WEBVTT 00:00:00.020 --> 00:00:04.620 Hello and welcome to Python Bytes, where we deliver Python news and headlines directly to your earbuds. 00:00:05.380 --> 00:00:07.600 This is episode 500, 00:00:08.240 --> 00:00:09.980 recorded October 13th, 00:00:10.260 --> 00:00:11.160 2026. 00:00:11.540 --> 00:00:12.560 I'm Michael Kennedy. 00:00:12.960 --> 00:00:13.860 And I'm Calvin Hendryx-Parker. 00:00:14.300 --> 00:00:15.120 I'm Nick Muoh. 00:00:15.480 --> 00:00:16.059 I'm Brett Cannon. 00:00:16.420 --> 00:00:19.800 I'm Carol Willing, and I sometimes pretend I'm Brett Cannon. 00:00:20.140 --> 00:00:21.660 And I have no problems being Carol either, 00:00:21.820 --> 00:00:22.420 so there you go. 00:00:22.660 --> 00:00:22.940 Everyone, 00:00:23.160 --> 00:00:25.000 thank you so much for being here for this special episode. 00:00:25.700 --> 00:00:26.580 Sign up to the newsletter, 00:00:26.880 --> 00:00:29.800 social links, all that kind of stuff is at the top of the show notes. 00:00:29.980 --> 00:00:31.740 So let's just go ahead and kick this off. 00:00:31.980 --> 00:00:34.560 We've got a really awesome, almost 10 years. 00:00:34.650 --> 00:00:36.200 I don't know if this is exactly 10 years or not, 00:00:36.380 --> 00:00:37.560 but almost 10 years. 00:00:37.740 --> 00:00:40.420 Certainly episode 500 is quite a thing. 00:00:40.760 --> 00:00:42.400 So Python Bytes has been around a long time. 00:00:42.680 --> 00:00:45.620 All of y'all have been here before in various ways. 00:00:46.180 --> 00:00:49.940 The one big missing component here, the person we're missing is Brian Okken. 00:00:50.420 --> 00:00:51.320 But don't worry, 00:00:51.700 --> 00:00:52.500 Brian was supposed to be here. 00:00:52.640 --> 00:00:53.120 Something came up, 00:00:53.320 --> 00:00:53.640 couldn't make it. 00:00:53.780 --> 00:00:57.900 So Calvin and I had a quick chat with him before and we'll splice that in. 00:00:57.980 --> 00:00:59.960 So we're going to be hearing that in just a little bit. 00:01:00.500 --> 00:01:00.680 Welcome, 00:01:00.920 --> 00:01:01.000 everyone. 00:01:01.360 --> 00:01:03.240 Thank you for taking the time to celebrate this milestone. 00:01:03.640 --> 00:01:03.880 Calvin, 00:01:04.420 --> 00:01:05.680 thanks for being here to keep it going. 00:01:06.040 --> 00:01:06.380 Love it. 00:01:06.480 --> 00:01:08.100 I'm glad to be the new co-host. 00:01:08.220 --> 00:01:10.580 And I'm excited for this crew for the 500 episode. 00:01:11.060 --> 00:01:14.880 And congratulations to all of you for getting us all the way to here. 00:01:15.260 --> 00:01:16.680 Thanks for calling us old, Brent. 00:01:16.780 --> 00:01:17.240 Appreciate it. 00:01:17.920 --> 00:01:18.820 No, I really do appreciate it. 00:01:18.820 --> 00:01:19.420 I'm just kidding, man. 00:01:19.420 --> 00:01:21.960 A person who's been doing Python was probably the, I think, 00:01:22.180 --> 00:01:25.160 at least officially the longest maybe on this call. 00:01:25.160 --> 00:01:26.440 I don't know, Caroline, I can duke it out. 00:01:27.100 --> 00:01:28.780 2000 when I started using Python. 00:01:29.240 --> 00:01:29.340 Oh, 00:01:29.540 --> 00:01:29.740 all right. 00:01:29.960 --> 00:01:30.420 Calvin wins. 00:01:30.860 --> 00:01:31.040 No, 00:01:31.260 --> 00:01:31.860 go Calvin. 00:01:32.460 --> 00:01:33.000 That's an accomplishment. 00:01:33.020 --> 00:01:33.180 Wait, 00:01:33.400 --> 00:01:33.480 no. 00:01:33.840 --> 00:01:34.500 Win in 2000. 00:01:34.600 --> 00:01:35.720 I started in 2000 as well. 00:01:36.040 --> 00:01:41.080 So it would have been early 2000, so probably like a February or March. 00:01:41.560 --> 00:01:41.980 You beat me. 00:01:42.140 --> 00:01:42.760 I started in fall. 00:01:43.180 --> 00:01:43.360 Okay. 00:01:44.660 --> 00:01:44.980 Wow. 00:01:45.120 --> 00:01:45.360 All right. 00:01:45.520 --> 00:01:46.500 Wow, that's crazy. 00:01:46.540 --> 00:01:48.080 Now we've established the pecking order. 00:01:48.260 --> 00:01:51.340 I'm sure I sit above Brett for most things. 00:01:52.280 --> 00:01:53.300 I'm not in the Python world. 00:01:53.760 --> 00:01:54.060 Amazing. 00:01:54.600 --> 00:01:57.280 Calvin, I was just thinking as I was getting ready for this show, 00:01:57.460 --> 00:01:59.160 we're on almost month 00:01:59.440 --> 00:01:59.620 four, 00:02:00.010 --> 00:02:02.640 if not past it, of you and I hosting the show together. 00:02:02.670 --> 00:02:03.580 It feels like we just started, 00:02:03.800 --> 00:02:04.720 but thanks for being here as well. 00:02:04.760 --> 00:02:05.220 It's been great. 00:02:05.370 --> 00:02:07.000 I really enjoyed jumping in. 00:02:07.160 --> 00:02:11.760 And this has been kind of a passion for me, a future dream that I thought would never 00:02:11.830 --> 00:02:12.140 come true. 00:02:12.210 --> 00:02:12.760 And here I am. 00:02:12.930 --> 00:02:13.320 I love it. 00:02:13.480 --> 00:02:13.800 Thank you, Michael. 00:02:14.260 --> 00:02:15.860 Yeah, it's awesome to be here with you as well. 00:02:16.420 --> 00:02:16.740 And now, 00:02:16.920 --> 00:02:17.520 like I said before, 00:02:17.800 --> 00:02:19.900 Brian wasn't able to join us in real time, 00:02:20.180 --> 00:02:21.400 but Calvin and I 00:02:21.340 --> 00:02:24.860 got the chance to speak with him beforehand to make sure he was part of the show. 00:02:25.200 --> 00:02:25.740 Let's have 00:02:25.880 --> 00:02:26.940 that conversation with Brian. 00:02:27.660 --> 00:02:28.260 Oh my goodness. 00:02:28.500 --> 00:02:30.920 It is Brian Okken. Thank you for joining us for the 00:02:31.020 --> 00:02:31.600 500 episode. 00:02:32.000 --> 00:02:32.960 Blast from the past. 00:02:33.280 --> 00:02:34.400 Brian Okken. Yeah, 00:02:34.620 --> 00:02:34.780 man. 00:02:35.060 --> 00:02:36.080 It's definitely a blast from the past. 00:02:36.080 --> 00:02:37.080 It's good to have you back in the show. 00:02:37.480 --> 00:02:37.780 500. 00:02:38.040 --> 00:02:39.060 That's a big number, 00:02:39.220 --> 00:02:39.320 man. 00:02:39.480 --> 00:02:40.740 And a lot's happened 00:02:40.900 --> 00:02:42.260 since you left. 00:02:42.480 --> 00:02:43.420 You published a book. 00:02:43.640 --> 00:02:44.360 I did. 00:02:44.580 --> 00:02:46.200 It's right there. 00:02:46.380 --> 00:02:46.520 Yes, 00:02:46.700 --> 00:02:47.120 it is. 00:02:47.460 --> 00:02:48.060 Lean TDD. 00:02:48.280 --> 00:02:48.959 We have the 00:02:48.980 --> 00:02:56.060 number one lean tdd author of for the pipe for the python bytes list and it may be the biggest 00:02:56.340 --> 00:03:00.960 new release of test driven development book in august of last of this year i don't know 00:03:02.820 --> 00:03:07.720 yes very very no no honestly it's it's a very good book and you and i we just got back from 00:03:08.080 --> 00:03:12.920 hanging out in san francisco together of all the things we ironically just randomly stayed at the 00:03:13.060 --> 00:03:18.940 same hotel for yeah it was it was actually kind of surreal we spent so michael and i um live in 00:03:18.960 --> 00:03:20.200 or near Portland. 00:03:20.600 --> 00:03:23.700 And we usually see each other at conferences somewhere else. 00:03:24.960 --> 00:03:25.700 You know what it is? 00:03:25.840 --> 00:03:26.560 It's so weird. 00:03:26.740 --> 00:03:27.940 I have so many friends like that. 00:03:28.100 --> 00:03:30.300 And it's because you don't have family. 00:03:30.510 --> 00:03:31.320 You don't have the kids. 00:03:31.390 --> 00:03:32.640 You don't have all this, you know, 00:03:32.890 --> 00:03:34.060 you could just kind of hang out and do whatever. 00:03:34.640 --> 00:03:35.360 Conferences are awesome. 00:03:35.530 --> 00:03:36.640 I think conferences are awesome. 00:03:36.940 --> 00:03:39.200 I mean, that's how I met most of you all 00:03:39.370 --> 00:03:41.280 is because of going to great Python conferences. 00:03:42.180 --> 00:03:42.360 Yeah. 00:03:42.660 --> 00:03:43.340 And Calvin, 00:03:43.760 --> 00:03:43.840 yeah, 00:03:44.140 --> 00:03:45.260 that Python party, 00:03:45.540 --> 00:03:45.680 man, 00:03:45.690 --> 00:03:47.400 it was the most epic party I've ever been to. 00:03:47.620 --> 00:03:48.520 So it was good. 00:03:48.760 --> 00:03:54.060 which one you're talking you're talking about select city yeah that was a really good one yeah 00:03:54.310 --> 00:03:58.260 the one that took over the top of the top of the bit the top of the top of the video 00:03:59.060 --> 00:04:05.860 no it was good plus i met um like on his name the brown bag guy um oh chris williams yeah yeah chris 00:04:05.940 --> 00:04:09.900 williams was there i mean we had charlie marsh there we had brett actually who's on the show 00:04:10.180 --> 00:04:15.460 was there so yeah he was really good yeah yeah and all and all those a lot of a lot of people 00:04:15.480 --> 00:04:22.180 had just started their um like startups that are going fantastic yeah yeah that was the rust wave 00:04:22.620 --> 00:04:28.300 we haven't really seen the ai wave party yet yeah it's the it no the ai wave is waving people 00:04:28.500 --> 00:04:34.400 goodbye when they get laid off because of it no well you know what seriously though we were just 00:04:34.540 --> 00:04:39.600 both in san francisco were you blown away at how many billboards were about ai that was weird things 00:04:39.660 --> 00:04:45.180 it was like going into a like a sci-fi futuristic movie or something it was well like because there 00:04:45.420 --> 00:04:50.500 wasn't anything else there was it took over the like i'm not sure i'm i mean i'm glad there weren't 00:04:50.560 --> 00:04:57.580 like cigarette billboards and alcohol billboards and and or like right you know the dispensary 00:04:57.780 --> 00:05:05.460 billboards divorce lawyers but having it all be like targeted towards stuff that uh just as i i 00:05:05.460 --> 00:05:11.260 still think it's a small percentage um uh ai is changing a lot but it's um it is sort of it's a 00:05:11.480 --> 00:05:16.960 kind of a middle upper middle class and higher sort of a phenomenon i don't i don't know if it's 00:05:17.120 --> 00:05:23.740 affecting everybody really other than it's taken over your search window anyway um yeah i i had a 00:05:23.960 --> 00:05:28.120 airport i had a ride to the airport i don't know maybe 25 minutes or something like that i mean 00:05:28.260 --> 00:05:31.560 in san francisco it could be three hours or it could be 10 minutes depending on the traffic but 00:05:31.580 --> 00:05:32.980 It was, I think, 25 minutes. 00:05:33.260 --> 00:05:35.080 And I think it was full of billboards. 00:05:35.190 --> 00:05:40.140 I think literally every single billboard I saw, at least that I noticed, was some kind of, if not AI, 00:05:40.360 --> 00:05:41.000 AI adjacent. 00:05:41.500 --> 00:05:44.940 Or there's all these companies that are like, yeah, we're the AI markdown editor. 00:05:45.120 --> 00:05:46.860 Like, you were just a markdown editor last year. 00:05:46.980 --> 00:05:47.420 Come on now. 00:05:47.630 --> 00:05:48.040 You know what I mean? 00:05:48.060 --> 00:05:49.260 Like that kind of thing as well. 00:05:49.400 --> 00:05:50.040 I don't get why. 00:05:50.140 --> 00:05:51.140 Who are they trying to convince? 00:05:52.200 --> 00:05:52.560 Investors. 00:05:52.960 --> 00:05:54.320 I think they're trying to convince investors. 00:05:54.480 --> 00:05:57.840 Anyone who's involved in this hype right now is raising money. 00:05:58.160 --> 00:05:58.380 Yeah. 00:05:58.590 --> 00:05:58.940 I don't know. 00:05:59.040 --> 00:06:02.800 How does your framework desktop in the basement feel about this, 00:06:02.960 --> 00:06:03.080 Colin? 00:06:03.540 --> 00:06:04.500 He's fine. He's right over here. 00:06:05.360 --> 00:06:07.040 He's on the standing desk over here. 00:06:07.570 --> 00:06:09.600 The local models are running just fine. 00:06:09.960 --> 00:06:10.520 A little hiccup. 00:06:10.910 --> 00:06:12.820 So I use it with Home Assistant in the house, 00:06:13.200 --> 00:06:16.360 which is another fine Python project that everyone should go check out. 00:06:16.760 --> 00:06:18.600 And you got a little confused this morning 00:06:18.940 --> 00:06:21.100 and thought that it couldn't control the home. 00:06:21.210 --> 00:06:23.540 Like I was using Qwen 3.8 27B, 00:06:23.750 --> 00:06:26.020 the one that's kind of all over the news today. 00:06:26.600 --> 00:06:29.560 And I just kind of forgot that it could actually control stuff. 00:06:29.940 --> 00:06:33.460 Maybe that's a good direction to go versus like take over all the things. 00:06:33.860 --> 00:06:34.900 But it does work. 00:06:35.080 --> 00:06:35.560 Works really well. 00:06:35.740 --> 00:06:35.980 Yeah. 00:06:36.220 --> 00:06:36.400 Awesome. 00:06:36.740 --> 00:06:36.820 Yeah. 00:06:36.820 --> 00:06:40.740 We got to get used to a non-deterministic world where like stuff just, it used to work yesterday. 00:06:40.940 --> 00:06:42.360 Now it doesn't necessarily work. 00:06:42.440 --> 00:06:43.880 Well, what's nice is you can have a discussion. 00:06:44.290 --> 00:06:46.240 Like I've got the home assistant voice set up. 00:06:46.450 --> 00:06:50.640 And so you can actually talk back and forth with the little box on the desk as opposed to kind of, 00:06:50.920 --> 00:06:56.240 I don't know, I won't utter the Amazon version of that box because it'll start going off. 00:06:56.460 --> 00:06:58.500 but you can't really have a discussion with those boxes, 00:06:58.580 --> 00:07:00.140 but you can have a discussion with Quinn 38. 00:07:00.480 --> 00:07:00.840 Oh, really? 00:07:01.220 --> 00:07:01.360 Yeah. 00:07:02.760 --> 00:07:03.500 You can go back and forth. 00:07:03.600 --> 00:07:04.380 I can interrupt it. 00:07:04.500 --> 00:07:04.880 I can, 00:07:05.120 --> 00:07:06.160 you know, it's all sitting on my desk, 00:07:06.360 --> 00:07:09.040 but just the satellite one dev board, 00:07:09.440 --> 00:07:11.080 which is getting a 1.1 release 00:07:11.780 --> 00:07:13.860 and their pre-orders are open now. 00:07:14.320 --> 00:07:15.660 So you're talking to it? 00:07:15.840 --> 00:07:16.460 I'm talking to it. 00:07:16.640 --> 00:07:17.280 Talking to my home assistant. 00:07:17.620 --> 00:07:17.820 Okay. 00:07:18.180 --> 00:07:18.820 Do you have arguments? 00:07:19.280 --> 00:07:19.600 All local. 00:07:20.880 --> 00:07:24.759 So it precedes like a prompt into the model 00:07:25.100 --> 00:07:29.120 and it precedes all of the devices that are in your home 00:07:29.520 --> 00:07:30.380 that home assistant knows 00:07:30.900 --> 00:07:33.200 with a preference for a specific room you're in. 00:07:33.320 --> 00:07:34.080 So if I'm in the study 00:07:34.460 --> 00:07:36.720 and the home assistant voice box is in the study, 00:07:36.780 --> 00:07:37.820 I've told it it's in the study, 00:07:38.240 --> 00:07:39.400 I can just say, turn off the lights 00:07:39.680 --> 00:07:40.660 and it knows it's in the study. 00:07:40.760 --> 00:07:41.420 There's lights in the study 00:07:41.460 --> 00:07:42.280 and it turns off the lights. 00:07:42.639 --> 00:07:42.960 Interesting. 00:07:43.780 --> 00:07:43.860 Yeah. 00:07:44.140 --> 00:07:44.640 That is clever. 00:07:44.760 --> 00:07:45.880 Here I'm back in the dark ages. 00:07:46.020 --> 00:07:47.900 I like go over to the wall and push a button. 00:07:48.120 --> 00:07:48.440 A wall? 00:07:49.660 --> 00:07:50.720 You could hurt your finger doing that. 00:07:50.820 --> 00:07:51.120 Don't do that. 00:07:52.080 --> 00:07:52.560 Touch that. 00:07:52.760 --> 00:07:54.380 They're conveniently located though. 00:07:54.600 --> 00:07:55.760 I mean, it's designed well. 00:07:55.920 --> 00:07:59.880 So the place where I enter the room is where the switch is. 00:08:00.900 --> 00:08:02.480 It's got incredible technology. 00:08:03.199 --> 00:08:04.160 It's kind of nice. 00:08:05.379 --> 00:08:06.800 Yeah, I just added it into the air. 00:08:07.060 --> 00:08:09.920 What happens when the Wi-Fi goes down? 00:08:10.999 --> 00:08:12.580 Well, that's on battery backup. 00:08:13.120 --> 00:08:15.020 And the Internet's not required. 00:08:15.760 --> 00:08:17.020 All the access points are PoE. 00:08:17.440 --> 00:08:18.660 There's a big battery in the basement, 00:08:19.320 --> 00:08:20.240 hooked up to the home assistant. 00:08:20.720 --> 00:08:21.440 It's all central. 00:08:21.700 --> 00:08:22.860 There's a battery up here on the framework. 00:08:23.460 --> 00:08:26.620 So I can last a couple hours without any Wi-Fi, 00:08:27.160 --> 00:08:27.260 internet, 00:08:27.500 --> 00:08:29.220 or power and still talk to my house. 00:08:29.520 --> 00:08:29.660 Okay, 00:08:30.120 --> 00:08:30.320 nice. 00:08:31.040 --> 00:08:33.740 I talk to my house too, but it just doesn't talk back. 00:08:34.340 --> 00:08:35.159 Mine can talk back. 00:08:35.880 --> 00:08:38.860 Hey, I wanted to say just thank you for stepping in. 00:08:40.759 --> 00:08:44.680 So I sort of just jumped out of Python Bytes without letting people know. 00:08:45.060 --> 00:08:46.460 And there was no hard feelings or anything. 00:08:46.660 --> 00:08:48.340 Michael and I are definitely still friends. 00:08:48.660 --> 00:08:49.560 He didn't kick me out. 00:08:50.580 --> 00:08:54.820 But it was this book and also my day job is pretty exciting. 00:08:55.140 --> 00:08:57.860 So I just sort of wanted to do other stuff. 00:08:58.520 --> 00:09:03.080 I really enjoyed Python Bytes and I'm glad it's not stopped just because I've stopped. 00:09:03.200 --> 00:09:05.480 I really enjoyed doing it. 00:09:05.580 --> 00:09:06.820 You had a great run. 00:09:06.900 --> 00:09:08.020 I mean, how many episodes, 00:09:08.480 --> 00:09:11.360 10 years nearly of Brian's voice in my earbuds. 00:09:12.060 --> 00:09:15.420 And now my Python Bytes, I'm like, oh, that's my voice. 00:09:15.960 --> 00:09:19.300 It was funny when I was watching PyBay. 00:09:19.560 --> 00:09:22.840 I was in the audience waiting for Michael Kennedy. 00:09:22.920 --> 00:09:23.720 I sat down right now. 00:09:24.400 --> 00:09:25.220 I kind of entered late. 00:09:25.500 --> 00:09:26.160 Sorry about that, Michael. 00:09:26.440 --> 00:09:27.280 But sat down, 00:09:27.520 --> 00:09:28.260 and he's up there. 00:09:28.260 --> 00:09:29.120 And I said, oh, 00:09:29.440 --> 00:09:30.320 that guy's kind of famous. 00:09:31.980 --> 00:09:35.900 And the person sitting next to me turned around and said, yeah, you're kind of famous too, 00:09:36.100 --> 00:09:36.240 man. 00:09:37.399 --> 00:09:38.320 So I'm like, 00:09:38.520 --> 00:09:39.700 no, not like Michael famous. 00:09:41.480 --> 00:09:42.240 Thanks for coming to talk. 00:09:42.260 --> 00:09:45.320 And your talk on Lean TD was really, really good. 00:09:45.720 --> 00:09:47.480 I think people really appreciated it as well. 00:09:47.520 --> 00:09:48.300 I'm waiting for that. 00:09:48.720 --> 00:09:51.600 One of the fun things about it is I got to talk about my career. 00:09:52.060 --> 00:09:52.660 So I, 00:09:53.060 --> 00:09:55.280 well, I didn't get to, I decided to. 00:09:55.460 --> 00:09:56.180 So the first few, 00:09:56.380 --> 00:09:57.620 like maybe five minutes of the talk 00:09:57.840 --> 00:10:00.720 are discussing my career in test and measurement, 00:10:00.980 --> 00:10:01.720 the RF world. 00:10:01.940 --> 00:10:02.380 That was fun. 00:10:02.540 --> 00:10:06.340 Yeah, if you want Brian's full story leading up to Lean TDD, 00:10:06.819 --> 00:10:08.820 eventually the videos for those talks should be out. 00:10:09.080 --> 00:10:12.340 And I do look forward to coming back for Python Byte's 00:10:12.920 --> 00:10:14.260 1,000th episode as well. 00:10:14.640 --> 00:10:14.740 Nice. 00:10:16.100 --> 00:10:16.240 Oh, 00:10:16.420 --> 00:10:16.480 yeah, 00:10:16.720 --> 00:10:16.860 yeah. 00:10:17.020 --> 00:10:18.980 We're going to have to start doing more than one episode. 00:10:19.900 --> 00:10:20.340 On the decade. 00:10:20.720 --> 00:10:21.480 No, I'm sure, 00:10:21.700 --> 00:10:21.900 honestly. 00:10:22.780 --> 00:10:23.780 It's very possible, 00:10:23.980 --> 00:10:24.860 which is very insane, 00:10:25.020 --> 00:10:25.140 honestly. 00:10:27.500 --> 00:10:30.580 Let's wrap this part out, this section out with just, 00:10:30.740 --> 00:10:32.220 we can maybe each look back. 00:10:32.480 --> 00:10:34.000 Think back to an innocent time. 00:10:34.440 --> 00:10:36.780 AI was a thing we made fun of because it was bad. 00:10:36.940 --> 00:10:37.640 There was no COVID, 00:10:37.940 --> 00:10:38.600 none of this stuff. 00:10:39.020 --> 00:10:40.180 Back to 2016, 00:10:40.640 --> 00:10:41.960 2017 when we started this. 00:10:42.459 --> 00:10:45.320 What's something notable that you all have seen change? 00:10:45.640 --> 00:10:46.740 I'll give you the final word. 00:10:46.800 --> 00:10:47.480 So Calvin, 00:10:47.680 --> 00:10:49.300 something from 2016, 00:10:49.600 --> 00:10:49.820 17, 00:10:50.000 --> 00:10:51.860 that era that really stands out for you. 00:10:52.399 --> 00:10:53.860 So anything from 2016, 00:10:54.020 --> 00:10:54.340 2017, 00:10:54.660 --> 00:10:55.240 around that time 00:10:55.660 --> 00:10:56.860 when you all started the podcast, 00:10:57.160 --> 00:10:58.980 I think the biggest thing that's changed for me 00:10:59.000 --> 00:11:01.320 has been packaging gotten way more sane. 00:11:01.620 --> 00:11:03.020 So I'm super excited about that. 00:11:03.400 --> 00:11:04.220 And the community is still here. 00:11:05.080 --> 00:11:06.040 PyCon still goes strong, 00:11:06.840 --> 00:11:09.220 even though we had a few years in there of weirdness 00:11:10.020 --> 00:11:10.660 around the world. 00:11:11.000 --> 00:11:12.040 And so I'm glad we're all still here. 00:11:12.260 --> 00:11:12.460 Yeah, 00:11:12.800 --> 00:11:14.100 packaging has absolutely changed. 00:11:14.440 --> 00:11:15.600 It still has a little ways to go. 00:11:15.760 --> 00:11:16.160 Way better. 00:11:16.160 --> 00:11:16.600 But what was it better? 00:11:16.680 --> 00:11:22.020 I'm still waiting for Python --build --arm my file someday. 00:11:22.460 --> 00:11:22.560 Right. 00:11:22.760 --> 00:11:22.960 The, 00:11:23.280 --> 00:11:23.340 the, 00:11:23.550 --> 00:11:23.960 a couple, 00:11:24.120 --> 00:11:24.760 there's a lot of stuff, 00:11:24.960 --> 00:11:27.840 but of course we don't really talk about Python 00:11:28.080 --> 00:11:28.880 two seven anymore, 00:11:29.280 --> 00:11:29.820 which is nice. 00:11:30.120 --> 00:11:30.320 Um, 00:11:30.530 --> 00:11:30.640 um, 00:11:30.900 --> 00:11:32.780 it's just gone or, 00:11:33.020 --> 00:11:33.440 you know, hopefully, 00:11:33.940 --> 00:11:35.880 and we'll get comments that people are saying 00:11:35.960 --> 00:11:37.480 like, I still have to deal with it, man. 00:11:37.960 --> 00:11:39.860 I'm sorry if you still have to deal with it. 00:11:40.110 --> 00:11:40.240 Uh, 00:11:40.560 --> 00:11:42.600 the other thing, packaging is a huge thing. 00:11:42.660 --> 00:11:43.060 I was just, 00:11:43.240 --> 00:11:44.900 I did, 00:11:45.160 --> 00:11:47.720 I've been updating a Python plugin or pytest plugin, 00:11:48.160 --> 00:11:49.540 pytest check recently. 00:11:50.100 --> 00:11:53.280 And the release process is push a tag. 00:11:54.080 --> 00:11:55.000 And then this, 00:11:55.400 --> 00:11:56.180 it's because it's this, 00:11:56.790 --> 00:12:00.600 the server builds it and talks with PyPI and does that. 00:12:00.670 --> 00:12:04.600 And I don't have to use Twine anymore by myself or anything like that. 00:12:04.670 --> 00:12:05.200 And the package, 00:12:05.520 --> 00:12:05.620 like, 00:12:05.800 --> 00:12:08.200 I'll have to just, I guess I'm marrying Calvin. 00:12:08.680 --> 00:12:11.100 The packaging system is something, 00:12:11.280 --> 00:12:13.140 it's so easy now that I don't think about it. 00:12:13.360 --> 00:12:15.780 Once I get it set up for a project, I just don't think about it. 00:12:16.300 --> 00:12:17.960 Yeah, and Nox and Tox. 00:12:18.450 --> 00:12:18.800 Oh, I still, 00:12:19.020 --> 00:12:19.300 yeah, 00:12:19.560 --> 00:12:20.240 Tox is great. 00:12:20.330 --> 00:12:20.880 I like Tox. 00:12:21.260 --> 00:12:22.540 We're just not left unsupervised. 00:12:23.480 --> 00:12:23.920 Unsupervised. 00:12:24.480 --> 00:12:29.460 Well, I was going to say, the thing that I was going to add is the two versus three wars 00:12:29.900 --> 00:12:31.440 and how many of us survived. 00:12:31.960 --> 00:12:32.560 But honestly, 00:12:32.780 --> 00:12:36.240 if you weren't around in that time or you weren't working in Python at the time, 00:12:36.460 --> 00:12:43.480 It was a long drawn out period where it was not entirely clear that Python was ever going to make it to Python 3. 00:12:43.640 --> 00:12:46.240 And then it would be permanently stuck in an older way. 00:12:46.540 --> 00:12:49.980 It was more shaky than I think people give it credit for being. 00:12:50.060 --> 00:12:54.260 And also, we just made a change that would have made the 2 to 3 transition easier. 00:12:54.600 --> 00:12:56.320 The making UTF-8 default. 00:12:56.820 --> 00:12:58.840 That should have been from the start. 00:12:58.960 --> 00:13:00.540 Just say it would have been a lot easier. 00:13:02.040 --> 00:13:04.140 So many cannot decode exceptions. 00:13:06.080 --> 00:13:08.360 Yeah, I was going to pick the two to three wars as well, 00:13:08.560 --> 00:13:09.900 but as a joke, of course. 00:13:09.950 --> 00:13:11.320 But honestly, that was a serious, 00:13:11.500 --> 00:13:12.100 serious thing. 00:13:12.290 --> 00:13:13.520 But I think I'll go for performance. 00:13:14.040 --> 00:13:15.400 Python has got a lot faster. 00:13:15.820 --> 00:13:21.640 The Python 310 Foundation and 311 especially made it quite a bit faster. 00:13:22.540 --> 00:13:23.240 And, you know, 00:13:23.440 --> 00:13:25.000 there's still continues today. 00:13:25.480 --> 00:13:26.720 There was a lot of, well, Python's too slow. 00:13:26.720 --> 00:13:30.220 We can't choose this for our project because obviously we got to use Go 00:13:30.470 --> 00:13:34.000 because that was the hot thing at the time before Rust really kicked into gear 00:13:34.330 --> 00:13:35.660 because we need better async. 00:13:35.840 --> 00:13:41.920 need faster x y and z and i think it's it's come a long way uh one final thing uh the or at least 00:13:41.920 --> 00:13:49.960 for me the um is that uh py test isn't considered niche anymore um it's uh pretty it's pretty obvious 00:13:50.050 --> 00:13:55.040 that people go with py test it's you don't have to defend it anymore which is cool and there was 00:13:55.040 --> 00:13:59.860 a book that probably helped maybe yeah who knows a couple couple books well i just want to take 00:14:00.240 --> 00:14:05.800 before before we wrap it up brian i just want to say you know thank you for 10 years of hard work 00:14:05.820 --> 00:14:09.980 People from the outside are like, oh, we just jump on for like 20 minutes or sometimes an hour. 00:14:10.100 --> 00:14:10.700 Then you get complaints. 00:14:10.740 --> 00:14:11.640 And it goes back to 20 minutes. 00:14:12.240 --> 00:14:14.360 No, you just jump on for a while and you have a quick chat. 00:14:14.420 --> 00:14:19.100 And there's so much behind the scenes that takes a lot of work to put this show together and keep it flowing. 00:14:19.720 --> 00:14:19.760 Yeah. 00:14:20.000 --> 00:14:20.600 Thanks for the hard work. 00:14:21.000 --> 00:14:21.400 Oh, you bet. 00:14:21.560 --> 00:14:21.780 Thank you. 00:14:22.640 --> 00:14:23.280 And yeah, 00:14:23.440 --> 00:14:24.420 thank you for keeping it going. 00:14:24.900 --> 00:14:28.120 So I'll just throw out also that Python people is going to come back. 00:14:28.280 --> 00:14:32.540 So I am going to do the podcasting thing more again. 00:14:32.840 --> 00:14:34.260 Well, you know, let us know in the episodes. 00:14:34.460 --> 00:14:34.560 Well, 00:14:35.020 --> 00:14:36.660 shout it out in the extras as you know how it goes. 00:14:37.040 --> 00:14:37.860 So thanks a lot. 00:14:38.050 --> 00:14:38.340 See you later. 00:14:39.060 --> 00:14:39.300 Awesome. 00:14:39.530 --> 00:14:40.440 Thank you so much, Brian, 00:14:40.620 --> 00:14:42.860 for being here and being in the show. 00:14:43.380 --> 00:14:45.920 Now let's get back to our group celebration. 00:14:46.960 --> 00:14:48.400 All right, let's jump straight into the items. 00:14:49.360 --> 00:14:50.240 And Nick, 00:14:50.730 --> 00:14:54.320 you have the probably most popular topic 00:14:54.450 --> 00:14:57.240 if you look at the derivative or second derivative 00:14:57.420 --> 00:14:59.740 of how fast it's got attention lately. 00:14:59.860 --> 00:15:00.160 Yeah. 00:15:00.230 --> 00:15:01.320 Tell us about this. 00:15:01.440 --> 00:15:01.800 This is crazy. 00:15:01.940 --> 00:15:04.600 If you have not heard about JEV, 00:15:04.760 --> 00:15:14.760 you have been living under a rock because it's the new kid in the block and it has made quite his presence known. 00:15:15.180 --> 00:15:16.700 It's been raving all over Twitter, 00:15:16.960 --> 00:15:17.640 X, Reddit. 00:15:18.280 --> 00:15:23.200 And what it is, is a new type of AI model that is not like your usual LLM. 00:15:24.280 --> 00:15:27.240 the folks from TypeSafe AI, 00:15:27.490 --> 00:15:31.880 these are the creators of JEV, I'll call it a system one decision model. 00:15:31.970 --> 00:15:35.100 And so system one, if you've read the book, Thinking Fast, 00:15:35.320 --> 00:15:35.840 Thinking Slow, 00:15:36.260 --> 00:15:36.600 they're like two 00:15:36.670 --> 00:15:37.460 different system types, 00:15:37.970 --> 00:15:39.480 the system one and system two. 00:15:39.920 --> 00:15:43.600 So JEV is like very focused on the system one 00:15:43.990 --> 00:15:45.380 type of inference. 00:15:45.870 --> 00:15:47.740 And what this means is that you have, 00:15:48.150 --> 00:15:49.660 instead of generating text, 00:15:50.420 --> 00:15:51.619 like a QN 00:15:51.640 --> 00:16:00.380 model or a ChatGPT model instead it makes decisions on either yes or no offers like gives 00:16:00.380 --> 00:16:06.060 you score probabilities for a rank of choices or it gives you like a score for like how how much 00:16:06.180 --> 00:16:11.260 something is like a range of like a range score about how some how much something is so it's very 00:16:11.620 --> 00:16:17.279 interesting and even though it sounds limited the use cases have been quite insane Vincent from 00:16:17.720 --> 00:16:25.120 Marimo did a video about it and he was benchmarking it against traditional LLMs because for the long, 00:16:25.200 --> 00:16:32.740 I'm a data scientist working and I work with a lot of ML models and classification is one of the 00:16:32.740 --> 00:16:35.160 things that, depending on the situation, 00:16:35.360 --> 00:16:36.800 can be really hard, can be really simple. 00:16:37.280 --> 00:16:38.020 And here comes 00:16:38.220 --> 00:16:42.260 this JEV model that is like a general classifier. 00:16:42.490 --> 00:16:45.679 And so he benchmarked it against a couple of 00:16:45.700 --> 00:16:49.620 real-world data sets and beat all of them. 00:16:49.790 --> 00:16:52.720 In most cases, he beat all of them. He even brought 00:16:53.070 --> 00:16:56.160 like an interesting trick he learned called SetFit. 00:16:56.320 --> 00:16:58.100 You should watch the video. He does a great 00:16:58.540 --> 00:17:05.120 explanation of it that allows you to train like very basic LLM models like BERT or 00:17:06.240 --> 00:17:09.560 QN embedding models and uses it for classification. 00:17:10.420 --> 00:17:12.959 Jev still does it better in terms of accuracy 00:17:12.959 --> 00:17:14.620 and so much cheaper. 00:17:15.199 --> 00:17:16.220 Now, the reason why I brought it up 00:17:16.360 --> 00:17:18.560 was because the use of JEV, 00:17:18.650 --> 00:17:19.880 even though it's really cheap, 00:17:20.130 --> 00:17:21.660 really fast, is really interesting. 00:17:22.020 --> 00:17:23.780 But there is another interesting point 00:17:23.920 --> 00:17:25.380 that Vincent made, 00:17:25.560 --> 00:17:27.699 whereby in most companies, 00:17:28.020 --> 00:17:29.800 there's a lot of data that is unlabeled 00:17:30.140 --> 00:17:32.080 and it is going to be more cost-effective 00:17:32.320 --> 00:17:33.680 to train your own local model 00:17:34.160 --> 00:17:35.860 on your own proprietary data 00:17:36.440 --> 00:17:38.220 than it is to use a general model, 00:17:38.320 --> 00:17:41.040 even if something as cheap and fast as JEV. 00:17:41.260 --> 00:17:51.180 So what you could do is to take the JEV model and help you create more accurate data sets, 00:17:51.620 --> 00:17:53.000 get you more cleaner labels. 00:17:53.600 --> 00:17:55.300 Is that kind of like supervised learning, 00:17:55.520 --> 00:17:56.700 but you use JEV as the supervisor? 00:17:57.580 --> 00:17:57.940 Yeah, exactly. 00:17:58.220 --> 00:18:04.460 So you take some general class and then you create a better data set so you can train supervised models. 00:18:05.100 --> 00:18:15.120 And this is how you like eke out a lot more performance and get better efficiency on very specific tasks to your like business. 00:18:15.440 --> 00:18:17.300 So that's one of the topics I wanted to bring up. 00:18:17.460 --> 00:18:18.740 It's a really interesting video. 00:18:19.320 --> 00:18:20.600 Jev is a really interesting model. 00:18:21.580 --> 00:18:22.580 You guys should go check it out. 00:18:22.920 --> 00:18:24.980 I've seen all this excitement about Jev and I'm like, 00:18:25.400 --> 00:18:27.020 I really want to have a use case for Jev. 00:18:28.520 --> 00:18:30.420 So I haven't had to use it, but I'm glad to hear that you have. 00:18:30.480 --> 00:18:31.160 And that's really cool. 00:18:31.280 --> 00:18:32.060 What do the rest of y'all think? 00:18:32.400 --> 00:18:39.460 Well, it's interesting that the number of classifier models that other labs are coming out with now, a lot of them are taking Quinn and fine tuning it to be a classifier. 00:18:39.470 --> 00:18:42.220 I think Microsoft just announced that's the one they're deploying, 00:18:43.060 --> 00:18:48.840 which I kind of like because I don't know these type secret folks or type secure folks very well. 00:18:49.060 --> 00:18:49.120 But, 00:18:49.390 --> 00:18:52.460 you know, you're sending them all their data because there's no option to use it locally. 00:18:52.840 --> 00:18:53.020 Yeah. 00:18:53.380 --> 00:18:57.120 Yeah, I think it was kind of interesting because when I first heard about Jev, it was like, 00:18:57.280 --> 00:18:57.800 what's the big deal? 00:18:58.020 --> 00:19:01.720 And I understood then I learned that all it did was the very key thing was, 00:19:01.800 --> 00:19:02.600 is does classification, 00:19:02.960 --> 00:19:03.040 right? 00:19:03.180 --> 00:19:06.260 It's not an open-ended write me this code is like answer this. 00:19:06.300 --> 00:19:06.400 Yes, 00:19:06.560 --> 00:19:07.760 no question or put, 00:19:07.980 --> 00:19:08.240 put every, 00:19:08.500 --> 00:19:09.660 put these things in a bucket. 00:19:09.880 --> 00:19:09.980 Right. 00:19:10.200 --> 00:19:15.320 And I think Nick's point about like using that as a quick and dirty way to 00:19:15.980 --> 00:19:19.620 label some data so that you can do your own training model with like scikit 00:19:19.760 --> 00:19:19.920 learn, 00:19:20.360 --> 00:19:23.340 which I will always give a shout out to because it's still my wife's favorite 00:19:23.640 --> 00:19:24.580 Python library in the world. 00:19:25.680 --> 00:19:26.760 It is a great thing, right? 00:19:26.840 --> 00:19:27.960 because I think it's one of these things 00:19:28.040 --> 00:19:29.120 we're also really starting to learn 00:19:29.340 --> 00:19:32.100 is using LLMs to help you 00:19:32.700 --> 00:19:34.200 build the automation for things 00:19:34.800 --> 00:19:36.500 is a better use of your time and money 00:19:36.820 --> 00:19:37.300 slash tokens 00:19:37.940 --> 00:19:39.700 than it is to like ask the models 00:19:39.840 --> 00:19:40.980 to do the same thing over and over 00:19:41.380 --> 00:19:42.220 rather than like, yeah, 00:19:42.400 --> 00:19:43.320 let's use it to help that, 00:19:43.740 --> 00:19:46.080 like use second learn to actually do like full, 00:19:46.200 --> 00:19:47.520 like I'm going to use the long term here. 00:19:47.600 --> 00:19:48.620 So I'm going to try to use the right term, 00:19:48.900 --> 00:19:50.280 but use second learn to build the local model 00:19:50.360 --> 00:19:51.320 that you need to do the classification. 00:19:51.760 --> 00:19:51.880 Yeah. 00:19:52.200 --> 00:19:53.160 That'll be the fastest, cheapest, 00:19:53.540 --> 00:19:54.840 best reproducible way to do it. 00:19:54.980 --> 00:19:56.380 And I think it's great that this is kind of 00:19:56.320 --> 00:19:59.320 helping push that narrative down more and more to more people. 00:19:59.840 --> 00:19:59.960 Yeah, 00:20:00.180 --> 00:20:04.940 it's so much more efficient to have like a model like Jev or a local model than 00:20:04.940 --> 00:20:07.420 to have like Opus running your like home assistant, 00:20:08.140 --> 00:20:08.240 right? 00:20:09.500 --> 00:20:12.340 That's like very cost, like cost not inefficient, 00:20:13.140 --> 00:20:13.640 cost efficient. 00:20:14.080 --> 00:20:14.380 Yeah. 00:20:14.670 --> 00:20:20.040 I mean, I think it seems like it would be something that could be used with like spaCy when you're 00:20:20.280 --> 00:20:23.720 annotating parts of natural language. 00:20:24.460 --> 00:20:27.300 And I think there's some creative ways that, 00:20:27.600 --> 00:20:29.760 you know, you can get decision maybe on a 00:20:29.920 --> 00:20:30.980 first pass or something. 00:20:31.600 --> 00:20:35.660 And I'm looking forward to looking at some of the research because I have heard there's 00:20:35.660 --> 00:20:38.120 a few others at the top of my head. 00:20:38.120 --> 00:20:39.160 I can't name them, 00:20:39.400 --> 00:20:45.080 but some former folks from DeepMind are doing something in the UK, 00:20:45.260 --> 00:20:45.380 I 00:20:45.480 --> 00:20:46.120 believe as well. 00:20:46.400 --> 00:20:46.900 So yeah, 00:20:47.300 --> 00:20:48.620 the real feature here is speed. 00:20:48.920 --> 00:20:53.480 The fact that you can feed it just hundreds of thousands of rows and get answers in microseconds 00:20:53.560 --> 00:20:54.140 and milliseconds. 00:20:54.200 --> 00:20:59.180 yeah yeah that's not a normal ai sort of experience right that's a big deal though 00:20:59.960 --> 00:21:05.840 it's a huge deal all right awesome i definitely think that's a a good pick brett tell us about 00:21:05.890 --> 00:21:11.840 this new council yeah i know uh what the heck am i talking about uh so for those you didn't hear 00:21:12.380 --> 00:21:18.479 um first two weeks of september we had an election i think uh this was covered on python bytes uh for 00:21:18.500 --> 00:21:19.680 Python Packaging Council. 00:21:19.960 --> 00:21:20.500 It is the 00:21:20.620 --> 00:21:21.240 inaugural one. 00:21:22.460 --> 00:21:24.400 And it was all set up via 00:21:24.560 --> 00:21:25.260 PEP772. 00:21:25.780 --> 00:21:26.420 And the 00:21:26.540 --> 00:21:27.000 elections happened. 00:21:27.200 --> 00:21:28.340 It was open to all 00:21:28.540 --> 00:21:30.360 PSF members who 00:21:31.420 --> 00:21:32.380 basically said 00:21:32.380 --> 00:21:33.240 they wanted to vote in it. 00:21:33.600 --> 00:21:34.320 And it 00:21:34.340 --> 00:21:34.460 happened. 00:21:34.700 --> 00:21:36.420 It ended on the 00:21:36.560 --> 00:21:38.300 15th or 16th, probably depending on where you 00:21:38.460 --> 00:21:39.040 were on the planet. 00:21:40.400 --> 00:21:41.340 I ended up on it. 00:21:42.420 --> 00:21:43.340 Ended up on it. 00:21:43.680 --> 00:21:44.460 Dolled stuff as well. 00:21:44.680 --> 00:21:44.920 Henry 00:21:45.620 --> 00:21:47.200 Schreiner and Ralph Gomers. 00:21:47.700 --> 00:21:51.880 So we're all in this inaugural council. 00:21:52.560 --> 00:21:56.700 Perdune and I are on it for two years as the top vote getters. 00:21:57.010 --> 00:22:00.160 And then Donald, Henry, and Ralph are going to be out for a year. 00:22:00.320 --> 00:22:01.360 And then going forward after that, 00:22:01.620 --> 00:22:03.640 all people on it are going to serve for two years. 00:22:04.640 --> 00:22:07.620 So interesting thing with this is it's the first one. 00:22:08.110 --> 00:22:11.340 I will say we've already met four times and actually already had to meet with the steering council. 00:22:11.700 --> 00:22:14.460 And a lot of it's been around logistics and how to organize ourselves. 00:22:15.220 --> 00:22:17.700 As any of you have had to start anything like this from scratch, 00:22:17.960 --> 00:22:19.720 it's about where are we going to keep our notes? 00:22:20.320 --> 00:22:21.900 Where are we going to let people reach us? 00:22:22.080 --> 00:22:22.760 And that kind of stuff. 00:22:22.980 --> 00:22:24.420 So we're still figuring things out. 00:22:25.040 --> 00:22:27.320 The one thing I will say about this council is 00:22:27.760 --> 00:22:32.260 it is explicitly meant to be in charge of the sentence specifications 00:22:32.900 --> 00:22:35.160 and the whole website in general for packaging up python.org 00:22:36.020 --> 00:22:37.780 where you can go and read about all the specs 00:22:38.450 --> 00:22:42.000 relating to the kind of the PyPI slash PyPA side of the world. 00:22:42.920 --> 00:22:43.240 But otherwise, 00:22:43.960 --> 00:22:45.920 PEP772, which established this council, 00:22:46.600 --> 00:22:48.400 did not try to dictate what this council is going to do. 00:22:48.540 --> 00:22:54.280 So we are actually also currently talking about what do we want to see our function be, right? 00:22:54.440 --> 00:22:59.980 Because we're under the auspices of the PSF, hence why they ran the election. 00:23:00.500 --> 00:23:03.820 But we're trying to figure out how do we represent people, 00:23:04.000 --> 00:23:04.140 basically, 00:23:04.480 --> 00:23:04.560 right? 00:23:04.620 --> 00:23:06.100 Like we represent specs, 00:23:06.270 --> 00:23:10.360 but we also want to somehow act as like the public point, I think, 00:23:10.640 --> 00:23:14.280 for and I'm speaking for myself as generalizing for my fellow council members. 00:23:14.880 --> 00:23:15.040 Basically, 00:23:15.200 --> 00:23:17.040 it's kind of the public face of packaging in Python, 00:23:17.260 --> 00:23:17.340 right? 00:23:17.500 --> 00:23:21.280 Who can you go talk to to maybe get funding to make something better? 00:23:21.720 --> 00:23:29.820 Or if we need to go talk to like integrators like the links distros or conda or anyone else who integrates Python and a bunch of other stuff. 00:23:30.160 --> 00:23:32.520 who can you go talk to to get some answers, 00:23:33.070 --> 00:23:36.680 get some questions answered or kind of talk to them to figure out what they need and all that stuff. 00:23:36.860 --> 00:23:39.080 Kind of similar to what the SC does around the language, 00:23:39.880 --> 00:23:41.060 but obviously it's basically about packaging. 00:23:41.760 --> 00:23:42.340 And probably honestly, 00:23:42.760 --> 00:23:44.420 I'm going to guess a little bit around workflow as well, 00:23:44.930 --> 00:23:48.020 because it doesn't fall basically stuff that doesn't fall under the language that way. 00:23:48.980 --> 00:23:49.260 But yeah, 00:23:49.470 --> 00:23:53.020 it was just basically I noticed admin and also I figured I'd bring that up. 00:23:53.960 --> 00:23:54.200 And yeah, 00:23:54.420 --> 00:23:56.040 it's as I said, we've just been going. 00:23:56.960 --> 00:23:59.560 We have a issue tracker on GitHub. 00:23:59.740 --> 00:24:02.480 We haven't made public yet because we're getting our initial notes and stuff set up, 00:24:02.640 --> 00:24:02.840 but we're 00:24:02.840 --> 00:24:04.340 going to try to be as transparent as possible. 00:24:05.020 --> 00:24:09.060 So like we're planning to keep our media notes up there and keep track of what we're going 00:24:09.060 --> 00:24:12.520 to talk about, what's on deck and all that and be accessible. 00:24:13.480 --> 00:24:16.240 Discuss the Python.org is where a lot of the packaging discussions happen. 00:24:16.660 --> 00:24:17.620 So you can find us there. 00:24:18.220 --> 00:24:18.680 And yeah, 00:24:19.160 --> 00:24:22.119 we're getting going, but we have high hopes that we can hopefully up the throughput 00:24:22.140 --> 00:24:29.960 a bit on stuff because poor Paul Moore used to be the vanilla dictator for packaging. 00:24:30.630 --> 00:24:36.200 Now it gets a bit of a break and at least spreading across the five of us will start 00:24:36.200 --> 00:24:36.800 some things a bit. 00:24:37.120 --> 00:24:41.280 And we can keep doing stuff such as talking about wheel variants. 00:24:41.510 --> 00:24:45.900 And I know there's wheel next being proposed and a whole bunch of kind of big projects that 00:24:46.080 --> 00:24:49.160 hopefully we can help go forward and make packaging for Python just better for everybody. 00:24:49.320 --> 00:24:52.880 Well, because packaging has always been such a simple and easy story in the Python world. 00:24:53.220 --> 00:24:54.640 So simple and so easy. 00:24:54.900 --> 00:24:55.300 I know. 00:24:55.920 --> 00:25:00.700 C and Fortran and all these other languages made it so easy for us to integrate that I 00:25:00.800 --> 00:25:01.940 don't know why anyone's ever complained. 00:25:02.160 --> 00:25:03.220 I'm glad you all are here. 00:25:03.560 --> 00:25:12.680 Yeah, but I will say like the journey of Python packaging from like cheese to eggs to wheels, 00:25:13.280 --> 00:25:18.300 it's actually been pretty amazing to see the evolution of Python packaging. 00:25:18.400 --> 00:25:18.980 So thanks, 00:25:19.300 --> 00:25:21.100 Brett, for championing that. 00:25:21.190 --> 00:25:22.800 I feel like you care so much about that. 00:25:23.220 --> 00:25:28.040 And that's why it's still you and others are making it such a pleasant, 00:25:28.260 --> 00:25:29.180 at least more pleasant 00:25:29.740 --> 00:25:30.100 experience. 00:25:30.680 --> 00:25:31.120 Well, thank you. 00:25:31.120 --> 00:25:32.040 I appreciate the kind words. 00:25:32.240 --> 00:25:32.400 Yeah, 00:25:32.580 --> 00:25:34.980 it has been a bit of a journey for packaging and Python in general, 00:25:35.160 --> 00:25:35.280 right? 00:25:35.360 --> 00:25:40.000 Like the creation of the Cheeshop slash Python Packaging Index, 00:25:40.260 --> 00:25:40.960 because people were too worried 00:25:41.080 --> 00:25:42.980 we'd upset managers if we called it the Cheeshop. 00:25:43.320 --> 00:25:44.500 FYI, if you go to a 404, 00:25:44.840 --> 00:25:47.660 it still links to the Cheeshop sketch on YouTube from Monty Python. 00:25:48.000 --> 00:25:48.320 for a reason. 00:25:48.720 --> 00:25:49.880 I'm glad that's still there 00:25:49.940 --> 00:25:52.060 because I will, being of an age, 00:25:52.140 --> 00:25:53.680 I will still call it the cheese shop. 00:25:55.160 --> 00:25:56.100 Yeah, you've got to do it 00:25:56.320 --> 00:25:56.960 on PyPI, though. 00:25:57.240 --> 00:25:58.160 Oh, PyPI, okay. 00:25:58.360 --> 00:25:59.240 Sorry if I said Python. 00:25:59.480 --> 00:26:00.000 That was my 00:26:00.240 --> 00:26:00.300 mistake. 00:26:00.460 --> 00:26:00.720 No, no. 00:26:01.900 --> 00:26:02.760 Yes, there it is. 00:26:03.220 --> 00:26:03.500 There you go. 00:26:05.840 --> 00:26:06.740 But yeah, it's been a journey, 00:26:06.980 --> 00:26:07.840 right? And it's been very driven 00:26:07.900 --> 00:26:08.880 by the community and, 00:26:09.240 --> 00:26:09.800 to be frank, 00:26:10.100 --> 00:26:11.280 very much driven by volunteers. 00:26:12.380 --> 00:26:13.900 So it's moved at the space, 00:26:14.160 --> 00:26:15.879 the speed of people having time 00:26:15.900 --> 00:26:16.200 to help, 00:26:16.550 --> 00:26:17.840 but it's been progressing 00:26:18.130 --> 00:26:19.660 forward and it's something I think everyone cares about. 00:26:20.040 --> 00:26:21.820 And as I kind of joked, doing the best 00:26:21.830 --> 00:26:23.840 we can in the face of Python being the glue 00:26:23.980 --> 00:26:25.760 language of the internet means we have to 00:26:25.860 --> 00:26:27.700 integrate with some languages who do not really 00:26:28.000 --> 00:26:29.800 have their up together in terms of 00:26:30.040 --> 00:26:30.780 packaging themselves. 00:26:31.170 --> 00:26:31.780 And so how do we 00:26:31.900 --> 00:26:32.540 work around that, 00:26:32.720 --> 00:26:33.020 right? Like 00:26:33.780 --> 00:26:35.700 we can go to the C community or 00:26:35.800 --> 00:26:37.840 C++ or any of these other compiled languages 00:26:37.870 --> 00:26:38.140 and go like, 00:26:38.360 --> 00:26:38.420 well, 00:26:38.580 --> 00:26:39.320 hey, how do you package? 00:26:39.520 --> 00:26:39.580 Like, 00:26:39.920 --> 00:26:40.080 oh, 00:26:40.400 --> 00:26:40.740 shell scripts. 00:26:41.540 --> 00:26:41.660 Oh, 00:26:41.910 --> 00:26:42.320 okay, great. 00:26:42.610 --> 00:26:43.640 So how do we integrate with that? 00:26:43.800 --> 00:26:43.920 Right? 00:26:44.140 --> 00:26:45.340 So it's been quite the journey. 00:26:45.740 --> 00:26:52.900 And I would like to think we've gotten to a point where the ergonomics are good and it's not painful anymore. 00:26:54.020 --> 00:26:58.780 But I think there's some opportunities for it to be better and definitely faster. 00:26:59.480 --> 00:27:01.140 And obviously the work's not done. 00:27:01.440 --> 00:27:01.660 Yeah, 00:27:01.900 --> 00:27:02.180 I mean, 00:27:02.440 --> 00:27:03.620 it's been, I mean, 00:27:03.840 --> 00:27:05.880 just the council itself, 00:27:06.740 --> 00:27:11.880 it's been a multi-year journey just to get the PEP approved and written. 00:27:12.780 --> 00:27:12.980 And, 00:27:14.100 --> 00:27:25.220 you know, it was great to see the number of folks that ran for the Packaging Council and really pleased by the people that were elected or chosen. 00:27:26.530 --> 00:27:30.200 I think it's a really great cross section of historical knowledge, 00:27:31.020 --> 00:27:32.020 scientific knowledge, 00:27:33.360 --> 00:27:33.580 you know, 00:27:33.940 --> 00:27:36.000 across a number of different tools. 00:27:37.040 --> 00:27:38.500 And, you know, 00:27:38.840 --> 00:27:44.180 having Brett having been part of the steering council when we did the inaugural steering council, 00:27:44.660 --> 00:27:46.100 bootstrapping something like this, 00:27:46.260 --> 00:27:47.420 there's some knowledge in there. 00:27:47.440 --> 00:27:48.420 So that's awesome. 00:27:48.860 --> 00:27:52.240 But one of the things I'm super happy about is, 00:27:52.620 --> 00:27:52.840 you know, 00:27:53.000 --> 00:27:56.940 having more emphasis on the standards and, 00:27:58.320 --> 00:28:00.580 you know, there's going to be different tool implementations. 00:28:00.740 --> 00:28:01.300 and that's great. 00:28:01.630 --> 00:28:04.680 I would love to see us get more of a focus 00:28:05.060 --> 00:28:06.940 on the consumers of packages 00:28:07.580 --> 00:28:10.740 and people who are using our packages on a day-to-day 00:28:11.120 --> 00:28:12.720 and making that more pleasant 00:28:13.090 --> 00:28:15.560 like when you're using a GPU and other things 00:28:15.820 --> 00:28:17.260 because the landscape has changed. 00:28:17.760 --> 00:28:20.800 So thank you, Brett, for continuing to serve us. 00:28:21.300 --> 00:28:21.480 Yeah, 00:28:21.700 --> 00:28:21.960 thanks, Brett. 00:28:22.560 --> 00:28:25.220 You have this legacy of launching councils. 00:28:25.760 --> 00:28:26.620 So that's very cool. 00:28:27.780 --> 00:28:31.200 I feel like it seemed like packaging was kind of solved. 00:28:31.310 --> 00:28:33.680 And then uv came along and we're like, well, now it's really solved. 00:28:34.020 --> 00:28:35.100 But then we also have AI, 00:28:35.700 --> 00:28:39.620 which means CUDA and all the crazy stuff that Wheel Next is trying to solve. 00:28:39.790 --> 00:28:44.680 And we've got the brand new PEP that got accepted for Emscripten and WebAssembly, 00:28:45.060 --> 00:28:46.180 which is a whole nother angle. 00:28:46.540 --> 00:28:48.400 And so I don't feel like it's solved for sure. 00:28:48.720 --> 00:28:48.960 Well, 00:28:49.140 --> 00:28:51.120 and Pixie has done a great job, 00:28:51.380 --> 00:28:51.460 too, 00:28:51.620 --> 00:28:52.880 in the scientific community. 00:28:53.140 --> 00:28:53.420 So, 00:28:53.780 --> 00:28:57.400 I mean, I think tools are kind of reflecting their users, 00:28:57.940 --> 00:29:00.420 and I think the standards are what makes that possible. 00:29:00.900 --> 00:29:06.140 So it's great to see more domains being represented within packaging. 00:29:06.400 --> 00:29:06.880 A question, 00:29:07.620 --> 00:29:14.820 Brett, is the council going to be consulting or working with our security developer in residence? 00:29:15.210 --> 00:29:15.680 Because of like, 00:29:16.100 --> 00:29:17.340 as you know, like we have all these, 00:29:17.530 --> 00:29:19.200 we have more CVEs being, 00:29:19.880 --> 00:29:20.520 you know, issued, 00:29:20.920 --> 00:29:22.420 especially attacking packaging. 00:29:23.080 --> 00:29:23.320 Yeah. 00:29:23.510 --> 00:29:23.660 So, 00:29:25.590 --> 00:29:25.780 yeah, 00:29:26.020 --> 00:29:26.920 there's Seth Michael Larson, 00:29:27.400 --> 00:29:30.460 who's the security developer in residence for Python itself. 00:29:30.940 --> 00:29:32.440 There's Michael Fiedler, 00:29:32.590 --> 00:29:34.640 who works on PyPI directly. 00:29:35.640 --> 00:29:36.040 And yeah, 00:29:37.130 --> 00:29:38.040 pretty much everyone on the council 00:29:38.150 --> 00:29:41.000 either knows either of them directly personally 00:29:41.930 --> 00:29:45.740 or because we're working under the PSF in a way, 00:29:46.860 --> 00:29:48.640 we know we can always just ask like, 00:29:48.750 --> 00:29:50.660 hey, can we talk to Mike about something or whatever? 00:29:51.250 --> 00:29:54.120 But I message Mike and Seth directly about stuff. 00:29:55.200 --> 00:29:56.560 So it's definitely there. 00:29:57.120 --> 00:30:00.060 And for instance, I will say part of my platform for running 00:30:00.400 --> 00:30:01.860 was on secure supply chain, 00:30:02.110 --> 00:30:02.240 right? 00:30:02.480 --> 00:30:08.020 So I will say I personally am interested slash invested in trying to see that improve. 00:30:08.200 --> 00:30:12.000 So security and how packaging plays into that is definitely going to be a thing. 00:30:12.280 --> 00:30:13.520 Now, I will say we, 00:30:14.100 --> 00:30:16.660 PyPI does not refer to us or vice versa. 00:30:17.080 --> 00:30:18.940 So it's very much going to be a collaboration. 00:30:20.060 --> 00:30:23.700 But we are definitely aware that it is something that we want to work with PyPI to make sure 00:30:23.700 --> 00:30:26.560 that anything we do will work well for them and everyone else in the ecosystem, 00:30:26.760 --> 00:30:26.820 right? 00:30:27.120 --> 00:30:29.800 Because as Carol pointed out, we have a pretty good, I think, 00:30:30.180 --> 00:30:31.200 breadth of coverage for the 00:30:31.200 --> 00:30:35.860 very broad community that we have in Python on this council to make sure that all corners, 00:30:36.200 --> 00:30:37.320 whether it's scripting, 00:30:37.780 --> 00:30:38.000 web, 00:30:38.360 --> 00:30:38.820 data science, 00:30:39.540 --> 00:30:40.780 or an LLM, 00:30:41.120 --> 00:30:43.260 you name it, I think someone on this 00:30:43.400 --> 00:30:46.000 council has a connection to somebody to make sure that we cover all the bases. 00:30:46.260 --> 00:30:46.820 And that includes 00:30:47.040 --> 00:30:47.200 security. 00:30:47.620 --> 00:30:50.920 And it's definitely something we have top of mind to make sure that we address. 00:30:51.280 --> 00:30:51.380 So. 00:30:51.740 --> 00:30:51.840 Yeah. 00:30:52.220 --> 00:30:53.020 That's a good question. 00:30:53.620 --> 00:30:57.519 The security side of any sort of packaging is definitely so front 00:30:57.540 --> 00:31:04.380 center these days you know but you know else is many many parsing many things speed michael speed 00:31:04.540 --> 00:31:10.560 is the feature we've talked about this before and it's interesting nick brought up jev we brought up 00:31:10.560 --> 00:31:15.180 the fact like this speed is a feature for like the a lot of the astral tools like rough and uv 00:31:15.780 --> 00:31:21.860 but most folks don't think of python as being fast at parsing json the this article here is a blog 00:31:21.840 --> 00:31:27.500 post from Daniel Lemire. He benchmarks a new wrapper around SIMD JSON, 00:31:27.720 --> 00:31:29.260 which is a C++ library 00:31:29.280 --> 00:31:29.980 I've never used, 00:31:30.140 --> 00:31:35.780 but he wrote a fast SIMD JSON wrapper around it and actually demonstrates parsing 00:31:36.100 --> 00:31:39.640 tens of gigabytes of JSON per second in Python. 00:31:40.160 --> 00:31:40.260 Now, 00:31:40.560 --> 00:31:40.900 there's a trick, 00:31:41.400 --> 00:31:41.560 obviously. 00:31:42.200 --> 00:31:43.080 Doing things in 00:31:43.280 --> 00:31:44.980 async, you know, is always not necessarily straightforward. 00:31:45.140 --> 00:31:46.280 Doing things threading is 00:31:46.420 --> 00:31:47.080 obviously never straightforward. 00:31:47.840 --> 00:31:50.220 Free threading brings a new wrinkle to all this, and a lot of 00:31:50.120 --> 00:31:55.840 folks have probably not tried out the free threading capabilities of Python 3.14 and now 3.15. 00:31:56.200 --> 00:32:00.360 This is actually a really good example, or at least a toy use case where boy, does it make a 00:32:00.660 --> 00:32:00.980 big difference. 00:32:01.280 --> 00:32:02.520 So he goes through, 00:32:03.740 --> 00:32:06.020 he's got a dual Xeon 64 core machine. 00:32:06.630 --> 00:32:07.440 So he's actually able 00:32:07.440 --> 00:32:14.520 to take that free threading and spread this workload across 64 different cores simultaneously. 00:32:15.160 --> 00:32:20.520 This is a GitHub archive snapshot of, I think, what, eight hours of GitHub data, 00:32:21.090 --> 00:32:21.440 which turns out 00:32:21.440 --> 00:32:25.040 to be about 7.6 gigabytes of JSON decompressed, 00:32:25.410 --> 00:32:26.820 2.2 million events, 00:32:27.440 --> 00:32:27.940 runs it through. 00:32:28.300 --> 00:32:28.860 The very 00:32:29.060 --> 00:32:31.560 first one he did was basically a non-threaded, 00:32:31.990 --> 00:32:32.120 non, 00:32:32.580 --> 00:32:34.440 it's like single-threaded version of this. 00:32:35.050 --> 00:32:37.060 So if you are looking at the standard JSON library, 00:32:37.630 --> 00:32:41.419 it only basically takes about 42 seconds to 00:32:41.440 --> 00:32:47.160 parsed the whole set, which are basically new line return JSON messages for every pull request that 00:32:47.160 --> 00:32:47.920 has happened in GitHub. 00:32:48.420 --> 00:32:51.940 So you can see here 0.18 gigabytes per second of throughput. 00:32:52.460 --> 00:32:52.580 The 00:32:53.200 --> 00:32:56.120 wrapper, he does the same thing for ORJSON, 00:32:57.160 --> 00:32:58.140 fast SIMD, 00:32:58.400 --> 00:33:00.280 JSON map mini, which is another approach he 00:33:00.400 --> 00:33:00.500 took. 00:33:00.780 --> 00:33:01.740 And then the lazy views, 00:33:01.920 --> 00:33:04.740 which is basically don't actually construct the dictionaries until I 00:33:04.940 --> 00:33:06.200 ask for them version of this. 00:33:06.640 --> 00:33:10.380 That got the 2.13 gigabytes per second. So that's obviously not as 00:33:10.320 --> 00:33:12.000 as fast as tens of gigabytes per second, 00:33:12.430 --> 00:33:13.880 then moving into free threading. 00:33:13.890 --> 00:33:16.720 So if you open up and start using all 64 cores, 00:33:17.280 --> 00:33:18.380 and still in one process, 00:33:18.940 --> 00:33:19.680 one Python process, 00:33:19.960 --> 00:33:21.800 64 cores on free threading, 00:33:23.120 --> 00:33:28.140 65.7 gigabytes per second of reading in that JSON file. 00:33:28.520 --> 00:33:30.240 So I think you could see some magic here 00:33:30.240 --> 00:33:33.660 if you combined a large JSON workload with tools like JEV, 00:33:34.020 --> 00:33:35.700 where you wanted to do some real-time processing 00:33:35.890 --> 00:33:38.240 and pull back the results as fast as possible. 00:33:38.800 --> 00:33:40.880 This is a very valid use case for this, 00:33:41.460 --> 00:33:42.220 but it's only one of many. 00:33:42.740 --> 00:33:45.420 The other bit here, he uses that there's a lazy trick 00:33:45.440 --> 00:33:46.140 in here, which I mentioned, 00:33:46.320 --> 00:33:48.500 which he doesn't instantiate the dictionaries 00:33:48.620 --> 00:33:49.500 until they're asked for. 00:33:49.860 --> 00:33:51.260 But in many cases that may be useful. 00:33:51.660 --> 00:33:54.320 You may only want to filter or pull back portions 00:33:54.440 --> 00:33:55.840 of the data and not the whole dataset. 00:33:56.040 --> 00:33:57.480 'Cause like for example, in this use case, 00:33:57.820 --> 00:34:01.380 I think each message is about 3K inside that dataset. 00:34:01.740 --> 00:34:04.580 So fairly large object each time. 00:34:05.180 --> 00:34:05.700 Let's see here. 00:34:06.320 --> 00:34:11.879 basically a 31 times speed up over single chord regular speed, 00:34:12.220 --> 00:34:12.840 which is pretty incredible. 00:34:14.060 --> 00:34:17.840 It's also a simple, you kind of got to take this a little bit of grain of salt. 00:34:18.080 --> 00:34:18.580 It's a simple use 00:34:18.780 --> 00:34:18.940 case, 00:34:19.679 --> 00:34:22.520 but if you've got the work that looks kind of like this, you may want to check out, 00:34:22.720 --> 00:34:22.960 he's what 00:34:22.990 --> 00:34:28.060 the maintainer of this wrapper around the simdjson library itself. 00:34:28.590 --> 00:34:29.260 The benchmarks, 00:34:29.800 --> 00:34:30.620 he links to them 00:34:30.790 --> 00:34:34.040 in the code. So in that blog post, you'll see the code used in the blog post. 00:34:34.500 --> 00:34:35.260 Here's all the, 00:34:35.379 --> 00:34:39.340 if you wanted to reproduce this yourself on your own hardware and see how that goes. 00:34:40.100 --> 00:34:40.980 And it's 00:34:41.120 --> 00:34:45.020 available on PyPI as fast as MDJSON. 00:34:45.179 --> 00:34:45.460 Actually, 00:34:45.679 --> 00:34:47.700 there's a new release since I looked at this 00:34:48.110 --> 00:34:48.679 an hour ago. 00:34:48.929 --> 00:34:52.540 So very active work going into this. I think it's getting a little bit of attention 00:34:52.850 --> 00:34:54.840 out there in the internets. 00:34:55.480 --> 00:34:56.240 I'm impressed. 00:34:56.780 --> 00:34:57.240 That's pretty awesome. 00:34:57.480 --> 00:34:59.640 And kind of a not explored free 00:34:59.760 --> 00:35:03.299 threading much because I don't have this kind of a workload. But boy, does that make a great 00:35:03.320 --> 00:35:09.080 argument for python is not slow once we got unleashed upon all the cores yeah i mean there's 00:35:09.080 --> 00:35:15.300 an old saying in the community right python's fast enough and yeah he really is and prototype it and 00:35:15.300 --> 00:35:19.620 if it's not fast enough then you start to look at what you have to do speed up and i think free 00:35:19.780 --> 00:35:23.000 threading adds a new wrinkle to it because originally it was always the answer yeah make 00:35:23.000 --> 00:35:26.840 sure you're using the proper data structure and algorithms right that usually got people most of 00:35:26.840 --> 00:35:30.820 the way and then at the end you'd throw in that extra wrapper traditionally would have been in 00:35:30.840 --> 00:35:32.460 like C or C++ nowadays, 00:35:32.900 --> 00:35:33.300 probably Rust. 00:35:33.640 --> 00:35:33.880 Probably Rust, 00:35:34.050 --> 00:35:34.140 yeah. 00:35:34.190 --> 00:35:35.800 But I think free threading throws in a new, 00:35:36.300 --> 00:35:36.620 well, 00:35:36.840 --> 00:35:38.020 can you go multi-threaded 00:35:38.140 --> 00:35:39.000 and get a win that way, 00:35:39.270 --> 00:35:39.380 right? 00:35:39.700 --> 00:35:40.740 And because, I mean, 00:35:41.040 --> 00:35:42.900 you want to make sure you go down 00:35:43.100 --> 00:35:45.080 the ladder of complexity, 00:35:45.400 --> 00:35:45.540 right? 00:35:45.590 --> 00:35:47.760 And the shortest one is write data structures, 00:35:48.100 --> 00:35:48.500 write algorithm, 00:35:48.880 --> 00:35:50.100 then it's throw in free threading, 00:35:50.500 --> 00:35:51.720 then you start talking about 00:35:51.750 --> 00:35:52.900 whether or not you want to go all the way out 00:35:53.060 --> 00:35:53.760 to a whole other language. 00:35:54.260 --> 00:35:54.340 Yeah, 00:35:54.530 --> 00:35:56.740 because each step gets progressively 00:35:57.030 --> 00:35:58.980 more difficult to triage and debug. 00:35:59.500 --> 00:35:59.860 Exactly. 00:36:00.340 --> 00:36:02.380 But I think that's a good point to me, Calvin. 00:36:02.500 --> 00:36:02.740 It's like, 00:36:02.970 --> 00:36:04.240 I think this is what's going to start happening 00:36:04.270 --> 00:36:06.500 is people are going to start using free threading, 00:36:06.660 --> 00:36:07.340 especially in the libraries 00:36:07.490 --> 00:36:08.960 where you directly don't have to maintain 00:36:09.370 --> 00:36:10.340 or manage the complexity 00:36:10.700 --> 00:36:11.960 and it will take care of it for you. 00:36:12.360 --> 00:36:12.940 In which case, 00:36:13.310 --> 00:36:14.740 I think things start to open up a bit 00:36:14.830 --> 00:36:16.620 under these new free threading builds. 00:36:17.080 --> 00:36:18.200 And IO is always a big one, 00:36:18.380 --> 00:36:18.440 right? 00:36:18.740 --> 00:36:21.640 How could we massively split apart a file, 00:36:21.980 --> 00:36:24.500 the giant file and do something fast with it with Python? 00:36:24.880 --> 00:36:25.480 So these benchmarks, 00:36:25.720 --> 00:36:26.420 I'll also note, 00:36:26.780 --> 00:36:28.460 this was post reading it into memory 00:36:28.680 --> 00:36:30.300 and then doing the processing 00:36:30.320 --> 00:36:32.920 timing started. So it was all in memory and then processed. 00:36:33.160 --> 00:36:36.080 So there was no IO from disk involved 00:36:36.100 --> 00:36:36.600 in those benchmarks. 00:36:37.040 --> 00:36:42.380 Is there a similar library in Rust that maybe got compared to this? 00:36:42.680 --> 00:36:42.780 Because 00:36:43.140 --> 00:36:43.980 it seems pretty cool, 00:36:44.140 --> 00:36:48.860 but it seems like a natural intention would be to do something similar in Rust. 00:36:49.280 --> 00:36:49.400 Oh, 00:36:49.580 --> 00:36:50.780 our JSON might be Rust. 00:36:51.300 --> 00:36:52.800 That was in part of it. 00:36:53.600 --> 00:36:55.000 It was in the second part. 00:36:55.160 --> 00:36:56.660 It was in the second part. They'd benchmarked it. 00:36:56.820 --> 00:36:58.180 This one is C++. 00:36:59.100 --> 00:36:59.400 Right. 00:36:59.820 --> 00:36:59.960 Yeah. 00:37:00.220 --> 00:37:01.120 It's only a matter of days, 00:37:01.360 --> 00:37:01.420 right? 00:37:01.560 --> 00:37:02.360 Then this will be rust. 00:37:04.960 --> 00:37:05.080 Yeah. 00:37:05.080 --> 00:37:07.340 I'll just tell my agent to go port it right now. 00:37:07.640 --> 00:37:07.760 Yeah. 00:37:07.870 --> 00:37:12.080 Like I'm sure someone's wrapped, maybe, or Jason's wrapped around Surday or something, 00:37:12.300 --> 00:37:13.320 because I know today's a big one. 00:37:13.420 --> 00:37:14.700 Or Armin just wrote, 00:37:14.860 --> 00:37:17.000 Armin Roenacker just released a new, 00:37:17.240 --> 00:37:18.260 Jason, 00:37:18.440 --> 00:37:22.540 a new alternative to Surday for very specific reasons. 00:37:22.900 --> 00:37:23.040 Yeah. 00:37:23.420 --> 00:37:24.900 It'd be interesting to know what that would be. 00:37:24.900 --> 00:37:27.520 We're talking tens of gigabytes per second. 00:37:28.040 --> 00:37:28.120 Yeah. 00:37:28.200 --> 00:37:29.460 If you have those kinds of problems, 00:37:29.760 --> 00:37:31.100 I'd love to talk to you actually, 00:37:31.760 --> 00:37:32.060 because that sounds 00:37:32.220 --> 00:37:32.520 pretty cool. 00:37:32.960 --> 00:37:33.380 Very cool. 00:37:33.740 --> 00:37:37.020 I am pretty sure that ORJSON is Rust. 00:37:37.280 --> 00:37:38.020 I think it is. 00:37:38.040 --> 00:37:39.120 It's 55% Python, 00:37:39.280 --> 00:37:40.140 44% Rust. 00:37:40.300 --> 00:37:41.120 It could be written in shell. 00:37:41.320 --> 00:37:41.680 I don't know. 00:37:41.960 --> 00:37:43.780 The core, it could be like the tight loop could be in shell. 00:37:43.860 --> 00:37:44.420 It's 1% shell, 00:37:44.760 --> 00:37:45.380 but unlikely. 00:37:46.320 --> 00:37:51.140 I mean, that's how we used to distribute installers was the executable compressed archives that 00:37:51.220 --> 00:37:52.120 would look like a shell script. 00:37:52.160 --> 00:37:52.400 You just, 00:37:52.700 --> 00:37:52.840 you know, 00:37:53.220 --> 00:37:54.620 sh passed the script. 00:37:55.020 --> 00:37:55.500 That's funny. 00:37:55.660 --> 00:37:56.080 All right, well, 00:37:56.560 --> 00:37:59.620 let's get on to something totally different and talk about speed. 00:38:00.020 --> 00:38:01.000 Okay, speed sounds great. 00:38:01.420 --> 00:38:01.800 Let's do it, Michael. 00:38:02.020 --> 00:38:02.940 Let's do it, speed. 00:38:03.540 --> 00:38:07.620 So I want to come back to a former fellow Portlander, 00:38:07.830 --> 00:38:08.440 Miguel Grinberg, 00:38:08.660 --> 00:38:10.120 who's now, I believe, in Ireland, 00:38:10.340 --> 00:38:10.620 Scotland, 00:38:10.780 --> 00:38:11.420 one of those two places, 00:38:11.910 --> 00:38:12.780 unless he's moved again. 00:38:13.200 --> 00:38:18.580 But he has a yearly Python performance benchmark that he puts out. 00:38:18.770 --> 00:38:24.720 So I thought, hey, it's kind of 315 week if you count seven-day rolling window instead of a calendar week. 00:38:25.000 --> 00:38:28.760 So we just had Python 3.15 come out and he said it's time again for this. 00:38:28.790 --> 00:38:29.580 So I thought, you know what? 00:38:29.800 --> 00:38:31.560 Let's have a look and just talk real quick about it. 00:38:31.640 --> 00:38:32.340 This one will be pretty short. 00:38:32.740 --> 00:38:34.600 It's got some interesting things. 00:38:34.780 --> 00:38:37.080 It's not a super complex benchmark, 00:38:37.480 --> 00:38:38.520 so keep that in mind. 00:38:38.980 --> 00:38:41.420 But it does involve a whole bunch of versions, 00:38:41.620 --> 00:38:42.080 all the, 00:38:42.280 --> 00:38:44.780 hold on, even one of the unsupported 00:38:44.930 --> 00:38:45.800 older 3.10s. 00:38:45.800 --> 00:38:48.380 I don't know, 3.10 sounds still pretty new to me, but it's not. 00:38:48.580 --> 00:38:53.220 And PyPy, the JIT compiled version or JIT running version for JIT. 00:38:53.460 --> 00:38:54.740 and node and rust. 00:38:54.980 --> 00:38:55.640 It also says, 00:38:55.860 --> 00:38:57.380 well, let's take the regular CPython, 00:38:57.900 --> 00:38:59.060 the one with the experimental 00:38:59.300 --> 00:39:00.880 JIT and the one with free threading, 00:39:01.260 --> 00:39:02.400 and then do all those comparisons. 00:39:02.920 --> 00:39:04.160 So let's look through. 00:39:04.380 --> 00:39:04.480 It 00:39:04.580 --> 00:39:08.420 says, hey, we're just going to do a single threaded Fibonacci type of pretty big Fibonacci, 00:39:08.760 --> 00:39:08.940 depending 00:39:09.160 --> 00:39:10.260 on how you write it, I suppose, 00:39:10.600 --> 00:39:11.280 40 numbers. 00:39:11.760 --> 00:39:15.160 Fibonacci gets big fast, but you can memoize it and stuff 00:39:15.180 --> 00:39:16.220 and it's a lot quicker. 00:39:16.440 --> 00:39:18.460 But I imagine this is trying to make it hard, right? 00:39:18.760 --> 00:39:19.300 Well, yeah, because 00:39:19.220 --> 00:39:22.500 it takes 15 seconds to answer the question on Python 3.10. 00:39:22.730 --> 00:39:23.580 So there's a bunch of charts, 00:39:23.610 --> 00:39:24.560 you can go look at the article. 00:39:25.070 --> 00:39:28.920 But if you think about performance of even 3.10, which had some of 00:39:29.000 --> 00:39:31.000 that faster CPython work done to it, 00:39:31.220 --> 00:39:34.480 it's 44% as fast as today's Python. 00:39:34.860 --> 00:39:37.440 So we've gotten 56% faster. 00:39:37.740 --> 00:39:39.040 That's a significant improvement, 00:39:39.390 --> 00:39:41.240 even just for this sort of straightforward stuff. And there's a 00:39:41.340 --> 00:39:42.720 lot of things that are not factored in here. 00:39:42.940 --> 00:39:46.060 But then 3.11 was 1.7 times faster. 00:39:46.520 --> 00:39:47.040 You'll see that 00:39:47.180 --> 00:39:52.120 3.11 and I think 3.14 are the two that make really big jumps a lot of the times here. 00:39:52.500 --> 00:39:53.840 So we've got 3.11, 00:39:54.140 --> 00:39:55.920 which is 1.7 times faster. 00:39:56.280 --> 00:40:01.600 And then the jump from 14 to 15 is only 3% faster. 00:40:01.760 --> 00:40:03.420 So not dramatically faster, 00:40:03.800 --> 00:40:05.160 but that's for this particular benchmark. 00:40:05.360 --> 00:40:07.480 There's other areas where it would be dramatically faster. 00:40:07.900 --> 00:40:09.240 And PyPy312, 00:40:09.780 --> 00:40:11.220 5.5 times as fast. 00:40:11.620 --> 00:40:13.020 Node, not as fast, 00:40:13.400 --> 00:40:14.680 but almost five times faster. 00:40:14.900 --> 00:40:15.060 Rust, 00:40:15.320 --> 00:40:17.020 you know, just a little 77 times faster. 00:40:17.920 --> 00:40:18.200 I don't know. 00:40:19.020 --> 00:40:21.380 I'll pull up the graph for you all. 00:40:21.580 --> 00:40:23.160 Quick thoughts before I go on to the next one. 00:40:23.380 --> 00:40:23.800 Anyone have thoughts? 00:40:24.180 --> 00:40:28.420 It still amazes me how PyPy consistently is, 00:40:29.100 --> 00:40:30.240 at least with these kinds of workloads, 00:40:31.079 --> 00:40:32.720 showcases how performant it is. 00:40:33.340 --> 00:40:33.760 I know. 00:40:33.760 --> 00:40:35.740 I always think I should really check this out. 00:40:35.910 --> 00:40:37.600 But I think the challenge is 00:40:38.060 --> 00:40:40.000 then you want to go and import some library 00:40:40.200 --> 00:40:41.540 and it's incompatible with PyPy. 00:40:41.630 --> 00:40:43.880 And you're like, well, it goes from pretty, 00:40:44.080 --> 00:40:46.580 you know, five times faster to impossible and doesn't run. 00:40:46.650 --> 00:40:46.960 Like, oh, 00:40:47.320 --> 00:40:48.340 I'm not doing that, you know? 00:40:48.560 --> 00:40:52.780 I do remember announcements with the latest release of PyPy 00:40:53.100 --> 00:40:58.020 that they're trying to interface better with the CPython EBI 00:40:58.420 --> 00:41:02.020 so that external libraries can just like work in PyPy. 00:41:02.320 --> 00:41:04.660 So that's an exciting news, but still, it's so amazing. 00:41:04.960 --> 00:41:08.860 What surprises me is Node still being faster than Python. 00:41:09.330 --> 00:41:10.460 And I really, 00:41:10.700 --> 00:41:12.880 I have respect for TypeScript and JavaScript, 00:41:13.280 --> 00:41:14.780 But still, it still amazes me. 00:41:14.860 --> 00:41:16.980 It just makes me realize how much engineering 00:41:17.200 --> 00:41:19.900 went to the V8 engine. 00:41:20.100 --> 00:41:20.180 Yeah. 00:41:20.200 --> 00:41:21.140 It's amazing. 00:41:22.040 --> 00:41:24.560 It's decades of fully funded, 00:41:24.900 --> 00:41:25.920 full team support 00:41:26.550 --> 00:41:27.320 versus Python, 00:41:27.700 --> 00:41:30.200 which as Michael alluded to, right? 00:41:30.800 --> 00:41:33.000 Just started maybe five, 00:41:33.160 --> 00:41:33.780 six years ago, 00:41:34.080 --> 00:41:35.300 getting external funded, 00:41:35.700 --> 00:41:37.740 let's make it faster kind of performance stuff, 00:41:38.000 --> 00:41:38.120 right? 00:41:38.200 --> 00:41:39.980 And having people like that and stuff 00:41:40.100 --> 00:41:42.100 start to bring stuff out from Cinder and all that, right? 00:41:42.280 --> 00:41:42.360 Yeah. 00:41:42.980 --> 00:41:43.680 It's, it's, 00:41:43.900 --> 00:41:44.780 it's timelines. 00:41:45.000 --> 00:41:45.600 The other thing though, 00:41:46.060 --> 00:41:49.900 is Michael's not shown us what the JIT numbers are with the experimental 00:41:50.140 --> 00:41:50.240 JIT. 00:41:50.560 --> 00:41:51.520 So we'll also keep going. 00:41:51.620 --> 00:41:52.040 Let's keep going. 00:41:52.540 --> 00:41:52.920 Let's see. 00:41:53.320 --> 00:41:54.740 So yeah. 00:41:54.740 --> 00:41:55.440 So we've got the, 00:41:55.840 --> 00:41:57.860 the next graph is we've got the standard, 00:41:58.060 --> 00:41:59.160 which is just what you get when 00:41:59.160 --> 00:41:59.680 you write Python. 00:41:59.840 --> 00:42:02.520 Then you've got the experimental JIT, which you can turn on and the free threading. 00:42:02.820 --> 00:42:03.800 And this is single threaded. 00:42:04.020 --> 00:42:06.520 So the free threading will not get better, 00:42:06.960 --> 00:42:07.760 but if it doesn't get worse, 00:42:08.160 --> 00:42:08.580 that's actually 00:42:08.580 --> 00:42:09.400 a huge improvement. 00:42:09.800 --> 00:42:11.400 And I think that's actually the biggest takeaway 00:42:11.980 --> 00:42:13.980 from this one is for this Fibonacci thing. 00:42:14.460 --> 00:42:15.740 It really doesn't that much of a difference, 00:42:15.940 --> 00:42:17.340 but Python 3.13, 00:42:17.600 --> 00:42:18.640 free threading was, 00:42:18.780 --> 00:42:19.800 I don't know, 20% slower. 00:42:20.220 --> 00:42:23.320 And now free threading is actually just on par 00:42:23.700 --> 00:42:25.020 with regular CPython. 00:42:25.040 --> 00:42:25.320 That's great. 00:42:26.260 --> 00:42:27.460 That's actually really good. 00:42:27.580 --> 00:42:28.720 And the JIT, as Brett points out, 00:42:28.740 --> 00:42:29.720 is like 20% faster. 00:42:30.020 --> 00:42:31.340 Yeah, I think Nick brought it up 00:42:31.340 --> 00:42:32.120 with like the compatibility. 00:42:32.500 --> 00:42:34.720 Like PyPy had problems with importing 00:42:35.120 --> 00:42:36.300 some package you want to use. 00:42:36.740 --> 00:42:38.080 Now, even the folks over at EVE Online 00:42:38.200 --> 00:42:39.080 are like, we're fast enough. 00:42:39.320 --> 00:42:39.520 finally, 00:42:39.960 --> 00:42:41.540 let's just use the box Python. 00:42:42.380 --> 00:42:43.420 And now we've got all the benefits 00:42:43.580 --> 00:42:45.580 of a giant library of ecosystem stuff. 00:42:46.380 --> 00:42:46.700 Exactly. 00:42:46.860 --> 00:42:47.180 Guess what? 00:42:47.640 --> 00:42:48.220 If we upgrade, 00:42:48.620 --> 00:42:50.500 all right, let's go a little farther. 00:42:50.600 --> 00:42:51.300 So the next one is the bubble sort. 00:42:51.300 --> 00:42:52.380 That was always a sketchy path. 00:42:52.460 --> 00:42:55.020 Like they were on such an experimental version of Python 00:42:55.500 --> 00:42:56.400 with their whole production. 00:42:56.640 --> 00:42:57.220 That is impressive. 00:42:57.819 --> 00:42:58.140 Yeah. 00:42:58.420 --> 00:43:00.700 They stuck it out with Stackless for a long time. 00:43:00.960 --> 00:43:01.420 Long time. 00:43:01.780 --> 00:43:02.380 Long time. 00:43:02.720 --> 00:43:03.320 Until this year. 00:43:03.680 --> 00:43:04.900 Yeah, I just had them on Talk Python. 00:43:05.140 --> 00:43:05.660 So yeah. 00:43:06.060 --> 00:43:06.200 All right. 00:43:06.220 --> 00:43:08.720 The next one is the Fibonacci with four threads. 00:43:08.980 --> 00:43:10.580 I think this gets pretty interesting 00:43:10.920 --> 00:43:13.680 and it shows some really interesting improvements. 00:43:14.220 --> 00:43:16.200 So we've got maybe if you run four threads 00:43:16.410 --> 00:43:17.640 to answer the Fibonacci question 00:43:18.080 --> 00:43:19.920 compared to the others like JIT and standard, 00:43:20.200 --> 00:43:21.280 it's, I don't know, 00:43:21.620 --> 00:43:21.840 20, 00:43:22.080 --> 00:43:23.940 25% time it takes. 00:43:24.040 --> 00:43:24.880 So that's four times faster, 00:43:25.280 --> 00:43:26.740 which if you add four threads 00:43:26.920 --> 00:43:27.960 and you get four times faster, 00:43:28.300 --> 00:43:28.840 wow, that's good. 00:43:28.980 --> 00:43:29.380 That's pretty good. 00:43:29.660 --> 00:43:30.900 People who haven't done multi-threading before, 00:43:31.130 --> 00:43:33.680 you rarely get as many times improvement 00:43:33.900 --> 00:43:35.660 as you have CPUs you've added to it, right? 00:43:35.800 --> 00:43:37.300 There's a lot of coordination and other things. 00:43:37.740 --> 00:43:38.620 It is usually not linear. 00:43:38.700 --> 00:43:39.060 So, 00:43:39.380 --> 00:43:39.580 yeah, 00:43:40.100 --> 00:43:41.160 this is actually pretty impressive. 00:43:42.280 --> 00:43:52.140 Just shows like for the longest time is actually kind of amazing that the language like Python that is compared to like C or Java is not as fast, 00:43:52.300 --> 00:43:56.380 was able to like garner the attention and love of so many people. 00:43:56.940 --> 00:43:58.860 And it was good enough for so many things. 00:43:59.420 --> 00:44:03.880 And now it's just like getting better and better now. It's actually quite remarkable. 00:44:04.220 --> 00:44:04.640 Absolutely. 00:44:05.200 --> 00:44:10.820 And also the JIT from 3.14 to 3.15 has gotten quite a bit better as well. 00:44:11.060 --> 00:44:11.540 So that's 00:44:11.610 --> 00:44:12.000 pretty interesting. 00:44:12.580 --> 00:44:13.100 What else, 00:44:13.260 --> 00:44:14.980 what else do we got to take away from here? I don't know. 00:44:15.240 --> 00:44:15.940 I would say. 00:44:16.210 --> 00:44:19.720 What was it, Brett, who said like Python is like the second best choice for everything. I think 00:44:19.900 --> 00:44:20.580 that's not true anymore. 00:44:20.960 --> 00:44:21.160 Yeah. 00:44:21.410 --> 00:44:22.020 I don't think it's, 00:44:22.210 --> 00:44:22.640 I don't know. 00:44:23.040 --> 00:44:24.020 There's certainly. 00:44:24.120 --> 00:44:28.580 Well, your AI will probably always choose Python for certain projects. 00:44:28.920 --> 00:44:30.100 So it's really the first 00:44:30.280 --> 00:44:30.700 choice for everything. 00:44:30.740 --> 00:44:31.380 It's the first choice. 00:44:31.550 --> 00:44:31.680 Yeah. 00:44:32.060 --> 00:44:33.860 Yeah, it's head-to-head on JavaScript, 00:44:34.050 --> 00:44:35.440 or if you ask it to do a web thing, 00:44:35.490 --> 00:44:37.380 it probably does JavaScript if you don't guide it. 00:44:37.610 --> 00:44:39.220 But if you ask it to do any sort of data thing, 00:44:39.310 --> 00:44:40.140 it's definitely doing Python. 00:44:40.620 --> 00:44:42.720 Yeah, playwright is Python. 00:44:43.919 --> 00:44:47.300 And to be fair to PyPy, right, 00:44:47.450 --> 00:44:49.260 they also have to keep up with the language, 00:44:49.570 --> 00:44:49.660 right, 00:44:49.860 --> 00:44:50.520 intended to change, 00:44:50.720 --> 00:44:53.620 hence why they're not on 3.14 or 3.15, 00:44:53.840 --> 00:44:57.300 while the built-in JIT just gets to piggyback on all the other work. 00:44:58.059 --> 00:44:58.780 Yeah, 100%. 00:44:58.800 --> 00:45:01.620 JIT comparison is not a direct one-to-one that way because of that. 00:45:01.820 --> 00:45:06.440 I also want to give a quick shout out to some of the new things we got that are utterly untested here. 00:45:06.780 --> 00:45:08.000 This is, I wrote one function, 00:45:08.240 --> 00:45:09.860 and in that one function, 00:45:10.120 --> 00:45:12.700 I ran it a bunch, sometimes with threads, sometimes with JIT, 00:45:12.880 --> 00:45:13.280 sometimes without, 00:45:13.540 --> 00:45:14.460 and it did a thing, 00:45:14.660 --> 00:45:14.780 right? 00:45:15.100 --> 00:45:24.340 What we don't see is we don't see the effect of lazy imports for really large applications or things that need super quick startup time, and then they're gone like a CLI tool. 00:45:24.720 --> 00:45:31.500 We don't see the specialized and adaptive interpreter or even the JIT really get to kick in very much because it's one or two functions. 00:45:31.740 --> 00:45:33.560 if you had 10,000 functions 00:45:34.040 --> 00:45:35.680 and it was a big, interesting application, 00:45:36.120 --> 00:45:37.960 those could also really start to make a difference. 00:45:38.860 --> 00:45:39.880 So the PSA, 00:45:40.060 --> 00:45:40.200 guys, 00:45:40.760 --> 00:45:42.740 the PSA is just upgrade your Python 00:45:43.000 --> 00:45:45.560 and you get performance for lunch. 00:45:45.560 --> 00:45:45.780 Oh, yeah. 00:45:46.180 --> 00:45:46.920 You will save money. 00:45:47.340 --> 00:45:47.540 Yeah. 00:45:47.920 --> 00:45:49.200 This is like legitimate free lunch. 00:45:49.560 --> 00:45:52.160 Yeah, no, I've had this discussion with people 00:45:52.420 --> 00:45:54.540 like when their managers don't want to punt 00:45:54.680 --> 00:45:55.700 and put off upgrading, 00:45:56.020 --> 00:45:57.340 the way to sell it to management is 00:45:57.680 --> 00:45:58.460 you will save money. 00:45:58.780 --> 00:45:59.700 Python will be faster. 00:45:59.800 --> 00:46:01.640 we'll use less memory very likely as well. 00:46:02.100 --> 00:46:03.720 And thus we will do more with less. 00:46:04.080 --> 00:46:05.260 So either we can scale higher 00:46:05.740 --> 00:46:07.420 or we can do the same things with less machines. 00:46:07.980 --> 00:46:09.700 So why not let us upgrade, 00:46:10.000 --> 00:46:10.060 please? 00:46:10.520 --> 00:46:11.400 So it's less of a pain 00:46:11.520 --> 00:46:13.740 because we only get to upgrade every two or four or five years. 00:46:14.040 --> 00:46:15.500 And it's a big hassle. 00:46:15.580 --> 00:46:16.920 It's like, no, upgrade continuously 00:46:16.960 --> 00:46:17.800 and you'll keep getting these ones. 00:46:18.100 --> 00:46:18.340 Well, 00:46:18.560 --> 00:46:21.060 and your libraries get far quicker too. 00:46:21.360 --> 00:46:24.040 Like Napari using lazy imports 00:46:24.240 --> 00:46:26.700 is far better than on a version 00:46:26.900 --> 00:46:28.520 that can't use lazy imports. 00:46:28.620 --> 00:46:28.940 So, 00:46:29.590 --> 00:46:29.720 yeah. 00:46:30.200 --> 00:46:36.100 And to quickly touch on what Michael said about the difficulty of judging performance on these micro benchmarks. 00:46:36.630 --> 00:46:36.760 Right. 00:46:36.810 --> 00:46:41.880 This is one reason PyPerformance that is used for benchmarking CPython and Python in general, 00:46:42.320 --> 00:46:53.620 tries to do benchmarks that are pulled from real world code and tries to be somewhat representative of actual work close to an extent while not taking hours and hours and hours to run. 00:46:53.960 --> 00:46:58.560 But another point that actually came up during the development 3.15 was the garbage collector, 00:46:59.020 --> 00:46:59.100 right? 00:46:59.270 --> 00:47:01.580 There was the incremental garbage collector put in. 00:47:01.940 --> 00:47:04.820 And we found out after people started to test with it, 00:47:05.140 --> 00:47:07.500 that it actually didn't work in large, 00:47:07.740 --> 00:47:11.280 certain key workloads and actually led to larger garbage collector pauses. 00:47:11.480 --> 00:47:14.740 Because I don't know how many people know this, but Python kind of has two garbage collectors, 00:47:14.960 --> 00:47:15.020 right? 00:47:15.100 --> 00:47:16.440 There's reference counting for most stuff, 00:47:16.700 --> 00:47:17.940 but then we have a garbage collector look for 00:47:18.060 --> 00:47:20.160 cycles because reference counting can't handle cycles. 00:47:20.700 --> 00:47:22.380 And we've had a generational one for a long time. 00:47:22.740 --> 00:47:26.840 and incremental will get to less pauses and hopefully better performance but we found out 00:47:26.840 --> 00:47:32.840 the one we did actually did make things better for most people but for certain people the the 00:47:33.160 --> 00:47:37.840 degenerate cases like the really bad edge case ones it was really slow and bad so we actually had to 00:47:37.900 --> 00:47:43.080 roll it back and um and mark shanning for instance currently has a proposal on how to kind of fix 00:47:43.200 --> 00:47:46.880 that but there's another point that it's really hard to measure unless people actually do this 00:47:46.940 --> 00:47:52.460 which is why we will always say as a psa from the python dev team please test the alphas and the 00:47:52.480 --> 00:47:52.820 And please, 00:47:53.130 --> 00:47:56.880 please at least test the release candidates on actual workloads to let us know 00:47:56.970 --> 00:47:58.080 where things may have gone wrong. 00:47:58.320 --> 00:48:01.360 Because if someone hadn't tested that and told us ahead of time, we would have shipped 00:48:01.900 --> 00:48:03.300 and we actually did in 3.14, 00:48:04.220 --> 00:48:06.100 a garbage collector that needed to get rolled back. 00:48:06.260 --> 00:48:08.140 So it just didn't hit people really badly. 00:48:08.420 --> 00:48:08.660 Absolutely. 00:48:09.000 --> 00:48:11.960 All right, let's have Carol close it out. 00:48:12.980 --> 00:48:14.340 And our favorite topic, 00:48:15.160 --> 00:48:17.900 artificial intelligence and tools around it. 00:48:18.980 --> 00:48:19.020 So, 00:48:19.740 --> 00:48:19.900 you know, 00:48:20.080 --> 00:48:24.840 after being in the scientific ecosystem and data science and Jupyter ecosystem, 00:48:25.640 --> 00:48:25.820 you know, 00:48:26.000 --> 00:48:28.400 I've watched the evolution of, 00:48:28.560 --> 00:48:29.700 you know, machine learning, 00:48:30.100 --> 00:48:30.560 deep learning. 00:48:31.110 --> 00:48:32.700 We call AI now. 00:48:33.840 --> 00:48:40.540 And one of the things Paul Everitt and I have been sort of talking over the past year about, 00:48:41.040 --> 00:48:41.120 hey, 00:48:41.480 --> 00:48:46.300 you know, so much of it is built on top of the work that we've done with Python, 00:48:46.760 --> 00:48:51.620 that it would be great to have a community benefit from it, 00:48:52.060 --> 00:48:54.000 not just large corporations 00:48:55.500 --> 00:48:58.360 steering the future of these tools. 00:48:58.880 --> 00:49:00.560 And so back in March, 00:49:01.860 --> 00:49:03.720 Georgie and Paul and I sat down and we 00:49:03.880 --> 00:49:09.400 started talking about, hey, how could we use more local capacity? 00:49:09.690 --> 00:49:12.140 How could we improve privacy, 00:49:13.180 --> 00:49:13.640 security, 00:49:14.420 --> 00:49:18.300 not leaking your data out to other services. 00:49:19.040 --> 00:49:23.140 And Paul did a keynote back at 00:49:23.380 --> 00:49:27.140 EuroPython this year and introduced what we're calling Satyrn. 00:49:27.780 --> 00:49:28.760 And really, 00:49:28.970 --> 00:49:31.360 Satyrn is looking at 00:49:31.760 --> 00:49:33.520 doing AI your way, 00:49:33.980 --> 00:49:37.820 which doesn't exclude the use of foundational models, 00:49:38.460 --> 00:49:39.960 but really emphasizes 00:49:39.960 --> 00:49:47.800 is that there's a spectrum of ways that you can use large language models and agents to suit your 00:49:48.120 --> 00:49:50.700 particular workflow because one size does not fit all. 00:49:51.020 --> 00:49:54.200 And from that EuroPython keynote, 00:49:54.390 --> 00:49:54.760 he had a 00:49:54.880 --> 00:49:55.500 call to action, 00:49:56.720 --> 00:49:58.920 mostly for the Europe community, 00:49:59.380 --> 00:50:02.860 but it has expanded out from that to kind of create 00:50:02.940 --> 00:50:10.500 a collective of individuals that are passionate about making it Python first and trying to make 00:50:10.780 --> 00:50:12.600 transparent and trusted AI. 00:50:13.600 --> 00:50:18.760 So I've kind of helped with the effort to kind of get the scaffolding in 00:50:19.000 --> 00:50:19.280 place. 00:50:19.600 --> 00:50:26.540 And I'm happy to say that when 3.15 got released, there was also a model that up on 00:50:26.600 --> 00:50:35.280 hugging face that the Satyrn team has trained on 315 that can run locally as well as on other 00:50:35.540 --> 00:50:35.640 things. 00:50:36.140 --> 00:50:41.120 And I think one of the key things is while we look at local first, 00:50:41.650 --> 00:50:43.100 we're also looking at 00:50:43.480 --> 00:50:43.940 trust first. 00:50:44.500 --> 00:50:44.720 And, 00:50:45.260 --> 00:50:52.440 you know, I could go back to my icon keynote back in 2023 when ChatGPT was still 00:50:52.340 --> 00:50:52.700 in beta. 00:50:53.180 --> 00:50:55.880 And one of the things that bothered me was, 00:50:56.170 --> 00:50:59.260 do I trust this large language model? 00:50:59.570 --> 00:50:59.820 Do I 00:51:00.000 --> 00:51:01.080 trust how it was trained? 00:51:01.310 --> 00:51:02.460 Do I trust it, 00:51:02.800 --> 00:51:03.060 you know, 00:51:03.300 --> 00:51:04.340 to be unbiased? 00:51:04.720 --> 00:51:07.100 And while there's a lot of value 00:51:07.340 --> 00:51:07.560 in it, 00:51:07.750 --> 00:51:11.620 you still have to verify that it's doing what you think it's doing. 00:51:12.100 --> 00:51:13.620 And so I think having 00:51:14.040 --> 00:51:15.920 more options where, 00:51:16.600 --> 00:51:18.580 you know, you don't always need a sledgehammer. 00:51:18.820 --> 00:51:20.340 Sometimes a hammer is good enough. 00:51:20.680 --> 00:51:21.880 And in this case, 00:51:22.120 --> 00:51:25.860 sometimes you can get a lot of good stuff the last six months out of local models. 00:51:26.340 --> 00:51:26.640 And, 00:51:27.540 --> 00:51:27.780 you know, 00:51:28.050 --> 00:51:35.460 the other key component is one of the keys that has made Python so successful is the language, 00:51:36.090 --> 00:51:37.560 the ecosystem of libraries, 00:51:38.060 --> 00:51:44.300 but also the huge commitment the community has about sharing learning materials, 00:51:44.860 --> 00:51:45.900 educating each other, 00:51:46.760 --> 00:51:47.580 spreading the knowledge. 00:51:48.000 --> 00:51:54.040 And if all Satyrn did was provide these education and learning opportunities, 00:51:54.630 --> 00:51:58.580 which is a big part of what we're taking from Jupyter, 00:51:58.920 --> 00:52:01.800 Jupyter really helped accelerate learning, 00:52:02.060 --> 00:52:03.840 sharing learning materials with folks. 00:52:04.120 --> 00:52:04.540 And, 00:52:05.030 --> 00:52:05.380 you know, 00:52:05.760 --> 00:52:09.300 there's a team around sharing educational content, 00:52:09.710 --> 00:52:11.380 both created by the team, 00:52:11.760 --> 00:52:15.260 as well as highlighting things that are available within the community. 00:52:15.470 --> 00:52:17.120 So it's still in the early stages, 00:52:17.620 --> 00:52:20.780 but I think there's good energy behind it. 00:52:21.060 --> 00:52:24.360 And I would love to see it grow more and more. 00:52:24.760 --> 00:52:37.200 And the reason we kind of started in Europe was we felt like the continent as a whole was more focused on security, 00:52:37.580 --> 00:52:37.740 privacy, 00:52:39.380 --> 00:52:41.140 and the goal impact of AI. 00:52:41.960 --> 00:52:43.140 Well, and being a European, 00:52:43.540 --> 00:52:50.720 how much do you want to say our entire AI future hangs on large American tech companies or Chinese companies? 00:52:51.420 --> 00:52:53.500 Maybe we should have some say in this, something like that, right? 00:52:53.640 --> 00:52:54.020 Yeah, 00:52:54.420 --> 00:52:54.860 well, I think, 00:52:55.040 --> 00:52:57.000 I mean, everybody should have a say in it. 00:52:57.200 --> 00:52:59.180 And I think, you know, 00:52:59.600 --> 00:53:01.560 some of the large companies, 00:53:01.940 --> 00:53:02.500 let's face it, 00:53:02.680 --> 00:53:08.440 they're beholden to their shareholders or investors to maximize profits, 00:53:08.960 --> 00:53:16.280 which doesn't always align with maximizing responsibility to the community and ethics and other things. 00:53:16.400 --> 00:53:16.720 So, 00:53:17.660 --> 00:53:20.160 you know, I think alternatives are always good. 00:53:20.380 --> 00:53:22.820 It helps push ecosystems forward. 00:53:23.140 --> 00:53:28.820 There's a lot of great stuff happening in bioscience and medical applications, 00:53:29.520 --> 00:53:34.740 particularly with rare diseases that the foundational models have been hugely helpful for. 00:53:35.220 --> 00:53:35.940 And so, 00:53:36.480 --> 00:53:38.140 you know, it's not one or the other. 00:53:39.000 --> 00:53:41.720 There's a spectrum of use cases. 00:53:42.300 --> 00:53:44.220 There should be a spectrum of tools and approaches. 00:53:44.840 --> 00:53:48.480 And as we navigate the understanding of it, 00:53:48.640 --> 00:53:50.760 sharing that becomes more and more important, 00:53:51.160 --> 00:53:52.440 which Python bites. 00:53:52.980 --> 00:53:55.740 And Python to me for sharing this on a regular basis. 00:53:56.280 --> 00:53:58.020 I love the fact there's two models. 00:53:58.120 --> 00:54:06.320 You've got the smaller Ornith model and then the other one based on that Quinn 3827B powerhouse of a local model. 00:54:06.580 --> 00:54:07.660 I can't wait to try both of them. 00:54:08.020 --> 00:54:08.160 Yeah. 00:54:08.320 --> 00:54:09.520 I mean, it's pretty interesting. 00:54:09.680 --> 00:54:13.880 And there's teams that, if you join the Discord, 00:54:14.260 --> 00:54:18.320 that are working on these, how do you test and train? 00:54:19.580 --> 00:54:26.560 And so we're trying to be open by default and the information and the workflows. 00:54:26.580 --> 00:54:26.900 So, 00:54:27.420 --> 00:54:28.540 you know, definitely reach out. 00:54:28.920 --> 00:54:31.780 And yeah, it's got some good energy behind it. 00:54:32.020 --> 00:54:35.140 Are there other languages who are interested in participating? 00:54:36.360 --> 00:54:37.120 I don't know. 00:54:37.450 --> 00:54:38.660 I mean, there might be. 00:54:39.010 --> 00:54:41.640 I mean, I could see maybe Rust having some interest, 00:54:42.140 --> 00:54:46.920 but for now we've kind of scoped it to Python and, 00:54:47.840 --> 00:54:48.020 you know, 00:54:48.380 --> 00:54:50.040 trying to train something specifically 00:54:50.060 --> 00:54:56.740 on Python 3.15 and the use cases that, you know, are new with 3.15 was something that 00:54:56.840 --> 00:54:57.680 was really interesting. 00:54:58.030 --> 00:54:59.100 But that's a really good question. 00:54:59.500 --> 00:55:01.500 I wonder where my R fellas, 00:55:02.080 --> 00:55:03.760 like the R language are, 00:55:03.920 --> 00:55:05.960 that would be a good language to add under 00:55:06.460 --> 00:55:06.840 the Satyrn. 00:55:06.840 --> 00:55:07.220 You know, 00:55:07.400 --> 00:55:07.940 yeah. 00:55:08.630 --> 00:55:09.000 I mean, 00:55:09.320 --> 00:55:09.700 in fact, 00:55:10.600 --> 00:55:12.920 Posit is having sort of a hackathon, 00:55:13.400 --> 00:55:15.460 I think, in November that I'm going 00:55:15.640 --> 00:55:15.760 to. 00:55:15.810 --> 00:55:19.740 So that might be a great thing to kind of kick around the tires on. 00:55:19.940 --> 00:55:20.140 Yeah, 00:55:20.510 --> 00:55:21.640 I want you to say, personally, 00:55:21.860 --> 00:55:26.620 I've always had trouble running local AI models, 00:55:27.270 --> 00:55:34.700 mainly because sometimes it was always hard to know which models could actually fit into my consumer-grade hardware. 00:55:35.380 --> 00:55:38.780 And so navigating that has always been not the fun part. 00:55:39.050 --> 00:55:45.300 So I'm really looking forward to participating in this because I believe you guys will probably do a better job of discussing. 00:55:45.780 --> 00:55:47.780 These are the kind of hardware you can run. 00:55:48.140 --> 00:55:50.360 This is the kind of models that can run on that hardware. 00:55:50.920 --> 00:55:52.040 This is some use cases, 00:55:52.260 --> 00:55:55.120 something as simple as categorizing or, 00:55:55.380 --> 00:55:55.960 you know, 00:55:56.360 --> 00:56:00.620 categorizing like your files to like helping you run a small, 00:56:01.120 --> 00:56:02.640 like automated business or something. 00:56:02.820 --> 00:56:04.820 That'll be cool to see what you guys do. 00:56:05.220 --> 00:56:09.480 Well, and looping back to the start where you were talking about Marimo, 00:56:09.840 --> 00:56:16.480 we've actually used some of the Marimo hub and the GPU access there to do some of this training. 00:56:17.320 --> 00:56:18.060 So, 00:56:18.260 --> 00:56:21.140 you know, so it doesn't have to be just your local laptop. 00:56:21.380 --> 00:56:21.980 It can be, 00:56:22.440 --> 00:56:22.640 you know, 00:56:23.000 --> 00:56:24.580 something that is a little beefier. 00:56:25.380 --> 00:56:26.900 Yeah, and looking at the size of the models, 00:56:27.120 --> 00:56:28.100 they're not ginormous. 00:56:28.200 --> 00:56:28.300 No. 00:56:28.360 --> 00:56:29.760 You know, they're 10 billion parameters, 00:56:30.200 --> 00:56:30.940 28 billion parameters. 00:56:31.240 --> 00:56:34.060 These are plausible on decent local machines. 00:56:34.540 --> 00:56:34.740 Yeah. 00:56:35.060 --> 00:56:38.980 I mean, they definitely work on my GPU Linux box. 00:56:40.100 --> 00:56:42.800 You can get it to work on a Mac as well. 00:56:42.880 --> 00:56:44.780 It's a little bit, you know, 00:56:45.060 --> 00:56:45.580 a little bit harder. 00:56:45.980 --> 00:56:47.240 Got to submit it to the app store. 00:56:47.280 --> 00:56:52.400 see if they approve it yeah no i think you just have to get the mlx you know version of the training 00:56:52.540 --> 00:56:56.980 done so you can accelerate it on the on the max yeah luckily i've got a Linux box here too so let's 00:56:57.000 --> 00:57:01.860 throw it on there cool let us know what you find out i just realized we're running out of planets 00:57:02.100 --> 00:57:10.320 to rename with a y i did i was not the namer uh that would be paul no i'm just glad you said 00:57:10.960 --> 00:57:16.640 after jupiter yeah i mean uranus is like one of the last ones with the u in it and yeah yeah 00:57:16.700 --> 00:57:21.380 number out yeah yeah yeah yeah good point brett it's brett it's gonna be okay we're going out to 00:57:21.380 --> 00:57:29.860 the orc cloud yeah fine yeah s2 s7 y5 i know you thought that was a u but no it's the y one 00:57:30.360 --> 00:57:35.460 it'll be good it'll be good all right well this is super interesting and the naming i guess 00:57:35.720 --> 00:57:42.920 probably came off of jupiter uh that i think when paul was looking at it yeah i mean definitely some 00:57:42.940 --> 00:57:46.960 of the things that Jupyter did in terms of accelerating the community. 00:57:47.780 --> 00:57:48.700 He wanted, 00:57:49.360 --> 00:57:49.540 you know, 00:57:49.760 --> 00:57:53.720 that ability to share knowledge as well as build things. 00:57:54.220 --> 00:57:54.600 And, 00:57:54.980 --> 00:57:55.900 you know, 00:57:56.320 --> 00:57:58.020 the Y just seemed to make 00:57:58.120 --> 00:58:01.620 sense. Plus there are other Satyrns out there, Satyrn AIs and whatnot. 00:58:01.830 --> 00:58:03.580 So why not add the Y? 00:58:03.700 --> 00:58:03.800 Yeah. 00:58:03.970 --> 00:58:05.320 Why not? Why not? Cool. 00:58:05.560 --> 00:58:07.460 Well, we're just about out of time, 00:58:07.960 --> 00:58:09.260 but I've got a couple of 00:58:09.280 --> 00:58:14.200 extras i'm gonna trim mine back but i know brett you want to there's some little release or 00:58:14.290 --> 00:58:19.900 something you wanted to talk about what is this i think you already talked about it yeah 315.0 00:58:20.100 --> 00:58:26.300 finals gone out so 310 also hit eol about a week prior i think two weeks prior i can't remember 00:58:26.720 --> 00:58:33.560 uh but yeah after the unfortunate new tradition of having a third rc happened um finals out it's 00:58:33.580 --> 00:58:41.240 good it's done uh obviously we're already working on 3.15.1 which should hopefully come out typically 00:58:41.540 --> 00:58:45.580 historically we do it in november um but then we'll be on the usual two-month cadence for 00:58:45.920 --> 00:58:51.080 bugfish releases but they're 15s out um a lot of great stuff in it as i know has been covered here 00:58:51.380 --> 00:58:56.380 and download however you download your latest python i'm gonna give a huge shout out to trey 00:58:56.400 --> 00:59:02.020 hunter who has some really good what's new in 315 that, 00:59:02.360 --> 00:59:02.560 you know, 00:59:02.940 --> 00:59:04.920 web developers and folks that 00:59:05.100 --> 00:59:10.680 are maybe not as focused on some of the scientific benefits for some of the performance stuff. 00:59:11.100 --> 00:59:12.360 It's really well done. 00:59:12.540 --> 00:59:14.800 And Barry Warsaw's what's newt, 00:59:15.000 --> 00:59:16.140 if you want to do a dungeon 00:59:16.420 --> 00:59:17.160 crawler version of it. 00:59:17.320 --> 00:59:18.700 That was so awesome. 00:59:19.100 --> 00:59:20.460 I really enjoyed that. 00:59:20.600 --> 00:59:21.760 That is fun. It was really fun. 00:59:22.220 --> 00:59:22.380 Agreed. 00:59:22.520 --> 00:59:22.620 Yeah. 00:59:22.800 --> 00:59:23.200 I was thinking, 00:59:23.640 --> 00:59:26.220 could I make a visual version to overlay on that? I don't know. 00:59:26.340 --> 00:59:26.680 We'll see. 00:59:26.980 --> 00:59:28.120 I will not be doing that. 00:59:28.560 --> 00:59:30.780 But I do want to give a quick shout out to this as an extra. 00:59:30.900 --> 00:59:31.800 Have you all seen this? 00:59:32.820 --> 00:59:33.280 This is insane. 00:59:33.900 --> 00:59:35.740 This craft apps thing. 00:59:35.960 --> 00:59:37.800 I think the guy's name is Ben. 00:59:38.040 --> 00:59:38.820 I thought it hit the news. 00:59:39.160 --> 00:59:39.260 Yeah. 00:59:39.660 --> 00:59:39.700 Yeah. 00:59:40.060 --> 00:59:43.960 This guy had a big falling out with Adobe over the subscription stuff. 00:59:44.140 --> 00:59:44.220 Surprise. 00:59:44.880 --> 00:59:45.440 And said, well, 00:59:45.620 --> 00:59:46.320 you know what I'm going to do? 00:59:46.600 --> 00:59:47.880 I'm going to recreate Photoshop. 00:59:48.920 --> 00:59:50.700 And I don't even know the apps that well. 00:59:50.760 --> 00:59:51.020 Let me see. 00:59:51.340 --> 00:59:52.120 I'm going to recreate Photoshop. 00:59:52.740 --> 00:59:55.120 And I'm going to create Illustrator. 00:59:55.840 --> 00:59:57.760 And a bunch of others, 00:59:58.960 --> 01:00:01.340 Lightroom, I don't know, a bunch of these, Acrobat, 01:00:01.710 --> 01:00:02.460 and recreated them. 01:00:02.890 --> 01:00:03.600 A whole creative suite. 01:00:04.160 --> 01:00:04.440 Almost. 01:00:04.780 --> 01:00:05.520 It's missing Audition. 01:00:05.650 --> 01:00:07.340 I can't wait for him to do Audition. 01:00:07.460 --> 01:00:07.820 Let's go. 01:00:08.440 --> 01:00:11.940 Because I pay a ton of money to edit this podcast and other stuff with Audition. 01:00:12.160 --> 01:00:13.100 But yesterday, 01:00:13.410 --> 01:00:15.780 when I was like, I'm going to throw this up here as a thing they're talking about. 01:00:16.180 --> 01:00:16.720 There were 12. 01:00:17.100 --> 01:00:17.760 Today is 13. 01:00:18.739 --> 01:00:21.520 And I got to check real quick and see what the 13th is. 01:00:21.570 --> 01:00:22.980 But it has Microsoft Word, 01:00:23.280 --> 01:00:23.740 PowerPoint, 01:00:24.460 --> 01:00:24.860 and Excel. 01:00:25.540 --> 01:00:27.040 I got the little deck craft thing. 01:00:27.420 --> 01:00:28.300 I know it's not the same, 01:00:28.640 --> 01:00:30.400 but I opened up my talk that I gave 01:00:30.530 --> 01:00:32.580 with all the animations and gradients and stuff 01:00:32.960 --> 01:00:35.900 for PiBay and it presented it just fine. 01:00:36.130 --> 01:00:36.540 I don't know. 01:00:36.810 --> 01:00:38.040 I don't know that it edits as well. 01:00:38.180 --> 01:00:39.040 But anyway, I just, 01:00:39.220 --> 01:00:39.900 I think, 01:00:40.090 --> 01:00:41.800 I wanted to give this a bit of a shout out 01:00:42.040 --> 01:00:42.720 because it's just, 01:00:42.940 --> 01:00:44.080 it's a weird time. 01:00:44.190 --> 01:00:44.720 And I think this, 01:00:44.980 --> 01:00:45.100 you know, 01:00:45.290 --> 01:00:47.780 looking back 10 years and looking forward 10 years, 01:00:48.020 --> 01:00:49.360 like this is kind of an interesting thing 01:00:49.470 --> 01:00:50.680 at the inflection point. 01:00:50.940 --> 01:00:52.280 Well, and especially because they talked about 01:00:52.280 --> 01:00:54.100 how they tried to do this as clean room, 01:00:54.320 --> 01:00:54.420 right? 01:00:54.600 --> 01:00:56.300 I think there's no source code to look at here. 01:00:56.390 --> 01:00:56.680 They literally, 01:00:56.870 --> 01:00:57.220 I think, 01:00:57.450 --> 01:00:59.360 threw a model at it and said, 01:01:00.299 --> 01:01:02.140 here's the thing I want you to reimplement. 01:01:02.520 --> 01:01:03.380 Go reimplement it. 01:01:03.800 --> 01:01:05.300 And here's the UI. 01:01:05.620 --> 01:01:06.940 Here's its feature list to make that. 01:01:07.839 --> 01:01:09.900 Go read the publicity material, 01:01:10.060 --> 01:01:10.140 right? 01:01:10.580 --> 01:01:13.520 There's a lot of documentation about how Photoshop is supposed to work. 01:01:13.920 --> 01:01:15.660 And if you give it access on the browser version, 01:01:15.860 --> 01:01:18.140 you can click through it and all that and crank away at it. 01:01:18.260 --> 01:01:21.760 Now, I'd be kind of curious as to how much their token budget was to pull that off. 01:01:22.020 --> 01:01:22.740 Very much curious. 01:01:23.200 --> 01:01:24.120 High. Very high. 01:01:24.360 --> 01:01:25.220 Very high, I'm sure. 01:01:26.360 --> 01:01:30.120 Because Cloud re-implemented a C compiler for the Linux kernel, 01:01:30.240 --> 01:01:33.040 and it cost them hundreds of thousands of dollars for something. 01:01:34.200 --> 01:01:34.320 Wow. 01:01:34.620 --> 01:01:38.680 But it's still kind of cool that this is what you theoretically can do 01:01:38.800 --> 01:01:41.140 with appropriate budget with LLMs. 01:01:41.500 --> 01:01:43.820 We talked about way back when. 01:01:44.920 --> 01:01:45.060 Yeah. 01:01:45.440 --> 01:01:48.100 I mean, way back to March, we talked about a clean room rewrite, 01:01:49.300 --> 01:01:50.980 and this was for Chardet. 01:01:51.380 --> 01:01:51.600 Yes. 01:01:52.060 --> 01:01:54.240 And this was like a big to-do. 01:01:54.410 --> 01:01:55.660 And this was just a library. 01:01:55.860 --> 01:01:59.080 Now we're like, oh, let's just redo giant suites of apps. 01:01:59.430 --> 01:01:59.660 You know? 01:02:00.150 --> 01:02:04.680 Yeah, it's a very interesting world we are entering for software. 01:02:05.360 --> 01:02:05.800 Very, 01:02:06.400 --> 01:02:11.520 I think a lot of the companies that their software was their moats 01:02:11.670 --> 01:02:15.220 because they put so much investment to it 01:02:15.450 --> 01:02:17.500 and the time it would take to reimplement it, 01:02:18.820 --> 01:02:19.700 deter competitors. 01:02:20.400 --> 01:02:20.900 But with AI, 01:02:21.400 --> 01:02:22.440 it's very interesting. 01:02:22.580 --> 01:02:25.440 Now, somebody could create a DaVinci Resolve 01:02:25.820 --> 01:02:28.720 specific to their own podcasting editing needs, 01:02:28.750 --> 01:02:31.080 and they don't need DaVinci Resolve 01:02:31.220 --> 01:02:33.580 or all the other ones that exist. 01:02:34.300 --> 01:02:35.640 So it's very interesting, 01:02:36.500 --> 01:02:39.480 the era of personal software we're entering into. 01:02:39.640 --> 01:02:40.000 Absolutely. 01:02:40.950 --> 01:02:41.880 Even if I never use this, 01:02:41.950 --> 01:02:43.560 I hope this puts pressure on companies 01:02:43.570 --> 01:02:45.220 to be more on top of things, 01:02:45.510 --> 01:02:46.100 newer versions, 01:02:46.280 --> 01:02:46.860 less expensive. 01:02:47.660 --> 01:02:50.060 Yeah, and I think for desktop software specifically, 01:02:50.380 --> 01:02:57.140 right where you can have where the actual desktop app gives the model something to work against to 01:02:57.340 --> 01:03:02.900 verify that it's working appropriately and the outcomes are what you expect is of versus like 01:03:03.200 --> 01:03:07.460 i'm going to re-implement slack or something it's like well there's all back end on that one that's 01:03:07.560 --> 01:03:12.400 way more complicated to try to recreate just based on how slack looks on the front end but for 01:03:13.100 --> 01:03:17.220 photoshop on this desktop right once again if you've got the token budget it can just turn away 01:03:17.240 --> 01:03:20.780 for quite a while and eventually just get probably a good enough axillin. 01:03:21.110 --> 01:03:21.380 I mean, 01:03:21.760 --> 01:03:22.280 or you just go use 01:03:22.380 --> 01:03:22.580 GIMP. 01:03:22.850 --> 01:03:23.120 I mean, 01:03:24.060 --> 01:03:24.120 yeah. 01:03:24.520 --> 01:03:27.980 I mean, how many features do you use of any given app? Probably 10%, right? If 01:03:27.980 --> 01:03:29.100 you just say I'm building it for me. 01:03:29.500 --> 01:03:31.580 Yeah. Like to Nick's point, just make it for you. Yep. 01:03:31.900 --> 01:03:32.020 All 01:03:32.040 --> 01:03:32.120 right. 01:03:32.300 --> 01:03:32.400 Well, 01:03:32.760 --> 01:03:36.300 that's enough of this, but I do think it's a really bizarre and interesting thing that 01:03:36.410 --> 01:03:39.400 happened this week. Anyone else got something I want to shout out before we get to the joke? 01:03:39.660 --> 01:03:41.740 I got a real time update for Nick. 01:03:42.200 --> 01:03:45.380 Nick asked about how he could know what models he could run 01:03:45.680 --> 01:03:46.520 on his hardware. 01:03:47.260 --> 01:03:48.820 I've got the screen up here for 01:03:49.779 --> 01:03:51.400 llmfit.org, and that's actually 01:03:51.480 --> 01:03:53.240 a TUI app that you can 01:03:53.900 --> 01:03:55.420 just load up. It detects your 01:03:55.660 --> 01:03:55.840 hardware. 01:03:56.130 --> 01:03:57.460 So, Michael, if you wanted to put that 01:03:57.600 --> 01:03:58.120 up on the screen. 01:03:58.400 --> 01:03:59.460 All right, give me something to search for. 01:03:59.540 --> 01:04:00.980 What am I looking for? No, I shared. 01:04:01.070 --> 01:04:01.780 No, I shared it already. 01:04:02.060 --> 01:04:02.800 Oh, your screen. 01:04:03.130 --> 01:04:04.000 Yeah, my screen. 01:04:05.200 --> 01:04:05.500 There we go. 01:04:06.070 --> 01:04:06.160 So, 01:04:06.450 --> 01:04:07.000 it detects 01:04:07.050 --> 01:04:07.500 your hardware. 01:04:08.060 --> 01:04:08.940 It loads up a whole bunch 01:04:08.950 --> 01:04:10.020 of models from different providers, 01:04:10.340 --> 01:04:10.820 like Olama, 01:04:11.160 --> 01:04:11.660 Lama CPP, 01:04:12.120 --> 01:04:12.860 MLX, Docker 01:04:13.040 --> 01:04:13.960 Runner Models, LM Studios, 01:04:14.380 --> 01:04:15.360 and 01:04:15.380 --> 01:04:20.540 you just filter for what models would run best and give you scores so you can actually see where it 01:04:20.540 --> 01:04:21.360 might be better fits. 01:04:22.160 --> 01:04:25.620 I mean be careful you're downloading potentially random models from 01:04:25.820 --> 01:04:30.020 potentially random places so you probably want to be careful which providers you pull those models 01:04:30.200 --> 01:04:34.940 in from but it's a cool little tool to see like what what actually would perform and is there a 01:04:34.980 --> 01:04:39.460 better version of that model that I'm already using that I should be using based on like quantization 01:04:39.980 --> 01:04:45.340 and the MLX optimization all those kinds of pieces so I thought that'd be pretty cool for folks who 01:04:45.360 --> 01:04:46.420 what model should I run? 01:04:46.840 --> 01:04:48.700 And the other one I got on my extras is the, 01:04:49.140 --> 01:04:50.380 my standard uv report, 01:04:50.920 --> 01:04:53.400 UV 0.13 makes Python 3.15, 01:04:53.930 --> 01:04:54.280 the default. 01:04:54.900 --> 01:04:56.520 So we keep it right up. 01:04:56.760 --> 01:04:57.000 Awesome. 01:04:57.200 --> 01:05:01.280 I've already run uv Python install 3.15, dash, 01:05:01.450 --> 01:05:01.540 dash, 01:05:01.800 --> 01:05:01.920 default, 01:05:02.160 --> 01:05:02.680 I'm good to go. 01:05:02.960 --> 01:05:03.740 Ooh, I need to do that. 01:05:04.680 --> 01:05:05.720 That way if you just type Python, 01:05:05.850 --> 01:05:07.680 it falls to that one rather than some other system. 01:05:07.720 --> 01:05:08.260 That's what I want. 01:05:08.700 --> 01:05:09.620 Or if you're on Windows, 01:05:09.830 --> 01:05:10.520 use the PyManager, 01:05:10.660 --> 01:05:12.780 just type PyManager install 3.15. 01:05:13.340 --> 01:05:14.540 T, get the free threaded version. 01:05:14.760 --> 01:05:14.980 Ooh. 01:05:16.680 --> 01:05:17.560 Brett, one-upping. 01:05:19.540 --> 01:05:19.820 All right. 01:05:20.040 --> 01:05:20.780 How about a joke? 01:05:21.200 --> 01:05:22.240 I give Steve the call out. 01:05:22.380 --> 01:05:25.640 He does all that, but I just know Windows has its own built-in manager 01:05:25.760 --> 01:05:26.680 that comes from Python.org, 01:05:26.880 --> 01:05:28.220 and Steve put a lot of time on it. 01:05:28.400 --> 01:05:32.000 And it's my North Star dream to get that on every major tier one 01:05:32.120 --> 01:05:34.620 or maybe even tier two platforms for Python someday. 01:05:35.020 --> 01:05:35.760 That would be awesome. 01:05:36.700 --> 01:05:40.240 Before we go on, you should give Nick a moment for at least one of his items on there. 01:05:40.300 --> 01:05:40.860 I think it's really important. 01:05:41.200 --> 01:05:41.560 Oh, okay. 01:05:41.780 --> 01:05:42.800 I didn't know you have some, Nick. 01:05:42.800 --> 01:05:42.980 Go ahead. 01:05:43.240 --> 01:05:43.600 Oh, yeah. 01:05:44.000 --> 01:05:47.040 I just wanted to quickly call out Black Python Devs. 01:05:47.440 --> 01:05:48.840 I'm wearing the T-shirt here. 01:05:49.720 --> 01:05:51.980 One of the amazing things that one of our members, 01:05:52.160 --> 01:05:52.460 Emmanuel, 01:05:52.620 --> 01:05:54.320 is doing is taking our 01:05:55.360 --> 01:05:59.660 Python Dev student ambassadors and going across campuses in Africa. 01:06:00.380 --> 01:06:06.980 So this is a really important initiative because if you guys don't know the continent of Africa, 01:06:07.660 --> 01:06:10.760 so many countries and so many youths there, 01:06:11.180 --> 01:06:13.460 we want to promote Python. 01:06:13.600 --> 01:06:16.700 for the next generation of Python developers. 01:06:16.880 --> 01:06:19.000 I think one of the CEOs of Anaconda 01:06:19.260 --> 01:06:22.020 says that he wants a world 01:06:22.200 --> 01:06:24.500 where the next billion users are Python users, 01:06:24.720 --> 01:06:25.460 something like that. 01:06:25.980 --> 01:06:29.240 Some of those users will be in Africa. 01:06:29.480 --> 01:06:31.820 So this is an amazing campaign. 01:06:32.480 --> 01:06:34.760 And if you go to Black Python devs 01:06:35.520 --> 01:06:36.900 and the links will be in the show notes, 01:06:36.900 --> 01:06:39.000 you can support that initiative. 01:06:39.540 --> 01:06:41.560 And I'm going to also shout out 01:06:41.580 --> 01:06:43.780 with Black Python Devs at PyOhio, 01:06:43.960 --> 01:06:46.260 they did a leadership summit again. 01:06:46.920 --> 01:06:47.940 And I mean, 01:06:48.260 --> 01:06:49.260 looking at the leaders 01:06:49.760 --> 01:06:52.740 that Python Devs has over the years grown, 01:06:53.980 --> 01:06:56.440 they have something really good going on 01:06:56.780 --> 01:06:58.960 and seeing those leaders emerge 01:06:59.200 --> 01:07:00.920 has been super rewarding 01:07:01.540 --> 01:07:03.840 and, you know, doing the good stuff. 01:07:04.100 --> 01:07:04.320 Awesome. 01:07:04.640 --> 01:07:05.200 Yeah, thank you. 01:07:05.500 --> 01:07:05.660 Yeah, 01:07:05.900 --> 01:07:06.160 thanks, Nick. 01:07:06.560 --> 01:07:07.620 Anything else? 01:07:08.220 --> 01:07:10.860 All right, I got a quick short joke here for us. 01:07:11.160 --> 01:07:12.480 This comes to us from Derek Chambers, 01:07:12.680 --> 01:07:14.840 sent us in, said, well, 01:07:15.060 --> 01:07:16.320 I felt this way at some point. 01:07:16.500 --> 01:07:19.680 So theoretically, a person's applying for a job, trying to look good. 01:07:19.980 --> 01:07:24.360 Interviewer says, says here on your resume that you're an exceptional Python developer. 01:07:24.500 --> 01:07:25.140 Can you prove it? 01:07:25.380 --> 01:07:25.460 Oh, 01:07:25.640 --> 01:07:25.840 sure, 01:07:26.260 --> 01:07:27.260 says the interviewee. 01:07:27.560 --> 01:07:28.340 Run any of my code. 01:07:28.380 --> 01:07:29.820 It raises exceptions a lot of the time. 01:07:30.900 --> 01:07:32.180 Oh, I'm a sucker for a good dad joke. 01:07:32.420 --> 01:07:32.660 Dot, dot, dot, bang. 01:07:33.160 --> 01:07:34.200 I love it. 01:07:34.540 --> 01:07:39.020 I also feel like I just reached dad level because I figured out the punchline before you got there. 01:07:39.100 --> 01:07:39.480 I was like, oh, 01:07:39.640 --> 01:07:41.040 wow, I finally leveled up my dad. 01:07:41.180 --> 01:07:42.580 Joke level is being an actual dad. 01:07:42.720 --> 01:07:43.120 There you go. 01:07:43.540 --> 01:07:45.280 I feel like now the new point is parenthood. 01:07:46.200 --> 01:07:46.580 That's right. 01:07:46.920 --> 01:07:47.000 Absolutely. 01:07:47.720 --> 01:07:47.840 All right. 01:07:47.860 --> 01:07:47.960 Well, 01:07:48.460 --> 01:07:48.660 Nick, 01:07:49.080 --> 01:07:49.300 Brett, 01:07:49.700 --> 01:07:49.960 Carol, 01:07:50.260 --> 01:07:51.320 thank you so much for being here. 01:07:51.480 --> 01:07:53.460 It's really nice that you came to celebrate this. 01:07:53.640 --> 01:07:54.140 And Calvin, 01:07:54.560 --> 01:07:54.900 thanks for helping. 01:07:55.380 --> 01:07:56.020 Happy 500. 01:07:56.520 --> 01:07:56.760 Yeah. 01:07:56.980 --> 01:07:57.740 Happy 500. 01:07:58.140 --> 01:07:59.440 And happy thank you to Brian, 01:07:59.860 --> 01:07:59.980 too. 01:08:00.420 --> 01:08:00.660 Absolutely. 01:08:00.900 --> 01:08:01.560 Thank you to Brian. 01:08:02.060 --> 01:08:02.280 Bye, all.