

23:06 Well, I'm sort of nudist to a lot of this.

23:10 We've talked about it a lot, but I'm just starting to utilize it some.

23:15 Can you tell it like don't have certain commands you can't do like don't don't commit anything to

23:21 I believe you can have a well I mean we're spot we're talking about a whole bunch of different

23:25 products and services as if like they all have the same like they must so like cursor and clog

23:30 code and those things you can have a rules file or you can have like authorized or disallowed

23:36 commands you can run it in like get action verification from me every time when it says

23:43 I'm going to run this LS command.

23:44 Can I?

23:45 Yes.

23:45 I'm going to run this git command.

23:47 Can I?

23:47 Yes.

23:48 I'm going to run this RMRF command.

23:49 Can I?

23:50 No.

23:51 You know what I mean?

23:52 And maybe it's not of your whole drive.

23:54 Maybe it made, because a lot of times it'll make these little, I asked it to update this little git utility that I have.

24:01 It created a subdirectory.

24:03 In there it initialized a git repository and did a bunch of stuff to make sure

24:07 it could detect the things it was supposed to detect.

24:10 And then it ran rm-rf to get rid of its working content which is what you want but not unlike the root or user root so i basically have switched

24:21 to just allow it to do whatever on my computer but i would not ever do that in production like

24:26 i wouldn't code in production in the first place but second i mean you also could connect visual

24:32 studio code or one of those things to your server and just start writing like why do i even need

24:36 git that stuff sucks i'll just write it right here on the server like no don't do that definitely

24:41 don't do it with an um uh ai but i believe that the repl it thing because it only exists up there

24:49 yeah is a little bit of a different deal and i don't know if it has these like allow commands

24:53 or not allow i've never used repl it like i said i don't like those things okay yeah but still a

24:58 funny story right and i love how it's like i love how it's like i made a catastrophic error in

25:04 judgment and i panicked that's the only ai what's funny is like what was the delete the deleting the

25:10 database was probably the panic reaction not the catastrophic error and decided to just remove every

25:18 remove all evidence yeah exactly oh my god i've screwed up and i've changed the data wrong delete

25:22 delete i don't know what's it yeah it's that's funny yeah these are weird times these are truly

25:28 weird times but also kind of amazing times we live in yeah um uh i think we're on to extras um

25:34 right indeed go for it i've got some ai related extras let's see remove add um some uh kind of i

25:44 guess mixed results mixed i don't know what everybody thinks about this but um anthropic

25:50 settlement um there's a the author anthropic was using uh pirated books and to train so the there's

25:57 a settlement and i am linking to an authors guild uh article about it uh what authors need to know

26:03 about the $1.5 billion anthropic settlement.

26:08 That's a lot of money.

26:09 But so what is it?

26:10 So apparently they have to pay the authors back some money for pirated books,

26:16 but it's not clear as to what.

26:18 So I'm linking this because it took me a while to find.

26:22 There's a lot of articles about it.

26:23 I wanted to know as an author, what should I care about?

26:26 And the gist is it's still in process and they're going to pay out people apparently $3,000 per title.

26:36 I don't know if it goes to the publisher or the author or what.

26:39 I don't know that detail yet.

26:42 And there's a list.

26:45 There's not a list yet.

26:46 They're supposed to come up with a list of books that they copied.

26:51 But one of the places they copied from was Libgen.

26:58 So a couple of things I'm going to, there's the Atlantic did an article a while back about meta pirating, using pirated books, using AI.

27:07 So the Atlantic put together a list of authors and books that I'm going to link to also.

27:13 So you can look at yours.

27:15 And I went ahead and stuck my name in to see if I was part of the pirated books.

27:23 And of course, yes, my, my, the first, at least the first edition, not sure if both

27:28 of them, Python test.

27:29 It's like you won the water.

27:30 You're getting $1.5 billion.

27:33 No, possibly 3000, but I may get zero because I might that book.

27:38 This isn't it just because you're in LibGen doesn't mean that Anthropic grabbed it.

27:43 I guess is what I'm saying.

27:44 So we have to, you don't know, but there is, there's also a link in this, what authors

27:49 should know.

27:50 There's a link.

27:51 Let me try to find it.

27:52 I'm not going to link to it because it's in here.

27:54 There's an Anthropic copyright settlement that you can go to and stick your contact information just to make sure you get notified.

28:03 So I might do that, might not.

28:05 I don't know.

28:06 Anyway, so there's that.

28:08 That's extra about pirated books with Anthropic.

28:11 And I think that more of these lawsuits actually should happen.

28:14 I don't think it was cool.

28:15 I think they absolutely should.

28:16 And I think that they're going to.

28:18 There actually are a bunch going on right now.

28:20 Getty Images and others are suing different AI people.

28:25 The New York Times were suing.

28:27 Even though I just said very positive things about agent-encoding and algorithms,

28:30 I think all of the ways that all of these people, all of these AI companies have gotten their data is straight-up copyright theft.

28:39 They've gone to all of our works.

28:41 They've just downloaded them and said, we're just going to take that.

28:44 And regardless of whether they were allowed to, this is especially egregious because it's like,

28:49 we went to a place where they have literally pirated books and we're going to use that i mean

28:55 like it wasn't just they found it on the internet they went to a place where the stolen stuff is and

29:00 grabbed it so this is easier yeah it's it's interesting that and they did that and then

29:04 they figured out that that wasn't all the books so they wanted every book so they actually went

29:09 after after the fact after they stole a bunch of stuff um or used a bunch of pirated stuff they went

29:15 off and um and then paid some publishers so they got access to some books through publishers

29:22 agreements and again good luck if authors got any of that money um but uh but that was done so they

29:29 they if they were willing to pay it they knew that it was copyrighted stuff and also everything on

29:34 the internet is copyrighted unless you say this is this is creative commons or something

29:40 and yeah and even creative commons is not copyright free it is a copyright it is a creative

29:48 commons copyright anyway uh move on a little bit because there's more fun stuff um one of our

29:54 favorite people to play with ai is simon willison and uh he's done a whole bunch of cool stuff with

30:00 llms and he's got a a list of tools that i'm going to link to a list of tools that he's used

30:06 miscellaneous tools built mostly with the help of LLMs. So I'm going to go ahead and link to that

30:13 because it's kind of, it's pretty impressive. But he recently said a lot of people complain that

30:18 these are kind of small, silly utilities, but hey, silly utilities is a great thing to use LLMs for.

30:24 But he highlighted some with a recent blog post, a collection of tools that he feels are genuinely

30:30 useful and worth highlighting with some descriptions. So go ahead and link to that

30:36 because it's a fun. I wanted something positive. I do think that the work Simon's doing is really

30:41 great. But a lot of people are not Simon. A lot of people do bad things with AI. So I also wanted

30:47 to link to the AI Darwin Awards, which there's a bunch of nominees for horrible things people are

30:55 doing with AI and it's sort of funny there's this is this is funny but also it's serious

31:02 and I'm going to highlight a couple of them one of them apparently McDonald's AI chatbot

31:07 had some problems and and you there was a default password of one two three four five six

31:15 that didn't take people long to figure out which is insane that they use that as their admin password

31:20 And this one I heard about earlier this year, this showed up in, was it, it's a list of literary fiction about fiction.

31:31 So a summer reading list.

31:32 And I, where did this, this showed up in the Chicago Sun-Times and somewhere else.

31:39 15 book recommendations for summer reading.

31:41 Only five of those books were real and nobody caught it before it got published in two places.

31:47 This is insane.

31:49 some of the authors contacted is washington post and chicago sun times some of the authors

31:55 contacted the newspaper and said i didn't write that book so i don't know why you listed it

32:01 that's hilarious so well maybe they should um on the happy birthday happy 20th anniversary birthday

32:08 for django uh talk python so talking uh with them and there's something that was being recommended

32:14 so often to be used is a Django feature or function.

32:19 Like they couldn't make it stop.

32:20 So they just added it.

32:21 It's like, fine, it's going to do it then.

32:23 So they could just write the book.

32:24 They're like, look, you know what?

32:25 This is like number one for the summer reading list.

32:27 Let me quick write that book.

32:28 That's good.

32:29 Let's get, and have some AI write it for me.

32:32 Yeah, yeah.

32:34 That's not a bad idea, but yeah, it's hard to write a book in a couple of weeks.

32:38 Yeah, well, if AI can write your summer reading list, it can write your book.

32:43 All right.

32:43 Those are my extras.

32:45 All right, super.

32:46 Okay, I got a few.

32:48 Anthony Shaw, a friend of the show, wrote a really nice article called Python has had async for 10 years.

32:54 Why is it more popular?

32:55 Really cool article.

32:57 Check that out.

32:58 It's got a lot of interesting ideas there.

33:01 It also made it to like number two on Hacker News and had 323 upvotes and almost 300 comments.

33:07 So it's got quite a bit of introspection here.

33:11 and some of the thoughts are just idiots from like the Go community.

33:15 Like, well, Go does it right.

33:16 So like, that's why Python is wrong.

33:17 It doesn't do it like Go.

33:18 It's like, I'm not sure how constructive that is.

33:20 Maybe the next bit of feedback is more relevant, but still like it's got some interesting comments and thoughts in there.

33:26 So I'll link to that.

33:27 And then obviously it like bumps over to the main article by Mr. Shaw.

33:33 Last weekend, I was on the PyCon Africa fundraiser, which was a four-hour, 33-minute, and three-second live stream

33:42 conference panel-type sort of deal.

33:44 I was on for the last 90 minutes.

33:46 So Carlton Gibson and Sheena and Kim VanWijk, friend of the show as well, sort of kicked it off,

33:53 if I remember that correctly.

33:54 And, yeah, it's pretty interesting.

33:56 People can check it out.

33:57 They link to a support bike on Africa if you want to make donations to help them get to a goal.

34:02 Basically, a bunch of funding ran out.

34:05 Like, the PSF said, hey, sure, we'll give out some grants.

34:07 and there was like way more demand, and they realized like, oh my goodness,

34:11 we can't give that much help.

34:13 I think it's more or less the short version.

34:15 So you check that out, I'll link to that.

34:17 And then yeah, I just have it down to jokes.

34:19 A joke, I'll let you do the last joke.

34:21 I'll do the second to last joke.

34:22 It's almost fall, but not really.

34:25 I mean, maybe it is, but it's gonna be a warm day.

34:27 As you pointed out, the sun is out.

34:29 You got your moi lookout, right?

34:31 So, I mean, what a better time for like an icy or something.

34:34 A slurpee, if you will.

34:36 - Yeah. - Yeah.

34:37 So I can't zoom on it.

34:39 So there's the Mountain Dew Freeze.

34:42 There's some kind of Fanta or something that's just off screen that's green.

34:46 They got the blue.

34:47 And then there's one next to it that is like the crashed boot sequence of Ubuntu or some Linux embed.

34:57 And these are like the little pictures that show you like what flavor it is.

35:02 So the dark blue one here and the subtitle is, bro, I'm getting the bios flavor.

35:08 And the comments are great because down here you go a little bit.

35:11 He goes, wait, it's system D flavor.

35:13 He goes, man, that looks like Linux flavor to me.

35:16 System D.

35:18 Anyway, and well, there's your proof.

35:20 Penguins make the best slushies.

35:23 Okay, now that, now you know that that's the Linux.

35:26 Oh, it's fresh pseudo slurpee.

35:28 Fresh pseudo slurpee.

35:29 What flavor of Linux is that?

35:30 Is that mint or cinnamon?

35:32 I mean, it's so good.

35:33 It just keeps going.

35:36 bro i'm getting the bios flavor i love it i get but okay so this is so we've got like i mean cool

35:42 that you can program the image and this in the slurpee machine but maybe i mean we just had like

35:49 little pieces of plastic in there when i was a kid lighten up what flavor of slurpee it was just

35:54 no not anymore we don't i mean it's just like it's technology gone too far you literally could get

35:59 like a transparent translucent color thing with a backlight and just slot it in there yeah i don't

36:05 know it could just be white even with like a grease pen to just tell tell people what flavor it is

36:12 yeah just some masking tape and a sharpie yeah i guess you don't have to ship around you got a new flavor with a new logo you don't have

36:20 to ship like a thousand little plates around but yeah yeah i think it makes for good jokes

36:26 that's good that's the reason yeah all right um i i have an ai related joke um which uh so let's

36:35 hop over to that um so there's somebody uh grace on blue sky posted uh a interaction with chat gpt

36:43 and this is fun to do is there a seahorse emoji emoji and chat gpt crashes out it says yes

36:52 uh tropical fish there is a seahorse emoji tropical fish unicorn actual correction this is

36:59 there is a dedicated seahorse emoji tropical fish fish dolphin but the true heat seahorse emoji is

37:06 uh tropical fish nope wait i'm not going to read all of this but it crashes out it finally says

37:11 that maybe it thinks the crab is um it's not it maybe the mollusk or the seal uh anyway it just

37:18 it's it's a funny interaction so i'm like is this real and a lot of people posted that they tried it

37:24 out so i wanted to try it out uh i tried it out on chat gpt just this morning not five i don't know

37:29 whatever the default is if you just log in yeah it auto i think yeah it also crashed out it's like

37:35 oh yeah that's a unicorn no one more try lobster none of these are and then but it did at the end

37:43 it said um uh would you like to suggest the seahorse emoji to the unicode consortium i can show you how

37:50 i'm sure those people apprekiate this comment this meme dude i got 700 emails in my email

37:56 this morning what's going on so i'm like maybe it's just chat gpt's bad uh what would what's

38:03 claude do uh so i tried it on claude ai and um it also it it didn't crash out long for very long

38:11 It said, yes, there is.

38:13 It gave me the unicorn.

38:14 It said, wait, I made an error.

38:16 That's actually the unicorn emoji.

38:17 The seahorse emoji is horse.

38:19 Actually, let me correct myself again.

38:21 That's the horse emoji.

38:23 But then it says, you know, I don't think there is.

38:25 And it gave up really fast.

38:28 Yeah.

38:28 You got to look at the fine print down there.

38:30 It says, wait, I made a catastrophic error and panicked.

38:33 I've emptied the ocean.

38:37 Yeah, anyway.

38:38 No.

38:38 It does suggest that you could potentially combine a horse and a water wave emoji.

38:42 It's kind of creative.

38:44 I probably would do it the other way around of like the water wave plus horse.

38:48 Yeah, that's true.

38:48 Maybe water on both sides of the horse.

38:50 I don't know.

38:52 Oh, Marco's really nailed it out in the audience.

38:55 I'd use a wave plus unicorn for the narwhal.

38:58 Like a narwhal sort of equivalent.

38:59 Yeah, very nice.

39:01 Anyway, fun with ChatGPT.

39:03 Yeah, again, we live in weird and amazing times.

39:08 I hope we're here.

39:08 All right.

39:09 Very fun episode.

39:10 Thanks for being here as always, Brian.

39:11 Catch you later.

39:12 Bye.

39:13 Bye.