1
00:00:02,299 --> 00:00:04,240
Welcome to The Bootloader, I'm Tod Kurt.

2
00:00:04,800 --> 00:00:05,620
And I'm Paul Cutler.

3
00:00:06,000 --> 00:00:06,980
The show works like this.

4
00:00:07,260 --> 00:00:10,560
Tod and I have each brought three things to share, and we'll talk about each one for

5
00:00:10,680 --> 00:00:11,480
about five minutes.

6
00:00:12,100 --> 00:00:15,260
For detailed show notes and transcripts, visit thebootloader.net.

7
00:00:16,000 --> 00:00:17,380
Tod, what's your first one for us?

8
00:00:17,980 --> 00:00:21,460
All right, so if you're in the US, there is a private company, well, really a collection

9
00:00:21,580 --> 00:00:26,680
of private companies, with 80,000 cameras displayed across the United States that track

10
00:00:26,680 --> 00:00:30,900
your car and you and sell that information to local police. And a synth

11
00:00:31,060 --> 00:00:34,760
YouTuber named Ben Jordan hacked them to show us how they work. If you're into

12
00:00:34,920 --> 00:00:38,820
synthesizers and watch YouTube like me, you may have heard of Ben Jordan. Around

13
00:00:38,920 --> 00:00:42,940
eight years ago he started doing music gear videos, I'm sorry, music gear reviews,

14
00:00:43,100 --> 00:00:46,940
studio tips, music theory lesson vids. These videos were good, insightful, and

15
00:00:47,000 --> 00:00:51,200
had a bit of legitimacy coming from a professional recording artist. About two

16
00:00:51,260 --> 00:00:54,140
years ago he stopped doing gear reviews because the synth gear industry has

17
00:00:54,100 --> 00:00:59,840
created a fairly toxic environment for these synth tubers, almost amounting to payola. So Ben just

18
00:01:00,000 --> 00:01:05,059
opted out. Nowadays his videos still occasionally feature music production techniques, but he's also

19
00:01:05,110 --> 00:01:10,800
turned his eye towards wider aspects of his world. Issues with AI, free software, hardware hacking,

20
00:01:11,250 --> 00:01:14,560
and most importantly for the last six months he's been working with security researchers and

21
00:01:14,680 --> 00:01:19,280
investigative journalists to look into these cameras and tracking systems made by a company

22
00:01:19,220 --> 00:01:19,940
called Flock.

23
00:01:20,900 --> 00:01:25,220
The first kind of these are cameras called ALPR, automatic license plate recognition

24
00:01:25,400 --> 00:01:25,780
systems.

25
00:01:26,520 --> 00:01:29,640
They blend into the various poles along the side of the road.

26
00:01:30,300 --> 00:01:34,220
The startup Flock, who makes the majority of them and license the data to law enforcement,

27
00:01:34,360 --> 00:01:38,060
claims they increase the number of arrests made and decrease crime.

28
00:01:38,620 --> 00:01:42,260
But as has been reported several times in the last several years, these systems are

29
00:01:42,280 --> 00:01:46,440
rife with OCR errors, because they've got like, you know, they're built in little machine

30
00:01:46,440 --> 00:01:50,480
language things and glitches that lead to arrests of random people having nothing to

31
00:01:50,520 --> 00:01:51,640
do with the crimes in question.

32
00:01:52,400 --> 00:01:56,640
And if you drive in the US, your car has been logged into this system.

33
00:01:56,940 --> 00:01:59,020
Like you don't even have a choice.

34
00:02:00,280 --> 00:02:05,880
Back in August 2025, Ben posted a video that shows how just using a modern gaming GPU,

35
00:02:06,040 --> 00:02:09,740
you can crack the Wi-Fi WPA2 encryption of these cameras.

36
00:02:10,759 --> 00:02:12,880
But also the physical security of these cameras is a joke.

37
00:02:12,880 --> 00:02:15,080
If you have access to one, you can just pull it off the pole really.

38
00:02:15,700 --> 00:02:17,800
you can easily pop out its internal micro SD card

39
00:02:17,880 --> 00:02:20,120
that contains unencrypted video it has recorded.

40
00:02:21,720 --> 00:02:24,140
So yeah, if you're a creep in a neighborhood,

41
00:02:24,340 --> 00:02:27,440
you can get access to all your neighbors, you know,

42
00:02:27,640 --> 00:02:28,240
goings on.

43
00:02:28,850 --> 00:02:32,680
But more concerning is what Ben posted in November, 2025,

44
00:02:32,940 --> 00:02:34,680
with the help of security researchers and journalists

45
00:02:34,940 --> 00:02:35,940
from 404 Media.

46
00:02:36,630 --> 00:02:38,940
There are these newer Flock safety cameras,

47
00:02:39,220 --> 00:02:40,600
that's capital Flock, capital safety.

48
00:02:41,370 --> 00:02:42,840
They were fully open and viewable

49
00:02:43,260 --> 00:02:44,960
from the IoT search engine, Shodan.

50
00:02:46,140 --> 00:02:49,320
By the way, did you know there's a search engine for IoT devices?

51
00:02:49,990 --> 00:02:53,580
You can go to and you can basically find various IoT things that are hooked up to the internet.

52
00:02:54,860 --> 00:02:59,320
These Flock safety cameras are PTZ cameras and they're designed to track people.

53
00:02:59,700 --> 00:03:02,600
They have built-in AI models to recognize and zoom in on people.

54
00:03:03,240 --> 00:03:06,380
They even zoom in enough to see what people are watching on their cell phones.

55
00:03:07,640 --> 00:03:12,719
He was able to just look at this stuff and see people using commercially available facial

56
00:03:12,720 --> 00:03:17,380
recognition software, public databases, and data breaches that's publicly available, he

57
00:03:17,380 --> 00:03:21,180
was able to identify several of these people that he saw, find their addresses, and discover

58
00:03:21,260 --> 00:03:23,760
details about their finances, health, and family relations.

59
00:03:24,020 --> 00:03:24,740
I mean, what the hell?

60
00:03:25,340 --> 00:03:30,160
What's damning is after all these vulnerabilities were reported to Flock, they did nothing.

61
00:03:30,720 --> 00:03:35,100
So Ben went on to one of these publicly available cameras and read the official response from

62
00:03:35,200 --> 00:03:40,000
Flock saying there was no security problem as the creepy AI camera zoomed in on him reading

63
00:03:40,140 --> 00:03:40,220
it.

64
00:03:41,160 --> 00:03:43,600
It's kind of a beautiful example of the hack.

65
00:03:44,440 --> 00:03:45,200
And it wasn't even a hack.

66
00:03:45,370 --> 00:03:49,840
It was just going to a public website, searching and seeing public URLs.

67
00:03:50,680 --> 00:03:55,000
But thanks to his reporting, some police departments sold on Flux promises have been canceling

68
00:03:55,040 --> 00:03:55,800
their contracts.

69
00:03:56,800 --> 00:03:57,980
So change is possible.

70
00:03:58,030 --> 00:04:01,980
I recommend that everyone go bug their city representatives and figure out if they're

71
00:04:02,080 --> 00:04:03,940
using these systems and get them to stop.

72
00:04:04,700 --> 00:04:08,600
From a privacy perspective, these things are bad, that they exist at all.

73
00:04:09,120 --> 00:04:13,120
And from a technical perspective, as a hacker, as an engineer, having them implemented so

74
00:04:13,360 --> 00:04:14,400
poorly is just embarrassing.

75
00:04:16,019 --> 00:04:18,760
So let's get them turned off and put Flock and similar companies out of business.

76
00:04:19,380 --> 00:04:22,960
Links to all this stuff and all the reporting from 404 is in the show notes.

77
00:04:23,820 --> 00:04:28,160
I've been a member of the EFF for about 20 years and they've been covering Flock for

78
00:04:28,260 --> 00:04:28,720
a while.

79
00:04:28,860 --> 00:04:32,780
They've got a number of blog posts just over the last six months about how scary it is.

80
00:04:32,800 --> 00:04:34,720
And it's none of the government's business.

81
00:04:35,240 --> 00:04:41,140
You can't tell me that Flock isn't selling this information sideways to data brokers as well as law enforcement.

82
00:04:41,680 --> 00:04:46,920
And we know that the government, the federal government, buys information from data brokers.

83
00:04:47,140 --> 00:04:54,180
Well, it's no one's business if someone's going to an abortion clinic, for example, or any kind of medical appointment.

84
00:04:54,500 --> 00:05:00,240
And it's pretty easy just using metadata like he did to figure out where people are going and what they're doing

85
00:05:00,300 --> 00:05:03,200
when you start cross-referencing the license plates with the facial recognition.

86
00:05:03,240 --> 00:05:04,620
And it's just scary.

87
00:05:05,080 --> 00:05:11,080
Yeah, I mean, one of the examples he had was just he saw a couple come out of a Home Depot

88
00:05:11,180 --> 00:05:14,280
with a bunch of brand new fancy tools loaded up into their SUV.

89
00:05:15,100 --> 00:05:19,720
And he could then see what the license plate of the SUV is and find out where they lived.

90
00:05:20,100 --> 00:05:23,780
So he could go and get those brand new tools, you know, if he was so inclined.

91
00:05:24,500 --> 00:05:28,479
Man, it's just like, you know, that Flock and the companies that use the Flock data

92
00:05:28,480 --> 00:05:33,640
are correlating all this video and AI detected information with other public

93
00:05:33,840 --> 00:05:38,520
datasets or even private datasets. And so it's just yeah let's let's try to knock

94
00:05:38,580 --> 00:05:42,320
this down a peg and make it not so easy for these people to do this sort of

95
00:05:42,400 --> 00:05:47,220
stuff. Anyway Paul what's your first one for this time? I'm a sucker for a good

96
00:05:47,280 --> 00:05:52,360
mechanical keyboard going back to an old IBM replica I had 25 years ago and I

97
00:05:52,440 --> 00:05:56,840
recently found out that my code keyboard is no longer made as WASD keyboards has

98
00:05:56,740 --> 00:06:00,820
gone out of business. I don't need a new keyboard, but I'm always window shopping

99
00:06:01,120 --> 00:06:05,040
and the Naya Connect caught my eye. The Naya Connect is the second product

100
00:06:05,120 --> 00:06:08,500
launch from Naya who has a Kickstarter campaign for their new modular keyboard

101
00:06:08,680 --> 00:06:12,300
the Naya Connect. They previously launched their split keyboard also on

102
00:06:12,460 --> 00:06:16,700
Kickstarter. The Naya Connect features a low-profile keyboard that's ultra thin

103
00:06:16,740 --> 00:06:21,260
at just under 15 millimeters. But what's innovative about it is that it's modular

104
00:06:21,440 --> 00:06:26,720
allowing users to customize their keyboard layout and functionality. You

105
00:06:26,720 --> 00:06:32,700
including a multi-pad, a 4x4 macro pad that can work as a number pad or anything you program it for,

106
00:06:33,180 --> 00:06:35,700
an extra column of six keys laid out vertically,

107
00:06:36,340 --> 00:06:42,880
four different input modules to let you replace your mouse, including a trackball, touchpad, a dial, and a floating joystick.

108
00:06:43,800 --> 00:06:48,160
And what's really neat about all of these modules is they're made for both right and left-handed users,

109
00:06:48,320 --> 00:06:51,060
with connections for each module on either side.

110
00:06:51,620 --> 00:06:58,220
It also includes Naya Flow, their custom software that allows you to program your keyboard layout and all of the modules.

111
00:06:58,840 --> 00:07:04,480
If you're quick, you can jump on the Kickstarter as it has about a week left from when this episode is being released.

112
00:07:05,240 --> 00:07:11,060
Options start at about $180 and quickly go up from there depending on which modules you choose.

113
00:07:11,280 --> 00:07:19,080
One of the options was over $400, so you're paying for it, but it's still a discount from what they're going to charge from a retail price.

114
00:07:19,620 --> 00:07:23,460
>> Yeah. One of the really cool things I love about this is

115
00:07:23,600 --> 00:07:24,860
their little mouse module,

116
00:07:25,000 --> 00:07:28,160
their input module is one thing that has

117
00:07:28,240 --> 00:07:31,600
a little round space in the middle and then four keys on the bottom.

118
00:07:31,980 --> 00:07:33,580
That round space in the middle can be

119
00:07:33,640 --> 00:07:37,280
a touchpad area or it can be a trackball,

120
00:07:37,720 --> 00:07:42,920
or it can be one of those six-dimensional nubbins

121
00:07:43,000 --> 00:07:45,840
like you use in CAD programs like the Space Mouse.

122
00:07:46,940 --> 00:07:48,540
All those parts can be swapped.

123
00:07:48,640 --> 00:07:50,400
So you can have one sort of input module,

124
00:07:50,520 --> 00:07:51,360
but then you can swap out,

125
00:07:51,420 --> 00:07:53,680
oh, I wanna have a touchpad, I wanna have a trackball.

126
00:07:54,300 --> 00:07:55,460
And it's just really clever.

127
00:07:56,220 --> 00:07:57,640
- Yeah, if you're using FreeCAD,

128
00:07:57,640 --> 00:08:00,720
you might want the space type joystick thing,

129
00:08:00,840 --> 00:08:02,100
I can't even describe it.

130
00:08:02,540 --> 00:08:04,640
And then you want a touchpad for your everyday use.

131
00:08:04,720 --> 00:08:06,700
And like you said, you just swap them in and out.

132
00:08:07,080 --> 00:08:09,200
And you can put them on either side of the keyboard.

133
00:08:09,450 --> 00:08:09,580
- Yeah, yeah, yeah.

134
00:08:09,940 --> 00:08:10,700
- Which I think is great.

135
00:08:10,700 --> 00:08:12,320
- Get one of each and put one on each side.

136
00:08:12,780 --> 00:08:12,980
- Right.

137
00:08:13,860 --> 00:08:14,220
(laughing)

138
00:08:14,920 --> 00:08:18,059
- Yeah, a friend of mine is a real big believer

139
00:08:18,060 --> 00:08:21,960
of the, they've got the Mac Magic Trackpad thing,

140
00:08:21,970 --> 00:08:23,440
which is like a really big trackpad,

141
00:08:23,800 --> 00:08:24,960
and they have that on one side,

142
00:08:25,190 --> 00:08:27,700
and they have just a normal fancy mouse on the other side,

143
00:08:28,040 --> 00:08:31,880
because doing some trackpad-y gestures are really natural,

144
00:08:32,000 --> 00:08:35,060
kind of like pinch and zoom and moving windows around

145
00:08:35,140 --> 00:08:36,080
is really nice with the trackpad.

146
00:08:36,430 --> 00:08:39,460
And so I see them doing no keyboard things

147
00:08:39,520 --> 00:08:41,020
where they're just mouse and trackpad,

148
00:08:41,659 --> 00:08:42,599
and they're really fast.

149
00:08:43,020 --> 00:08:44,159
So I think that's what I would do

150
00:08:44,159 --> 00:08:44,840
if I had to get this keyboard.

151
00:08:45,880 --> 00:08:47,420
- Yeah, I use the Apple trackpad too,

152
00:08:47,500 --> 00:08:53,160
It does make a lot of stuff easier though. There's there's times where I wish I had a mouse for really detailed use

153
00:08:54,040 --> 00:08:54,980
But I could go either way

154
00:08:55,720 --> 00:08:57,500
Totally. What's your next one for us?

155
00:08:58,460 --> 00:09:03,880
Okay, so very exciting news the ESP 32 finally supports 5 gigahertz Wi-Fi. Yay

156
00:09:05,720 --> 00:09:06,820
Everybody get your streamers

157
00:09:07,380 --> 00:09:12,860
So if you don't know ESP 32 chips and the dev boards that have them are pretty much the best way to do Wi-Fi with

158
00:09:12,740 --> 00:09:16,340
microcontrollers in my opinion. They're fast, they have lots of memory, they have

159
00:09:16,380 --> 00:09:20,240
a rich built-in set of peripherals for driving things like displays, doing audio

160
00:09:20,440 --> 00:09:24,300
I/O, you know buttons, all that kind of stuff. And you can program them either

161
00:09:24,600 --> 00:09:29,900
using the vendors supplied ESP-IDF SDK, or you can use Arduino or MicroPython,

162
00:09:30,080 --> 00:09:33,020
CircuitPython, you can even use the really modern languages like Rust and

163
00:09:33,200 --> 00:09:37,740
Swift. But the term ESP32 is pretty misleading. It's not a description of a

164
00:09:37,820 --> 00:09:42,320
chip architecture, it's more like a brand, kind of like saying Ford for a car.

165
00:09:42,320 --> 00:09:45,620
there are multiple chip architectures in the ESP32 family,

166
00:09:46,220 --> 00:09:47,780
with the current more modern one

167
00:09:47,940 --> 00:09:49,360
being the RISC-V based ones.

168
00:09:50,140 --> 00:09:52,320
And it used to be, if you were to say ESP32,

169
00:09:52,690 --> 00:09:54,500
you meant Wi-Fi microcontroller.

170
00:09:55,060 --> 00:09:56,600
But even now that's not true

171
00:09:56,800 --> 00:09:59,940
because they recently announced an ESP32-P4

172
00:10:00,580 --> 00:10:01,640
that doesn't even have Wi-Fi.

173
00:10:02,500 --> 00:10:03,700
It has a ton of GPIO,

174
00:10:03,880 --> 00:10:05,500
it seems really good at driving big displays,

175
00:10:05,760 --> 00:10:07,260
but it's not a Wi-Fi chip.

176
00:10:07,880 --> 00:10:10,560
In fact, their dev board ships with two ESP32s on it,

177
00:10:10,960 --> 00:10:17,000
this new P4 chip and then like a ESP32-C6 or something little Wi-Fi coprocessor, kind of like

178
00:10:17,040 --> 00:10:20,780
what Adafruit does with their AirLeft boards. The problem with all these different variations

179
00:10:21,090 --> 00:10:28,120
of the ESP32 Wi-Fi chips though is they've all been 2.4 gigahertz Wi-Fi. Even the most recently

180
00:10:28,460 --> 00:10:34,920
announced ESP32-C6 chip that boasts Wi-Fi 6 support is still only on 2.4 gigahertz.

181
00:10:35,750 --> 00:10:40,440
This 2.4 gigahertz spectrum is the original one that the Wi-Fi was on. It's what like the old

182
00:10:41,059 --> 00:10:44,360
cordless cell phones are on, it's what Bluetooth is on, it's what your microwave

183
00:10:44,520 --> 00:10:49,040
oven is on, and it's still pretty good for Wi-Fi for longer distance stuff, but

184
00:10:49,100 --> 00:10:53,000
if you want higher data rates, then 5 gigahertz is pretty much the norm

185
00:10:53,060 --> 00:10:57,760
that everyone uses in their home and office setups. And it's to the point

186
00:10:57,860 --> 00:11:02,260
where a lot of Wi-Fi access points will encourage you to turn off the 2.4

187
00:11:02,420 --> 00:11:06,540
gigahertz side of things. So some people have gotten home with their little

188
00:11:06,940 --> 00:11:09,860
ESP32 dev board trying to get on their local Wi-Fi, and they've not been able to

189
00:11:09,800 --> 00:11:16,760
because they've had their 2.4 gigahertz side of things turned off. But now there's the ESP32-C5

190
00:11:17,020 --> 00:11:23,360
chip. It's a dual band Wi-Fi 6 ESP32 that does both 2.4 gigahertz and 5 gigahertz.

191
00:11:24,580 --> 00:11:30,940
And while both Espressif, who makes the ESP32, and Waveshare have had these larger dev boards,

192
00:11:31,140 --> 00:11:37,680
the sort of big for like testing out stuff, that use the ESP32-C5, just recently, like two weeks

193
00:11:37,560 --> 00:11:44,920
ago, Seed started selling a Xiao board with the ESP32-C5. This is great if you already have a

194
00:11:45,040 --> 00:11:50,880
Wi-Fi project using a CutiePie or a Xiao, you might be able to just pop it out and swap in this new

195
00:11:50,960 --> 00:11:59,100
ESP32-C5 Xiao to get your project on 5 gigahertz Wi-Fi. And the other cool thing is that the ESP32-C5

196
00:11:59,180 --> 00:12:05,480
also supports Matter and Thread protocols, so I suspect you will be seeing this chip in this Xiao

197
00:12:05,480 --> 00:12:08,400
board and maybe Adafruit will make a cutie pie version of it.

198
00:12:09,300 --> 00:12:11,820
We'll be seeing those in a lot of home automation projects in the future.

199
00:12:12,860 --> 00:12:17,180
Links to the C Studio blog post about this in the show notes.

200
00:12:18,010 --> 00:12:18,720
And I can't wait to get one.

201
00:12:18,800 --> 00:12:20,600
They're currently sold out because I didn't get there fast enough.

202
00:12:22,300 --> 00:12:26,160
Do you know if the C5 you can choose which band to use in software?

203
00:12:27,000 --> 00:12:27,600
I think so.

204
00:12:28,000 --> 00:12:33,680
One of the really cool things about the ESP-IDF, which is the SDK that Espressif uses, that

205
00:12:33,680 --> 00:12:37,580
Specifics is written for their their chips is they cover all the chips

206
00:12:37,800 --> 00:12:42,180
You just want to use one API, but they'll have different flags to specify things

207
00:12:42,220 --> 00:12:45,340
I'm pretty sure I saw in the ESPN it thing

208
00:12:45,400 --> 00:12:47,500
There's this way to specify what what band you want to be on

209
00:12:47,840 --> 00:12:53,400
I do know you can tell what band you're on when you after you've connected. So there's that at least I

210
00:12:54,200 --> 00:12:58,200
Was one of those users that when I first got into microcontrollers four or five years ago

211
00:12:58,120 --> 00:13:05,000
I had 2.4 gigahertz turned off and it took me probably a week to figure out why I couldn't get this thing on a network

212
00:13:06,459 --> 00:13:08,440
Drove me crazy. So I

213
00:13:08,940 --> 00:13:13,180
Bet I'm not the only one out there. Yeah. Yeah. No, it's it's a harder. It's a harder problem

214
00:13:13,200 --> 00:13:18,460
So I'm glad I'm glad that that hurdle has been overcome at least a little bit. All right Paul. So what's your what's your second one?

215
00:13:19,100 --> 00:13:22,780
It's no secret that we're fans of CircuitPython and Adafruit products on the show

216
00:13:23,220 --> 00:13:25,760
So let's take a moment and celebrate the Adafruit fruit jam

217
00:13:26,080 --> 00:13:33,020
The Fruit Jam was first released at the end of July in 2025 and then went out to thousands of Adabox subscribers last November.

218
00:13:33,750 --> 00:13:40,680
If you haven't heard of the Fruit Jam, it's Adafruit's take on building a mini-computer using the RP2350B microcontroller.

219
00:13:41,320 --> 00:13:47,780
With all the extra GPIO in the RP2350B, Adafruit was able to add DVI out using an HDMI port,

220
00:13:48,080 --> 00:13:52,380
Two USB ports for USB hosts to hook up gamepads, mice, and keyboards.

221
00:13:53,160 --> 00:13:54,360
A microSD card slot.

222
00:13:55,080 --> 00:13:57,400
And a Stemma QT port to add additional sensors.

223
00:13:58,060 --> 00:14:00,440
All in a size that's barely bigger than a credit card.

224
00:14:00,860 --> 00:14:02,960
This literally turned it into a microcomputer.

225
00:14:03,760 --> 00:14:07,760
Not only is the product innovative, I'm blown away by the community's response

226
00:14:07,810 --> 00:14:10,640
and all the cool software that is available to run on the Fruit Jam.

227
00:14:11,320 --> 00:14:12,900
First up is Fruit Jam OS.

228
00:14:13,640 --> 00:14:16,220
Once you have CircuitPython 10 installed on your Fruit Jam,

229
00:14:16,300 --> 00:14:19,560
you can install the OS like you would any other CircuitPython program.

230
00:14:19,920 --> 00:14:25,500
It includes a few games like Breakout, Minesweeper, and a Flappy Bird clone, an IRC client of

231
00:14:25,520 --> 00:14:31,540
all things, utility apps like PyPaint, a text editor, and LarsIO Paint, which you can use

232
00:14:31,540 --> 00:14:36,380
to create MIDI music, and a couple screensavers like the Flying Toaster and the Matrix, and

233
00:14:36,600 --> 00:14:36,680
more.

234
00:14:38,120 --> 00:14:44,460
Sean Carolan wrote a Pac-Man clone originally for the SeedWIO terminal, and Cooper Delrimple

235
00:14:44,460 --> 00:14:47,040
ported it to work on the Fruit Jam and Fruit Jam OS.

236
00:14:47,550 --> 00:14:49,460
I never would have thought you could write Pac-Man

237
00:14:49,560 --> 00:14:51,860
in just CircuitPython, but that's what Sean did.

238
00:14:52,680 --> 00:14:55,160
- Yeah, it's really good too. (laughs)

239
00:14:55,460 --> 00:14:56,940
- I'll be having Sean on a future episode

240
00:14:56,940 --> 00:14:59,420
of the CircuitPython show just to talk about Pac-Man.

241
00:15:00,820 --> 00:15:01,620
Speaking of Cooper,

242
00:15:01,900 --> 00:15:03,540
who is a CircuitPython community member,

243
00:15:04,160 --> 00:15:06,240
he's been doing a ton of stuff with the Fruit Jam

244
00:15:06,400 --> 00:15:07,840
and doing it behind the scenes.

245
00:15:08,210 --> 00:15:10,040
He was one of the first to create a game for it

246
00:15:10,100 --> 00:15:11,500
with Fruitress, a Tetris clone.

247
00:15:11,980 --> 00:15:13,680
He's also worked on game pad support,

248
00:15:13,940 --> 00:15:16,600
a Fruit Jam store, and more things that I can list.

249
00:15:17,700 --> 00:15:20,080
Back in December, my last guest on the CircuitPython show

250
00:15:20,140 --> 00:15:21,100
was Dan Cogliano.

251
00:15:21,980 --> 00:15:24,540
Dan ported the Z Machine, the engine that runs Zork,

252
00:15:24,640 --> 00:15:26,740
to the Fruit Jam so he can play Zork 1-3

253
00:15:27,040 --> 00:15:28,320
or other Z Machine games.

254
00:15:28,880 --> 00:15:30,480
He also has made a Moon Miner game,

255
00:15:30,940 --> 00:15:34,200
and he's working on a maze game

256
00:15:34,320 --> 00:15:35,820
that you'll be able to play on it as well.

257
00:15:36,860 --> 00:15:39,720
But where the Fruit Jam really shines is as an emulator.

258
00:15:40,440 --> 00:15:42,340
So many of the early computers have been ported

259
00:15:42,340 --> 00:15:49,300
work with the Fruit Jam. Mac emulator? Check. It can run System 2.0 through System 7.5.5.

260
00:15:50,000 --> 00:15:57,460
Apple II emulator? Check. Intel 286? Check. But my favorite emulator is the RetroJam by

261
00:15:57,660 --> 00:16:03,600
Frank Hoedemakers. It's written in C using the Pico SDK and can emulate most 8-bit gaming

262
00:16:03,720 --> 00:16:09,380
system from the 80s, as well as my favorite 16-bit system, the Sega Genesis/Mega Drive.

263
00:16:11,500 --> 00:16:16,200
Emulators include the Nintendo Entertainment System, the Sega Game Gear, and Master System,

264
00:16:16,640 --> 00:16:18,640
and the Nintendo Game Boy and Game Boy Color.

265
00:16:19,250 --> 00:16:22,340
It works with both a gamepad or a keyboard, which is wild.

266
00:16:22,560 --> 00:16:26,940
I would have never thought you could use a keyboard to play some of these old 8-bit games.

267
00:16:27,780 --> 00:16:30,780
It also includes the ability to save your games with some of the emulators.

268
00:16:31,680 --> 00:16:35,420
The Fruit Jam is only $40 and is available directly from Adafruit, though it's out of

269
00:16:35,540 --> 00:16:36,900
stock at the time of this recording.

270
00:16:37,480 --> 00:16:39,800
For $40, it really packs a punch.

271
00:16:40,220 --> 00:16:42,940
Yeah, and if you if you want to play this play this kind of stuff right now

272
00:16:42,940 --> 00:16:45,000
And you don't have you can't get a free gem

273
00:16:45,270 --> 00:16:49,700
you can cobble together like most of the functionality of a fruit jam by using a

274
00:16:50,360 --> 00:16:57,300
RP 2350 feather and the TLB 320 DAC and and the RP 2350

275
00:16:58,020 --> 00:16:59,520
HST to DVI

276
00:16:59,960 --> 00:17:03,500
Connector like like you can basically like before before the food free gem existed

277
00:17:03,750 --> 00:17:08,100
We were all trying to get this stuff to work by using these like like five or six parts

278
00:17:08,100 --> 00:17:12,680
And it's like they're actually pretty cheap to get it all assembled. So if you want to do it right now you can still

279
00:17:15,020 --> 00:17:15,380
Also

280
00:17:16,560 --> 00:17:19,540
for other cool or I mean like most of or many of the

281
00:17:20,100 --> 00:17:24,440
things you listed are on the adafruit - playground comm site and

282
00:17:25,319 --> 00:17:27,199
There's a lot of other really cool

283
00:17:28,160 --> 00:17:29,760
Projects that are for there for fruit jam on there

284
00:17:29,760 --> 00:17:34,940
If you just search for fruit jam, like one of the ones that I I've seen that I really liked is Sam Blaney

285
00:17:34,980 --> 00:17:40,900
He's made a couple of MIDI projects that use like you plug in a USB MIDI device into the fruit jam

286
00:17:40,900 --> 00:17:42,680
And you can like use it to do stuff

287
00:17:43,580 --> 00:17:44,700
but also he's created some like

288
00:17:45,480 --> 00:17:49,780
video test patterns and sort of video screensaver type things to help you sort of like

289
00:17:50,240 --> 00:17:56,120
Learn how the fruit jam does video and I found those very very useful. Yeah, both Sam and Cooper

290
00:17:56,200 --> 00:18:00,560
I've really knocked it out of the park with with all their community driven programs that they've written for the fruit jam

291
00:18:01,200 --> 00:18:03,560
Totally. What's your next one for us?

292
00:18:04,020 --> 00:18:08,480
All right, neural networks in CircuitPython with native code.

293
00:18:09,020 --> 00:18:13,120
A few weeks ago, I came across a really interesting blog post by Ashish Patel on

294
00:18:13,120 --> 00:18:16,520
how to embed native code versions of trained TensorFlow or

295
00:18:16,720 --> 00:18:18,720
PyTorch models in a CircuitPython.

296
00:18:19,500 --> 00:18:21,000
So if you're interested in machine learning or

297
00:18:21,140 --> 00:18:24,040
embedding native code in a CircuitPython, this is a really useful read.

298
00:18:24,920 --> 00:18:28,080
TensorFlow and PyTorch are both Python-based machine learning tools for

299
00:18:28,140 --> 00:18:29,880
building neural networks and other learning models.

300
00:18:30,280 --> 00:18:31,920
Normally, these run on real computers, but

301
00:18:31,740 --> 00:18:34,340
we've seen TensorFlow models being run on embedded dev

302
00:18:34,440 --> 00:18:35,860
boards for several years now.

303
00:18:36,320 --> 00:18:39,180
In fact, Adafruit made an Edge Badge version

304
00:18:39,210 --> 00:18:42,740
of their PyGamer board that could do TensorFlow-based voice

305
00:18:42,820 --> 00:18:46,080
recognition using TensorFlow Lite in Arduino.

306
00:18:46,380 --> 00:18:48,720
These models are trained on a large, fast computer.

307
00:18:48,900 --> 00:18:50,320
And then the resulting models exist

308
00:18:50,320 --> 00:18:53,700
as a blob of generated code with essentially a single function

309
00:18:54,260 --> 00:18:55,980
that takes an input and returns a result.

310
00:18:56,880 --> 00:18:58,980
This code is small enough to run on microcontrollers.

311
00:18:59,580 --> 00:19:01,940
And there are several mechanisms for sticking them

312
00:19:01,980 --> 00:19:05,100
into your Arduino code or Rust code or Python code

313
00:19:05,120 --> 00:19:07,780
or whatever, but getting them into CircuitPython

314
00:19:07,820 --> 00:19:08,280
has been tricky.

315
00:19:09,220 --> 00:19:12,220
You can run a pure Python version of your model,

316
00:19:12,700 --> 00:19:14,180
and Ashish shows you how to do this,

317
00:19:14,440 --> 00:19:16,440
but that runs pretty slowly as you might expect.

318
00:19:17,420 --> 00:19:20,500
But Ashish's blog post gives a step-by-step process

319
00:19:20,860 --> 00:19:22,620
starting with the training of the model

320
00:19:22,740 --> 00:19:24,899
and converting the result into C

321
00:19:24,900 --> 00:19:29,440
using this tool called onnx2c.exe, whatever.

322
00:19:30,460 --> 00:19:32,960
Onnx is the Open Neural Network Exchange,

323
00:19:33,760 --> 00:19:34,460
something I didn't know about.

324
00:19:34,520 --> 00:19:36,720
It's a standard way of representing machine learning models.

325
00:19:37,700 --> 00:19:40,000
Once you run this tool, you have a resulting C code

326
00:19:40,160 --> 00:19:44,040
that's just a bunch of tensor vectors

327
00:19:44,600 --> 00:19:47,020
and no CircuitPython-specific code in it.

328
00:19:47,500 --> 00:19:50,160
So the next step is to show how you can embed that C code

329
00:19:50,220 --> 00:19:51,000
in a CircuitPython.

330
00:19:51,560 --> 00:19:52,500
And he does it in a way

331
00:19:52,520 --> 00:19:53,540
that's pretty clean and understandable.

332
00:19:54,520 --> 00:19:57,620
There's several little niggly details you need to do

333
00:19:57,700 --> 00:20:01,980
to make the CircuitPython source tree see your code

334
00:20:02,360 --> 00:20:05,020
to add your new native code that you're putting in there.

335
00:20:05,620 --> 00:20:07,680
It's more frustrating than just sticking your code

336
00:20:07,860 --> 00:20:09,400
in your board's definition directory,

337
00:20:09,460 --> 00:20:10,760
which is kind of what I would hope.

338
00:20:11,500 --> 00:20:13,960
But it shows exactly how all they need to do.

339
00:20:14,400 --> 00:20:17,240
So even if you're not interested in machine learning,

340
00:20:17,720 --> 00:20:20,400
but are interested in adding native code to CircuitPython,

341
00:20:20,740 --> 00:20:21,860
this post is pretty useful.

342
00:20:22,040 --> 00:20:26,620
It goes into much more detail than a gist I put up a couple years ago.

343
00:20:26,620 --> 00:20:31,360
And I think it's actually more up to date because the way that CircuitPython is compiled

344
00:20:31,460 --> 00:20:33,000
kind of evolves over time.

345
00:20:33,000 --> 00:20:36,580
And so what I wrote in 2022 in a gist is probably out of date now.

346
00:20:37,500 --> 00:20:39,620
But yeah, thank you Ashish for that blog post.

347
00:20:39,900 --> 00:20:40,200
Super handy.

348
00:20:40,800 --> 00:20:42,360
My partner is the smart one in the house.

349
00:20:42,480 --> 00:20:44,080
She's getting her doctorate in machine learning.

350
00:20:44,260 --> 00:20:48,540
So I read Ashish's article and most of it went over my head, but she would probably

351
00:20:48,600 --> 00:20:49,340
understand it.

352
00:20:50,840 --> 00:20:55,360
But I need to reach out to Ashish because he needs to be on the CircuitPython show.

353
00:20:55,460 --> 00:20:56,880
What he's done is pretty cool.

354
00:20:57,080 --> 00:21:01,600
And the instructions do make sense because he does take you through it step by step.

355
00:21:02,240 --> 00:21:02,360
Yeah.

356
00:21:02,780 --> 00:21:02,880
Yeah.

357
00:21:02,980 --> 00:21:08,940
And it's so nicely embeddable, these little models, because they are literally just a

358
00:21:08,940 --> 00:21:12,240
single C function that takes one input and returns one output.

359
00:21:13,460 --> 00:21:14,860
So it's really handy, really useful.

360
00:21:15,540 --> 00:21:17,040
All right, Paul, what's your final one for this time?

361
00:21:17,240 --> 00:21:22,120
I've been following on social media Washington State's proposed bill that basically adds

362
00:21:22,280 --> 00:21:25,880
DRM to all 3D printers so they can't print ghost guns.

363
00:21:27,470 --> 00:21:32,320
On January 25th Adafruit published a must-read blog post that has all the details and more.

364
00:21:33,320 --> 00:21:37,100
Washington State Representative Osman Salahuddin has proposed two bills.

365
00:21:37,680 --> 00:21:43,759
HB 2320 adds criminal liability around digital firearm manufacturing code and classifies

366
00:21:43,760 --> 00:21:46,860
the machines that make these parts as part of the firearms law.

367
00:21:47,940 --> 00:21:56,060
HB 2321 requires any 3D printer sold or transferred in Washington after July 1, 2027 to include

368
00:21:56,340 --> 00:22:00,840
mandatory DRM that has blocking features that can detect a firearm's blueprint.

369
00:22:01,510 --> 00:22:06,520
As Adafruit wrote and shared, printers would have to examine every file, compare it against

370
00:22:06,520 --> 00:22:12,440
a centrally maintained list of prohibited designs, reject flag jobs, and be engineered

371
00:22:12,440 --> 00:22:16,600
to resist bypass by users with significant technical skill.

372
00:22:17,660 --> 00:22:20,120
Which is pretty much everyone who owns a 3D printer.

373
00:22:22,580 --> 00:22:25,000
While I'm all for gun control, this is just crazy.

374
00:22:25,140 --> 00:22:28,940
You can't outlaw the tools to create something just because it creates something you don't

375
00:22:29,040 --> 00:22:29,320
like.

376
00:22:29,820 --> 00:22:30,420
What's next?

377
00:22:30,600 --> 00:22:33,020
Outlawing drill presses and milling tools as well?

378
00:22:33,540 --> 00:22:34,060
Exactly.

379
00:22:35,430 --> 00:22:38,060
You can't tell me that people won't find a way around this.

380
00:22:38,200 --> 00:22:40,360
When it comes to DRM, they always do.

381
00:22:40,800 --> 00:22:45,380
What's stopping someone from adding a Raspberry Pi or a custom microcontroller and flashing

382
00:22:45,540 --> 00:22:48,860
Clipper on it, which is done all the time in the 3D printing community?

383
00:22:49,840 --> 00:22:52,600
DRM never works and this is a terrible idea.

384
00:22:53,220 --> 00:22:57,260
If you live in Washington state, reach out to your elected representatives and tell them

385
00:22:57,360 --> 00:22:58,800
that this is a terrible idea.

386
00:22:59,580 --> 00:23:03,080
This is definitely something to keep an eye on because if one state is thinking about

387
00:23:03,160 --> 00:23:04,620
it, I bet others will be too.

388
00:23:05,620 --> 00:23:09,680
The language as it's proposed is way too broad, and I'll leave you with this quote

389
00:23:09,680 --> 00:23:11,240
from the Adafruit blog post.

390
00:23:12,080 --> 00:23:14,420
None of these controls stop determined actors.

391
00:23:14,740 --> 00:23:20,520
What they do is burden lawful users, destroy open source innovation, and force proprietary

392
00:23:20,680 --> 00:23:23,620
stacks and cloud services, not safety.

393
00:23:24,799 --> 00:23:26,240
Yeah, it's so dumb.

394
00:23:27,000 --> 00:23:29,300
This sort of stuff seems to pop up every couple of years.

395
00:23:29,460 --> 00:23:36,140
And it seems like every time I see the video of the little desktop mill, not even a CNC

396
00:23:36,080 --> 00:23:41,420
mill, just a manual mill of like someone who can create a gun in a couple of hours just

397
00:23:41,460 --> 00:23:42,460
out of bar stock.

398
00:23:42,940 --> 00:23:46,180
And it's like, it's not the tools, man.

399
00:23:47,020 --> 00:23:47,120
Right.

400
00:23:50,100 --> 00:23:52,520
We just don't need more DRM in the world.

401
00:23:53,100 --> 00:23:54,300
Yeah, yeah, no doubt.

402
00:23:55,340 --> 00:23:56,020
Well, that's our show.

403
00:23:56,760 --> 00:23:58,400
Before we go, I'd like to take a moment.

404
00:23:58,720 --> 00:24:02,440
I live in the outer suburbs of Minneapolis, which is currently occupied by ice.

405
00:24:03,240 --> 00:24:07,280
If you're looking for ideas on how you can help support those on the ground, please visit

406
00:24:08,140 --> 00:24:09,180
StandWithMinnesota.com.

407
00:24:09,940 --> 00:24:12,820
Thanks for listening, and until next time, stay positive.

408
00:24:13,080 --> 00:24:14,040
(bright music)