Here’s a quick cheatsheet on setting and reading environment variables across common OS’s and languages.
When developing software, it’s good practice to put anything you don’t wish to be public, as well as anything that’s “production-environment-dependent” into environment variables. These stay with the local machine. This is especially true if you are ever publishing your code to public repositories like Github or Dockerhub.
Good candidates for environment variables are things like database connections, paths to files, etc. Hosting platforms like Azure and AWS also let you easily set the value of variables on production and testing instances.
I switch back and forth between Windows, OSX and even Linux during development. So I wanted a quick cheatsheet on how to do this.
Writing Variables
Mac OSX (zsh)
The default shell for OSX is now zshell, not bash. If you’re still using bash, consider upgrading, and consider using the great utility “Oh My Zsh.”
and add export BASEBALL_TEAM=Seattle Mariners to this file. Be sure to restart a new terminal instance for this to take effect, because ~./zshenv is only read when a new shell instance is created.
bash shell (Linux, older Macs, and even Windows for some users)
export BASEBALL_TEAM=Seattle Mariners
echo $BASEBALL_TEAM
Seattle Mariners
printenv
< prints all environment variables >
// permanent setting
sudo nano ~/.bashrc
// place the export BASEBALL_TEAM=Seattle Mariners in this file
// restart a new bash shell
Windows
Right click the Windows icon and select System
In the settings window, under related settings, click Advanced System Settings
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
Revisiting the Enchanted Rose project four years later, this time with an open-source build based on Raspberry Pi.
Enchanted Rose 2.0 with Falling Petals
Got a performance of Beauty and the Beast coming up? For about $200, you can build this Enchanted Rose with Falling Petals.
Dropping The Petals
There are a few approaches to make the petals drop. You’ve got a microcomputer (a Raspberry Pi) which can send electrical current HIGH or LOW on an output pin. How do you translate that into petals physically dropping on command?
In version 1.0, I used servo motors to pull fishing line. That fishing line was attached to a spring-loaded magnet in the bud of the rose. When the fishing line was pulled, the magnet would retract into the bulb and drop the petal. This worked very well, but it’s a delicate mechanism and wouldn’t ship well, and I wanted the 2.0 design to work for shipment.
After horsing around with solenoids (which didn’t work very well because they (a) were too bulky and (b) weren’t strong enough), the design I ultimately settled upon was to use individual, cheap 6v air pumps to push air through a small opening, and blow out a toothpick, to which a petal was attached. I found that I could “overclock” the air pumps at 12V (higher than rated) to push slightly more air out than rated. As long as the rose petals remained relatively light, bingo, it works. Here’s an in-progress video:
Just some notes to the person who bought this prop, but it gives you an idea of the functionality of the final 2.0 design. It also shows how the cloche can interfere with the falling petals if not set properly. TIP: Find as wide a cloche as you can.
Here’s an earlier video showing mock petals which were much lighter — tissue paper:
Design Goals
Drop 4 petals on cue, one by one
Turn on and off stem lights
Turn on and off accent lights, set color
Easy to reset between productions
Wireless — allow to be set on a pedestal on stage
Control via mobile web browser
Durable enough to ship
Do NOT require that it connect to a local internet
Easy to use interface that an elementary school stage crew could use
Easy on-site setup. Ideally, just turn it on and it works.
Parts List
2 half breadboards
Raspberry Pi: I used a Raspberry Pi Zero 2W, but any Raspberry Pi which has onboard wifi will work, such as a 3B+, 4, Zero W, etc. Do not choose a Pico; that’s a different product line. You’ll also need a MicroSD card to store the operating system.
Cloche: This was the widest and tallest one I was able to find for easy ordering. But I really did prefer the one I was able to find for the 1.0 build 4 years ago. This one is just barely big enough. Shop around — you may have better luck finding a wider one in a local arts and crafts store. You really want one that is fairly wide, because the petals need to fall freely and not get hung up against the side.
Enchanted Rose with silk petals – look for lightweight petals. The standard one on Amazon is too heavy — I went to Etsy to shop for mine.
Toothpicks — You are looking for ones which slide in and out of the aluminum tubing very easily, but which also block the most amount of air (to build up pressure — you need some force on these to pop the petal.)
Aluminum metal tubing (small) — The purpose of these aluminum tubes is to eliminate a lot of friction that existed between the toothpick and the pneumatic tube. In other words, toothpicks and these tubes sort of act like a blowdart. By using aluminum tubing as a gasket within the pneumatic tubing, a lot of the friction is reduced.
Cigar box, wooden box or other case for components below the cloche
Pipe insulation foam for vibration-dampening, to just fit around the pumps
Parts List
Before You Begin
You’ll want to get your Raspberry Pi set up with Raspbian OS on a micro-SD card. Make sure you have SSH access to your RPi. You might also want to set this up to present its own SSID/wifi network. I’ve got a blog post on that here.
If you’re new to Raspberry Pi, here’s one video to get you started in setting up the operating system and SSH:
I also loved using Visual Studio Code’s “SSH” remote features. Setting this up can be a tiny bit tricky, but stick with it. Once you get this working, it’s incredibly productive. It allows you to use your desktop (Mac, Windows or Linux) to work with files and directories on your RPi.
NeoPixels will also require ROOT-level permissions on your Raspberry Pi to install some drivers. Now, there are all kinds of security folks telling you that using root on your Raspberry Pi is a bad practice for stuff like this. But since this is a simple stage prop, I had no problem with the idea of just using my root user and a password for the entire build, and it saved a lot of file permission headaches to just be “root” and be done with it:
Set a password for the root user, reboot your pi, and connect in as root from then on. While it may not be “best practice” for security, trust me, you’ll save a lot of headaches with installing libraries etc.
Assembly
There’s both a hardware circuitry part and a software part. Let’s get some of the basic circuitry working first.
For ease of explanation, the hardware components can be thought of as three different “zones,” all powered by a single Raspberry Pi. These zones don’t need to talk to one another, except that all ground wires should be connected together and also to the RPi’s GND pin.
There are two zones powered by 4.5V: The Fairy lights and the Neopixel lights. There is one zone powered by 12V: the pump zone.
(DO NOT power the Neopixel lights with the 8 AA batteries (12V) used to power the pumps, as you will permanently damage the Neopixels!)
Zone 1: Control the stem lights (4.5V)
Let’s start here, because it’s the easiest circuit. These fairy lights are controlled by 3 AA batteries (4.5V.) You want to be able to have a GPIO pin on the Raspberry Pi toggle tell a transistor to allow current to flow to power these lights.
Cut the fairy light wire, and strip the ends. It’s a bad idea to power the LED directly from the Raspberry Pi, as altogether, the current draw might be more than the Pi can handle. So, you’ll want to use a standard Transistor (like a TIP 102) to switch the current.
Now, run a wire from the GPIO pin of choice — I chose pin 25 (BCM) — to the “base” pin of the transistor. When this is toggled HIGH, it will close the circuit of the LED fairy lights and thus turn them on or off programmatically.
[CIRCUIT DIAGRAM HERE]
Zone 2: Control the 4 pumps (12V)
The main mechanism to drop these petals is to blow air through tubes which have loose petals. I chose to cut some aluminum tubing to reduce the diameter of airflow and also to reduce the friction with the wooden toothpicks:
A close-up of one of the petals. You can hot-glue a toothpick into red silk. Placing these petals into the metal tubes allows the pump to then “pop” them out when they turn on.
For the electronics, you need a diode for each of the 4 pump circuits, for the reasons described in this video (where this video says “solenoid”, substitute in the “pump” in your mind — the same principle applies):
In my Enchanted Rose prop, I built a breakout board with a bunch of these MOSFET transistors and diodes. I overbuilt this board, making it with 6 different circuits when it only really needed 4. It’s messy looking, but pretty simple. Each of these circuits has a MOSFET transistor and a diode. Take care to get the direction of the diode correct — it needs to be pointing toward the positive load. (A diode is rather like a backflow valve in plumbing — it allows current to flow in only one direction, toward the “ring” on the diode.) The yellow wires in this photo below come from the Raspberry Pi (GPIO pin) and when they are HIGH, they allow current to pass from the “Base” of the transistor to the “Emitter.”
Basically you want four circuits which allow the GPIO pins on the RPi to turn on and off the motor. A single such circuit looks like this. (Note that instead of a 9V battery, I “overclocked” it with a 12V battery pack — I found these pumps can handle short bursts of 12V.)
Each pump motor needs a circuit like the below. Since RPi cannot output 9V, we’ll make use of a transistor to switch a supplied 9V current on and off. (Note that in production, I ended up boosting the power supply to 8 AA batteries, or 12V. The motors can handle this higher voltage for short duration, even though they’re rated for 6-9V.)
Wiring diagram for pump motors
When you make six such circuits on a PCB board and don’t spend too much time cleaning up the build, it looks like this:
I cut insulation foam for these motors (pipe insulation) which helps reduce the vibration sound a bit.
Battery pack; there are pumps inside these foam cylinders; the foam helps dampen some sound
Since these motors generate current spikes when they are powering down and powering up, it’s very important to protect your RPi with a diode, as follows (the directionality matters! pay attention to the line on the diode and the positive lead from the motor):
Zone 3: Control the Neopixel accent lights (4.5V)
DO NOT POWER THE NEOPIXEL WITH ANYTHING GREATER THAN 5V, AS YOU WILL PERMANENTLY DAMAGE THEM.
The python code to control the Neopixel needs a set of libraries called “CircuitPython” to work. You’ll need to follow the NeoPixel Adafruit guide to installing CircuitPython.
I strongly recommend you read through the NeoPixel to Raspberry Pi Wiring Guide on Adafruit, which is likely to be more up-to-date than anything I’m providing here. I went to a 74AHCT125 level-shifter chip to map the Raspberry Pi’s 3.3v output to proper 5V levels. Note that the power cord shown in the diagram below simply feeds out to the 4.5V battery pack. The ground rail needs to be connected to the RPi’s ground rail. All the grounds in this project all connect together, and connect back to the Rpi’s ground. You can and should run both the LED fairy lights and the Neopixels off the same battery pack.
I’ve published two Github repos that should really accelerate your work here.
First, let’s run through what you need, because it’s helpful to see this at a high level. If you want to go the easiest possible route, set your RPi to render its own wifi network, and make sure you configure your Raspberry Pi so that it is known as 196.168.11.1 on its own wifi network; that’s what the software is built for.
You will need the following software components set up:
Website software to deliver a nice user interface so that people don’t need to call an API directly to change the lights, drop the petals. See this Next JS Front-End Repository on Github. Run this on an Apache web server on the Raspberry Pi. To do this, set up an Apache Web Server on your Raspberry Pi. Make sure it’s working by visiting the IP address of your Raspberry Pi from a web browser. Next, take the “out” folder from the repository above, and place it in your Raspberry Pi’s /var/www/html folder. You should then have a web front-end that is calling locally to the device’s Flask port (hardcoded in the source as: http://192.168.11.1:5001).
Set up the web server and the API, and make sure they start properly at Raspberry Pi boot-up.
Make your Raspberry Pi its own local wifi network, so you don’t have to rely upon tapping into the router or a network wherever your production happens.
Configuring Raspberry Pi for “Ad Hoc Network”
Now, how do we get a mobile phone to reach the Raspberry Pi? Of course, if we know the networking details at the production venue, we could in theory just have the RPi join the wifi network of the theatre, and then somehow go to the theatre’s router settings and look for its address. But that’s a pain and doesn’t scale well.
So I was excited to hit upon the idea of having the prop create its own “Ad Hoc Network” (i.e., it’s own Wifi SSID that you can join.) Not only does this eliminate the need to have the stage manager at each playhouse somehow find a way to connect the prop into their own local network, but it removes any requirement for a display or keyboard to ship with the device for configuration purposes.
All they you to do is plug in the device, wait for boot-up, and it presents its own Wifi Network (“EnchantRose”.) You then simply enter the password, bring up a web browser on their phone, and visit a pre-determined IP address which I’ve configured: http://192.168.11.1.
It took a while to figure out how to configure Raspberry Pi to do this, since the instructions have changed seemingly with every release of Raspbian, the RPi operating system. But I figured it out by adopting existing instructions, and blogged about it here. The steps in that blog post provide all you need to do to get your Raspberry Pi setting its own network. To ensure the Enchanted Rose repository code works, be sure to set your network to the instructions given — you want the RPi to be able to reference itself on its own network as 192.168.11.1.
What’s most exciting about this is that it’s a pattern that can has many applications in the world of “Internet of Things” — run Apache2 on your Raspberry Pi to host a website, and render that website visible via the Rpi’s own password-secured wifi network.
Current Status
The new version 2.0 prop is fully functional. I’ve sold it and shipped it to one of the many production managers who have inquired. As this post continues to get indexed by Google, I get hit up for these builds; at this writing it’s over a dozen such requests, from FL, CA, HI, UK, NJ, CO and more.
UPDATE, late January 2023: This 2.0 prop is sold to a production in Hawaii, and has been shipped off.
I’ve decided for now that the logistics and expense of shipping these out to productions is just too complicated, but I’d love to hear from you if you’ve built one of these. If you’ve got specific questions on assembly or the software, drop me a note. On with the show!
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
The Raspberry Pi microcomputer is great for building “Internet of Things” devices. This technical post describes the steps to get your Raspberry Pi to broadcast its own local network and bridge to an Ethernet network if available. There are other guides showing how to do this on the Internet, but they are often woefully out of date, leading people down real rabbit holes.
The Raspberry Pi (RPi) microcomputer is great for building cheap “Internet of Things” devices. In this tech note, I outline how to set up a Raspberry Pi device so that it broadcasts its own local wifi network, so you can join it from a mobile phone. From there, you could launch a web browser and point it at the RPi’s website, which in turn controls the board’s I/O devices. This means you could use a mobile phone, say, and a Raspberry Pi device, and control it without any larger Internet connection.
The steps below also essentially turn the RPi into a “wifi access point,” which basically can take a hardline Internet network (LAN) and present a wifi front-end (WLAN.)
Turn your RPi into a wifi access point, or also its own ad-hoc network
So how do you get a Raspberry Pi to create its own wifi network? There are several “How To’s” on the web, but nearly all of them I followed are out of date. You need to install a few services, set up some configuration to define your network, and then reboot the Pi once everything is all set.
One great use-case is to build a standalone device which can be controlled via a web interface from, say, a mobile device. RPi’s can easily run web servers like Apache, so once you set up an RPi to broadcast its own private wifi network, you can easily interact with the Raspberry Pi’s devices and sensors via your website.
You’ve probably seen this kind of behavior if you’ve set up a wifi printer, or smart switch, or wifi security camera. These devices often have modes where they broadcast their own local wifi network, then you use a browser or a configuration app to “join” it, you do some setup process, and then it in turn uses what you’ve input to restart and join your home network with those configuration details thus supplied.
Enchanted Rose Prop 2.0
I had a use-case for this “ad hoc networking” with a stage prop I’m building.
A few years ago, I built an Enchanted Rose prop for my daughter’s school production of “Beauty and the Beast.” It let the stage manager drop petals and turn on lights on queue. It was based on Arduino and Bluetooth. I blogged the instructions in a two-part series. As a result thanks to Google, every couple of months or so, I get an inquiry from a production manager from somewhere in the world who wants to know more about how to build such a prop. Invariably, they follow these instructions and then hit a dead-end (which I’ve now noted in the posts.) The problem is, the version 1.0 device that I built is based upon an Arduino microcomputer (fine) with a Bluetooth add-on board which is now discontinued (not fine.) Worse, its controller was a proprietary Swift app which I wrote in a many-years-out-of-date dialect of Swift, which had to be installed straight from XCode onto a machine, as I opted not to publish it in the App Store. Apple has made many changes to Swift since then. The app as originally written no longer compiles, and Apple itself make it very difficult to distribute “one off” apps to people. (You can’t just post the app download somewhere, or email them a link — Apple requires that the developer add the person to an ad-hoc distribution list.)
So I began to think about how to rebuild it in a more open-source way.
Motivation for Ad-Hoc Network
At the heart of the version 2.0 of this prop is a Raspberry Pi Zero 2W, which runs its own little Apache web server to control lights and the falling petals. Ideally, a stage manager would simply need a mobile phone or iPad or some kind of web browser on a device equipped with wifi to communicate with the prop.
I’ve heard interest in this prop from stage managers in the UK, California, Colorado, Texas and more, and I wanted to re-build this prop in a more robust, even shippable way. So on the mechanical front, instead of using delicate springs and fishing line, it now uses pumps and air to push the petals off the rose. I’ve got the motors and circuits all working fine. Before moving on to the final aesthetics, I’m now working through matters related to the deployment setting(s) this prop will run in.
There’s a high school in California putting on this play in March. I don’t know what network environment is at that school, nor should I particularly care. I certainly don’t want to request and configure wifi passwords for the prop to “join” before shipping the device out. Rather than equip the prop with some kind of user interface to allow it to “join” the local WAN (needlessly wasteful screen and keyboard and instructions), it’s far better for it to broadcast its own tiny network, and be controlled from backstage just on its own. You’ve probably seen this type of “ad hoc” network when setting up, say, a printer or security camera, smart speaker, or smart switch.
But wow, getting ad-hoc networking up and going in Raspberry Pi is complicated!
THAR BE DRAGONS. It has taken a full day to get this up and running. Not only is Linux itself pretty arcane, but a much bigger problem is that so many of the instructions on ad-hoc networking for Raspberry Pi’s are wildly out of date. The Raspberry Pi Foundation changes these methods seemingly with every release of the OS.
So, to save my future self (and anyone Googling who lands here) many headaches… After much tinkering, the following instructions successfully configured a Raspberry Pi Zero 2W device to create and broadcast its own wifi network with a password. It allows clients (such as mobile browsers) to join it, and visit its control webpage hosted by the device on its Apache server. In short, the below just plain worked for me. Many, many other “how-to’s” on the web did not. A day wasted; I don’t want to do that again.
After you complete the below, you should be able to use a wifi device and “join” the network that the Raspberry Pi broadcasts. I successfully did so from both a Mac desktop and an iPhone. I could also then visit the apache page hosted on the Pi from these devices. This opens up a world of possibilities for “build and deploy anywhere” devices.
There are a lot of steps here, but once you’re finished configuring your RPi, it should have its own private wifi network, and yet still be connectable via Ethernet. This is called an “Access Point” configuration.
Prerequisite: Use Raspbian “Buster,” a prior version of Raspbian OS
Make sure you use Raspbian “Buster” for the instructions. “Buster” is not, in fact, the most current version of the OS.
For the instructions below to work, you mustuse the older “Buster” version of Raspbian. Let me repeat, because some of you will be very tempted to use the latest version of Raspbian OS, and then will wonder why the network isn’t showing up. The Ad-hoc networking steps described below have only been tested to work on the “Buster” version. In fact, I tried and failed to get them to work on the latest (“Bullet”) version of Raspbian OS. Perhaps I did something wrong; perhaps by now it’s all resolved. But I tried twice, and only “Buster” worked perfectly.
The Raspberry Pi Foundation is great, but it’s pretty frustrating that they’re constantly tweaking the network setup code and drivers. There. are so many web pages which refer to old versions and drivers. (One clue you’re looking at an outdated how-to: if you see mention of “/etc/network” or the “interfaces” file. This method is discontinued and you’ll see no mention of it below.)
If you are dedicating your Raspberry Pi to its own ad-hoc network, I highly recommend you start over with it, back up whatever files you have on the SD card, and then use the Raspbian OS imager to create a bootable card with a fresh copy of Raspbian OS “Buster” on it. So, grab an archived copy of Buster. Unzip it. Then, launch the Raspberry Pi Imager, and click the GEAR icon to:
ENABLE SSH with a password, and
Set it up with your home wifi’s network name (SSID) and password.
After your card is written, you should be able to pop it out and put it in your RPi, and turn on the device. After a few moments, you should be able to ssh into it (Mac) or use PuTTY (Windows) to connect. To find the device’s IP address, go into your wifi’s router and look for a recently joined Raspberry Pi device. (Generally, the IP address will be the same from session to session.)
For Mac OSX, this looks something like:
ssh pi@192.168.x.y
where x and y are numbers. Every Internet device connected to your wifi is assigned a different address; this will vary from device to device. The default password for “pi” is “raspberry”.
Later in these setup instructions, we’re going to have it create its own wifi network with its own subnet, and connect on the backend with Ethernet. This is called a “Wide Area Network Access Point” or “Wifi Access Point” configuration, in which the RPi will basically have two IP addresses: one over the wire, and the other wireless.
1. Set up “root” user
Instead of the default user “pi”, for some operations you may need root access. For instance, NeoPixel libraries need Raspberry Pi “root” permission to run. So it’s best to set up a root user password first thing:
This will create a folder in /var/www/html which the Apache web server will run. The install will also ensure that it starts at boot time.
3. Set up Ad Hoc Networking
OK this is the crucial part. HAVE PATIENCE, and follow this closely. (Full notes are at this Github location.)
Setting up a Routed Wireless Access Point
We will want to configure our Raspberry Pi as a Wireless Access Point, so that it broadcasts a wifi ssid, and so mobile devices can connect to it. Ideally, the RPi would also remain connectable to an Ethernet network for development purposes, to install new libraries, etc.
So a “Routed Wireless Access Point” is perfect for our needs.
AGAIN, BE SURE YOU ARE WORKING ON RASPBIAN “BULLET”, NOT “BULLSEYE” OR LATER.
You can find out what OS version you’re running with the following command:
A Raspberry Pi within an Ethernet network can be used as a wireless access point, creating a secondary network. The resulting new wireless network (called “Enchanted Rose” in my case) is entirely managed by the Raspberry Pi.
A routed wireless access point can be created using the inbuilt wireless features of the Raspberry Pi 4, Raspberry Pi 3 or Raspberry Pi Zero W, or by using a suitable USB wireless dongle that supports access point mode. It is possible that some USB dongles may need slight changes to their settings. If you are having trouble with a USB wireless dongle, please check the forums.
This documentation was tested on a Raspberry Pi Zero W2 running a fresh installation of Raspberry Pi OS Buster.
Before you Begin
Ensure you have root access to your Raspberry Pi. The network setup will be modified as part of the installation: local access, with screen and keyboard connected to your Raspberry Pi, is recommended.
Connect your Raspberry Pi to the Ethernet network and boot the Raspberry Pi OS.
Ensure the Raspberry Pi OS on your Raspberry Pi is up-to-date and reboot if packages were installed in the process.
Take note of the IP configuration of the Ethernet network the Raspberry Pi is connected to:
In this document, we assume IP network 192.164.4.0/24 is configured for the Ethernet LAN, and the Raspberry Pi is going to manage IP network 192.168.11.0/24 for wireless clients.
Please select another IP network for wireless, e.g. 192.168.10.0/24, if IP network 192.168.11.0/24 is already in use by your Ethernet LAN.
Have a wireless client (laptop, smartphone, …) ready to test your new access point.
Install Access Point and Management Software
In order to work as an access point, the Raspberry Pi needs to have the hostapd access point software package installed:
In order to provide network management services (DNS, DHCP) to wireless clients, the Raspberry Pi needs to have the dnsmasq software package installed:
Finally, install netfilter-persistent and its plugin iptables-persistent. This utility helps by saving firewall rules and restoring them when the Raspberry Pi boots:
Software installation is complete. We will configure the software packages later on.
Set up the Network Router
The Raspberry Pi will run and manage a standalone wireless network. It will also route between the wireless and Ethernet networks, providing internet access to wireless clients. If you prefer, you can choose to skip the routing by skipping the section “Enable routing and IP masquerading” below, and run the wireless network in complete isolation.
Define the Wireless Interface IP Configuration
The Raspberry Pi runs a DHCP server for the wireless network; this requires static IP configuration for the wireless interface (wlan0) in the Raspberry Pi. The Raspberry Pi also acts as the router on the wireless network, and as is customary, we will give it the first IP address in the network: 192.168.11.1.
To configure the static IP address, edit the configuration file for dhcpcd with:
This section configures the Raspberry Pi to let wireless clients access computers on the main (Ethernet) network, and from there the internet.
NOTE
If you wish to block wireless clients from accessing the Ethernet network and the internet, skip this section.
To enable routing, i.e. to allow traffic to flow from one network to the other in the Raspberry Pi, create a file using the following command, with the contents below:
Enabling routing will allow hosts from network 192.168.11.0/24 to reach the LAN and the main router towards the internet. In order to allow traffic between clients on this foreign wireless network and the internet without changing the configuration of the main router, the Raspberry Pi can substitute the IP address of wireless clients with its own IP address on the LAN using a “masquerade” firewall rule.
The main router will see all outgoing traffic from wireless clients as coming from the Raspberry Pi, allowing communication with the internet.
The Raspberry Pi will receive all incoming traffic, substitute the IP addresses back, and forward traffic to the original wireless client.
This process is configured by adding a single firewall rule in the Raspberry Pi:
Filtering rules are saved to the directory /etc/iptables/. If in the future you change the configuration of your firewall, make sure to save the configuration before rebooting.
Configure the DHCP and DNS services for the wireless network
The DHCP and DNS services are provided by dnsmasq. The default configuration file serves as a template for all possible configuration options, whereas we only need a few. It is easier to start from an empty file.
Rename the default configuration file and edit a new one:
interface=wlan0 # Listening interface
dhcp-range=192.168.11.2,192.168.11.20,255.255.255.0,24h
# Pool of IP addresses served via DHCP
domain=wlan # Local wireless DNS domain
address=/gw.wlan/192.168.11.1
# Alias for this router
The Raspberry Pi will deliver IP addresses between 192.168.4.2 and 192.168.4.20, with a lease time of 24 hours, to wireless DHCP clients. You should be able to reach the Raspberry Pi under the name gw.wlan from wireless clients.
NOTE
There are three IP address blocks set aside for private networks. There is a Class A block from 10.0.0.0 to 10.255.255.255, a Class B block from 172.16.0.0 to 172.31.255.255, and probably the most frequently used, a Class C block from 192.168.0.0 to 192.168.255.255.
There are many more options for dnsmasq; see the default configuration file (/etc/dnsmasq.conf) or the online documentation for details.
Ensure Wireless Operation
Note, I successfully skipped this subsection
Countries around the world regulate the use of telecommunication radio frequency bands to ensure interference-free operation. The Linux OS helps users comply with these rules by allowing applications to be configured with a two-letter “WiFi country code”, e.g. US for a computer used in the United States.
In the Raspberry Pi OS, 5 GHz wireless networking is disabled until a WiFi country code has been configured by the user, usually as part of the initial installation process (see wireless configuration pages in this section for details.)
To ensure WiFi radio is not blocked on your Raspberry Pi, execute the following command:
Add the information below to the configuration file. This configuration assumes we are using channel 7, with a network name of EnchantedRose, and a password AardvarkBadgerHedgehog. Note that the name and password should not have quotes around them, and the passphrase should be between 8 and 64 characters in length, or else “hostapd” will fail to start.
Note the line country_code=US: it configures the computer to use the correct wireless frequencies in the United States. Adapt this line and specify the two-letter ISO code of your country. See Wikipedia for a list of two-letter ISO 3166-1 country codes.
To use the 5 GHz band, you can change the operations mode from hw_mode=g to hw_mode=a. Possible values for hw_mode are:
a = IEEE 802.11a (5 GHz) (Raspberry Pi 3B+ onwards)
b = IEEE 802.11b (2.4 GHz)
g = IEEE 802.11g (2.4 GHz)
Note that when changing the hw_mode, you may need to also change the channel – see Wikipedia for a list of allowed combinations.
Troubleshooting hostapd: If for some reason your access point is not coming up, try running hostapd manually from the command line: sudo hostapd /etc/hostapd/hostapd.conf
You’ll likely get some kind of error message back.
Running the new Wireless AP
Now restart your Raspberry Pi and verify that the wireless access point becomes automatically available.
Once your Raspberry Pi has restarted, search for wireless networks with your wireless client. The network SSID you specified in file /etc/hostapd/hostapd.conf should now be present, and it should be accessible with the specified password.
If SSH is enabled on the Raspberry Pi, it should be possible to connect to it from your wireless client as follows, assuming the pi account is present: ssh pi@192.168.11.1.
If your wireless client has access to your Raspberry Pi (and the internet, if you set up routing), congratulations on setting up your new access point!
If you encounter difficulties, contact the forums for assistance. Please refer to this page in your message.
Once this is done, the network should be “EnchantedRose”, and the main web address should be 192.168.11.1.
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
This is the third in a three-part series introducing revolutionary changes in AI-generated art. In Part I: AI Art Comes of Age, we traced back through some of the winding path that brought us to this point. Part II: Deep Learning and The Diffusion Revolution, 2014-present, introduced three basic methods for generating art via deep-learning networks: GANs, VAEs and Diffusion models.
But what does it all mean? What’s at stake? In this final installment, let’s discuss some of the opportunities, legal and ethical questions presented by these new Engines of Wow.
Opportunities and Disruptions
We suddenly have robots which can turn text prompts into relevant, engaging, surprising images for pennies in a matter of seconds. They can compete with custom-created art taking illustrators and designers days or weeks to create.
Anywhere an image is needed, a robot can now help. We might even see side-by-side image creation with spoken words or written text, in near real-time.
Videogame designers have an amazing new tool to envision worlds.
Bloggers, web streamers and video producers can instantly and automatically create background graphics to describe their stories.
Graphic design firms can quickly make logos or background imagery for presentations, and accelerate their work. Authors can bring their stories to life.
Architects and storytellers can get inspired by worlds which don’t exist.
Entire graphic novels can now be generated from a text script which describes the scenes without any human intervention. (The stories themselves can even be created by new Chat models from OpenAI and other players.)
Storyboards for movies, which once cost hundreds of thousands of dollars to assemble, can soon be generated quickly, just by ingesting a script.
It’s already happening. In the Midjourney chat room, user Marcio84 writes: “I wrote a story 10 years ago, and finally have a tool to represent its characters.” With a few text prompts, the Midjourney Diffusion Engine created these images for him for just a few pennies:
Industrial designers, too, have a magical new tool. Inspiration for new vehicles can appear by the dozens and be voted up or down by a community:
Motorcycle concept generated by Midjourney, 2022
These engines are capable of competing with humans. In some surveys, as many as 87% of respondents incorrectly felt an AI-generated image was that of a real person. Think you can do better? Take the quiz.
I bet you could sell the art below, generated by Midjourney from a “street scene in the city, snow” prompt, in an art gallery or Etsy shop online. If I spotted it framed on a wall somewhere, or on a book cover or movie poster, I’d have no clue it was computer-generated:
A group of images stitched together becomes a video. One Midjourney user has tried to envision the aging destruction of a room, via successive videoframes of ever-more decaying description:
These are just a few of the things we can now do with these new AI art generation tools. Anywhere an image is useful, AI tools will have an impact, by lowering cost, blending concepts and styles, and envisioning many more options.
Where do images have a role? Well, that’s pretty much every field: architecture, graphic design, music, movies, industrial design, journalism, advertising, photography, painting, illustration, logo design, training, software, medicine, marketing, education and more.
Disruption
The first obvious impact is that many millions of employed or employable people may soon have far fewer opportunities.
Looking just at graphic design, approximately half a million designers employed globally, about 265,000 of whom are in the United States. (Source: Matt Moran of Colorlib.) The total market size for graphic design is about $43 billion per year. 90% of graphic designers work freelance, and the Fortune 500 accounts for nearly one-fifth of graphic design employment.
That’s just the graphic design segment. Then, there are photographers, painters, landscape architects and more.
But don’t count out the designers yet. These are merely tools, just as photography ones. And, while the Internet disrupted (or “disintermediated”) brokers in certain markets in the ’90s and ’00s (particularly in travel, in-person retail, financial services and real estate), I don’t expect that AI-generation tools means these experts are obsolete.
But the AI revolution is very likely to reduce the dollars available and reshape what their roles are. For instance, travel agents and financial advisers very much do still exist, though their numbers are far lower. The ones who have survived — even thrived — have used the new tools to imagine new businesses and have moved up the value-creation chain.
Who Owns the Ingested Data? Who Compensates the Artists?
Is this all plagiarism of sorts? There are sure to be lawsuits.
These algorithms rely upon massive image training sets. And there isn’t much dotting of i’s and crossing of t’s to secure digital rights. Recently, an artist found her own private medical records in one publicly available training dataset on the web which has been used by Stability AI. You can check to see if your own images have been part of the training datasets at www.haveibeenstrained.com.
But unlike most plagiarism and “derivative work” lawsuits up until about 2020, these lawsuits will need to contend with not being able to firmly identify just how the works are directly derivative. Current caselaw around derivative works generally requires some degree of sameness or likeness from input to final result. But the body of imagery which go into training the models is vast. A given creative end-product might be associated with hundreds of thousands or even millions of inputs. So how do the original artists get compensated, and how much, if at all?
No matter the algorithm, all generative AI models rely upon enormous datasets, as discussed in Part II. That’s their food. They go nowhere without creative input. And these datasets are the collective work of millions of artists and photographers. While some AI researchers go to great lengths to ensure that the images are copyright-free, many (most?) do not. Web scraping is often used to fetch and assemble images, and then a lot of human effort is put into data-cleansing and labeling.
The sheer scale of “original art” that goes into these engines is vast. It’s not unusual for a model to be trained by 5 million images. So these generative models learn patterns in art from millions of samples, not just by staring at one or two paintings. Are they “clones?” No. Are they even “derivative?” Probably, but not in the same way that George Harrison’s “My Sweet Lord” was derivative of Ronnie Mack’s “He’s So Fine.”
In the art world, American artist Jeff Koons created a collection called Banality, which featured sculptures from pop culture: mostly three dimensional representations of advertisements and kitsch. Fait d’Hiver (Fact of Winter) was one such work, which sold for approximately $4.3 million in a Christie’s auction in 2007:
Davidovici acknowledged that his sculpture was both inspired by and derived from this advertisement:
It’s plain to the eye the work is derivative.
And in fact, that was the whole point: Koons brought to three dimensions some of banality of everyday kitsch. In a legal battle spanning four years, Koons’ lawyers argued unsuccessfully that such derivative work was still unique, on several grounds. For instance, he had turned it into three dimensions, added a penguin and goggles on the woman, applied color, changed her jacket, the material representing snow, changed the scale, and much more.
While derivative, with all these new attributes, was the work not then brand new? The French court said non. Koons was found guilty in 2018. And it’s not the first time he was found guilty — in five lawsuits which sprang from this Banality collection, Koons lost three, and another settled out of court.
Unlike other “derivative works” lawsuits of the past, generative models in AI are relying not upon one work of a given artist, but an entire body of millions of images from hundreds of thousands of creators. Photographs are often lumped in with artistic sketches, oil paintings, graphic novel art and more to fashion new styles.
And, while it’s possible to look into the latent layers of AI models and see vectors of numbers, it’s impossible to translate that into something akin to “this new image is 2% based on image A64929, and 1.3% based on image B3929, etc.” An AI model learns patterns from enormous datasets, and those patterns are not well articulated.
Potential Approaches
It would be possible, it seems to me, to pass laws requiring that AI generative models use properly licensed (i.e., copyright-free or royalty-paid images), and then divvy up that royalty amongst its creators. Each artist has a different value for their work, so presumably they’d set the prices and AI model trainers would either pay for those or not.
Compliance with this is another matter entirely; perhaps certification technologies would offer valid tokens once verifying ownership. Similar to the blockchain concept, perhaps all images would have to be traceable to some payment or royalty agreement license. Or perhaps the new technique of Non Fungible Tokens (NFTs) can be used to license out ownership for ingestion during the training phase. Obviously this would have to be scalable, so it suggests automation and a de-facto standard must emerge.
Or will we see new kinds of art comparison or “plagiarism” tools, letting artists compare similarity and influence between generated works and their own creation? Perhaps if a generated piece of art is found to be more than 95% similar (or some such threshold) to an existing work, it will not retain copyright and/or require licensing of the underlying work. It’s possible to build such comparative tools today.
In the meantime, it’s the Wild West of sorts. As has often happened in the past, technology’s rapid pace of advancement has gotten ahead of where legislation and currency and monetary flow is.
What’s Ahead
If you’ve come with me on this journey into AI-generated art in 2022, or have seen these tools up close, you’re like the viewer who’s seen the world wide web in 1994. You’re on the early end of a major wave. This is a revolution in its nascent stage. We don’t know all that’s ahead, and the incredible capabilities of these tools are known only to a tiny fraction of society at the moment. It is hard to predict all the ramifications.
But if prior disintermediation moments are any guide, I’d expect change to happen along a few axes.
First, advancement and adoption will spread from horizontal tools to many more specialized verticals. Right now, there’s great advantage to being a skilled “image prompter.” I suspect that like photography, which initially required extreme expertise to create even passing mundane results, the engines will get better at delivering remarkable images the first pass through. Time an again in technology, generalized “horizontal” applications have concentrated into an oligopoly market of a few tools (e.g., spreadsheets), yet also launched a thousand flowers in much more specialized “vertical” ones (accounting systems, vertical applications, etc.) I expect the same pattern here. These tools have come of age, but only a tiny fraction of people know about them. We’re still in the generalist period. Meanwhile, they’re getting better and better. Right now, these horizontal applications stun with their potential, and are applicable to a wide variety of uses. But I’d expect thousands of specialty, domain specific applications and brand-names to emerge (e.g., book illustration, logo design, storyboard design, etc.) One set of players might generate sketches for creating book covers. Another for graphic novels. Another might specialize in video streaming backgrounds. Not only will this make the training datasets much more specific and outputs even more relevant, but it will allow the brand-names of each engine to better penetrate a specific customer base and respond to their needs. Applications will emerge for specific domains (e.g., automated graphic design for, say, blog posts.)
Second, many artists will demand compensation and seek to restrict rights to their work. Perhaps new guilds will emerge. A new technology and payments system might likely emerge to allow this to scale. Content generally has many ancillary rights, and one of those rights will likely be “ingestion rights” or “training model rights.” I would expect micropayment solutions or perhaps some form of blockchain-based technology to allow photographers, illustrators and artists to protect their work from being ingested into models without compensation. This might emerge as some kind of paywall approach to individual imagery. As is happening in the world of music, the most powerful and influential creative artists may initiate this trend, by cordoning off their entire collective body of work. For instance, the Ansel Adams estate might decide to disallow all ingestion into training models; right now however it’s very difficult to prove whether or not those images were used in the training of any datasets.
Third, regulation might be necessary to protect vital creative ecosystems. If an AI generative machine is able to create works auctioned at Christie’s for $5 million, and it may well soon, what does this do to the ecosystem of creators? It is likely necessary for regulators to protect the ecosystem for creators which feeds derivative engines, restricting AI generative model-makers from just fetching and ingesting any old image.
Fourth, in the near-term, skilled “image prompters” are like skilled photographers, web graphic designers, or painters. Today, there is a noticeable difference between those who know how to get the most of these new tools from those who do not. For the short term, this is likely to “gatekeep” this technology and validate the expertise of designers. I do not expect to this to be extremely durable, however; the quality of output from very unskilled prompters (e.g., yours truly) now meets or exceeds a lot of royalty-free art that’s out there from the likes of Envato or Shutterstock.
Conclusion
Machines now seem capable of visual creativity. While their output is often stunning, under the covers, they’re just learning patterns from data and semi-randomly assembling results. The shocking advancements since just 2015 suggest much more change is on the way: human realism, more styles, video, music, dialogue… we are likely to see these engines pass the artistic “Turing test” across more dimensions and domains.
For now, you need to be plugged into geeky circles of Reddit and Discord to try them out. And skill in crafting just the right prompts separates talented jockeys from the pack. But it’s likely that the power will fan out to the masses, with engines of wow built directly into several consumer end-user products and apps over the next three to five years.
We’re in a new era, where it costs pennies to envision dozens of new images visualizing anything from text. Expect some landmark lawsuits to arrive soon on what is and is not derivative work, and whether such machine-learning output can even be copyrighted. For now, especially if you’re in a creative field, it’s good advice to get acquainted with these new tools, because they’re here to stay.
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
A revolutionary insight in 2015, plus AI work on natural language, unleashed a new wave of generative AI models.
In Part I of this series on AI-generated art, we introduced how deep learning systems can be used to “learn” from a well-labeled dataset. In other words, algorithmic tools can “learn” patterns from data to reliably predict or label things. Now on their way to being “solved” via better and better tweaks and rework, these predictive engines are magical power-tools with intriguing applications in pretty much every field.
Here, we’re focused on media generation, specifically images, but it bears a note that many of the same basic techniques described below can apply to songwriting, video, text (e.g., customer service chatbots, poetry and story-creation), financial trading strategies, personal counseling and advice, text summarization, computer coding and more.
Generative AI in Art: GANs, VAEs and Diffusion Models
From Part I of this series, we know at a high level how we can use deep-learning neural networks to predict things or add meaning to data (e.g., translate text, or recognize what’s in a photo.) But we can also use deep learning techniques to generate new things. This type of neural network system, often comprised of multiple neural networks, is called a Generative Model. Rather than just interpreting things passively or searching through existing data, AI engines can now generate highly relevant and engaging new media.
How? The three most common types of Generative Models in AI are Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs) and Diffusion Models. Sometimes these techniques are combined. They aren’t the only approaches, but they are currently the most popular. Today’s star products in art-generating AI are Midjourney by Midjourney.com (Diffusion-based) DALL-E by OpenAI (VAE-based), and Stable Diffusion (Diffusion-based) by Stability AI. It’s important to understand that each of these algorithmic techniques were conceived just in the past 6 years or so.
My goal is to describe these three methods at a cocktail-party chat level. The intuition behind them are incredibly clever ways of thinking about the problem. There are lots of resources on the Internet which go much further into each methodology, listed at the end of each section.
Generative Adversarial Networks
The first strand of generative-AI models, Generative Adversarial Networks (GANs), have been very fruitful for single-domain image generation. For instance, visit thispersondoesnotexist.com. Refresh the page a few times.
Each time, you’ll see highly* convincing images like this, but never the same one twice:
As the domain name suggests, these people do not exist. This is the computer creating a convincing image, using a Generative Adversarial Network (GAN) trained to construct a human-like photograph.
*Note that for the adult male, it only rendered half his glasses. This GAN doesn’t really understand the concept of “glasses,” simply a series of pixels that need to be adjacent to one another.
Generative Adversarial Networks were introduced in a 2014 paper by Ian Goodfellow et al. That was just eight years ago! The basic idea is that you have two deep-learning neural networks: a Generator and a Discriminator. You can think of them like a Conterfeiter and a Detective respectively. One Deep Learning model, serving as the “Discriminator” (Detective), learns to distinguish between genuine articles and counterfeits. It penalizes the generator for producing implausible results. Meanwhile, a Generator model learns to “generate” plausible data, which, if it “fools” the discriminator, becomes negative training data for the Discriminator. They play a zero-sum game against each other (thus it’s “adversarial”) thousands and thousands of times, and with each adjustment to the Generator and Discriminator’s weights and attributes, the Generator gets better and better at “learning” how to construct something to fool the Discriminator, and the Discriminator gets better and better at detecting fakes.
GANs have delivered pretty spectacular results, but in fairly narrow domains. For instance, GANs have been pretty good at mimicking artistic styles (called “Neural Style Transfer“) and Colorizing Black and White Images.
GANs are cool and a major area of generative AI research.
An encoder can be thought of as a compressor of data, and a decompressor, something which does this opposite. You’ve probably compressed an image down to a smaller size without losing recognizability. It turns out you can use AI models to compress an image. Data scientists call this reducing its dimensionality.
What if you built two neural network models, an Encoder and a Decoder? It might look like this, going from x, the original image, to x’, the “compressed and then decompressed” image:
Variational Autoencoder, high-level diagram. Images go in on left, and come out on right. If you train. the networks to minimize the difference between output and input, you get to a compression algorithm of sorts. What’s left in red are lower-dimension representation of the images.
So conceptually, you could train an Encoder neural network to “compress” images into vectors, and then a Decoder neural network to “decompress” the image back into something close to the original.
Then, you could consider the red “latent space” in the middle as basically the rosetta stone for what a given image means. Run that algorithm numerous times over multiple images, encoding it with the text of the labeled images, and you would end up with the condensed encoding of how to render various images. If you did this across many, many images and subjects, these numerous red vectors would overlap in n-dimensional space, and could be sampled and mixed and then run through the decoder to generate images.
With some mathematical tricks (specifically, forcing the latent variables in red to conform to a normal distribution), you can build a system which can generate images that never existed before, but which have some very similar properties to the dataset which was used to train the encoder.
Is there another method entirely? What else could you do with a deep learning system which can “learn” how to predict things?
In March 2015, a revolutionary paper came out from researchers Sohl-Dickstein, Weiss, Maheswaranathan and Ganguli. It was inspired by the physics of non-equilibrium systems: for instance, dropping a drop of food coloring into a glass of water. Imagine you saw a film of that process of “destruction”, and could stop it frame by frame. Could you build a neural network to reliably predict what a reverse might look like?
Let’s think about a massive training set of animal images. Imagine you take an image in your training dataset, and create multiple copies of the image, each time systematically adding graphic “noise” to it. Step by step, more noise is added to your image (x), via what mathematicians call a Markov chain (incremental steps.) You apply a normally-distributed distortion, let’s say, Gaussian Blur.
In a forward direction, from left to right, it might look something like this. At each step from left to right, you’re going from data (the image) to pure noise:
But here’s the magical insight behind Diffusion models. Once you’ve done this, what if you trained a deep learning model to try to predict frames in the reverse direction? Could you predict a “de-noised” image X(t) from its more noisier version, X(t+1)? Could you could read each step backward, from right to left, and try to predict the best way to remove noise at each step?
This was the insight in the 2015 paper, albeit with much more mathematics behind it. It turns out you can train a deep learning system to learn how to “undo” noise in an image, with pretty good results. For instance, if you input the pure-noise image in the last step, x(T), and train a deep learning network that its output should be the previous step x(T-1), and do this over and over again with many images, you can “train” a deep learning network to subtract noise in an image, all the way back to an original image.
Do this enough times, with enough terrier images, say. And then, ask your trained model to divine a “terrier” from random noise. Gradually, step by step, it removes noise from an image to synthesize a “terrier”, like this:
Screen captured video of using the Midjourney chatroom (on Discord) to generate: “terrier, looking up, cute, white background”
Images generated from the current Midjourney model:
“terrier looking up, cute, white background” entered into Midjourney. Unretouched, first-pass output with v3 model.
Wow! Just slap “No One Hates a Terrier” on any of these images above, print 100 t-shirts, and sell it on Amazon. Profit! I’ll touch on some of the legal and ethical controversies and ramifications in the final post in this series.
Training the Text Prompts: Embeddings
How did Midjourney know to produce a “terrier”, and not some other object or scene or animal?
This relied upon another major parallel track in deep learning: natural language processing. In particular, word “embeddings” can be used to get from keywords to meanings. And during the image model training, these embeddings were applied by Midjourney to enhance each noisy-image with meaning.
An “embedding” is a mapping of a chunk of text into a vector of continuous numbers. Think about a word as a list of numbers. A textual variable could be a word or a node in a graph, or a relation between nodes in a graph. By ingesting massive amounts of text, you can train a deep learning network to understand relationships between words and entities, and numerically pull out how closely associated some words and phrases are with others. They can be used to cluster together the sentiment of an expression in mathematical terms a computer can appear to understand. For instance, embedding models are now able to interpret semantics and relationships between words, like “royalty + woman – man = queen.”
An example on Google Colab took a vocabulary of 50,000 words in a collection of movie reviews, and learned over 100 different attributes from words used with them, based on their adjacency to one another:
So, if you simultaneously injected into the “de-noising” diffusion-based learning process the information that this is about a “dog, looking up, on white background, terrier, smiling, cute,” you can get a deep learning network to “learn” how to go from random noise (x(T)) to a very faint outline of a terrier (x(T-1)), to even less faint (x(T-2)) and so on, all the way back to x(0). If you do this over thousands of images, and thousands of keyword embeddings, you end up with a neural network that can construct an image from some keywords.
Incidentally, researchers have found that about T=1000 is about all you need in this process, but millions of input images and enormous amounts of computing power are needed to learn how to “undo” noise at high resolution.
Let’s step back a moment to note that this revelation about Diffusion Models was only really put forward in 2015, and improved upon in 2018 and 2020. So we are just at the very beginning of understanding what might be possible here.
In 2021, Dhariwal and Nichol convincingly note that diffusion models can achieve image quality superior to the existing state-of-the-art GAN models.
Up next, Part III: Ramifications and Questions
That’s it for now. In the final Part III of Engines of Wow, we’ll explore some of the ramifications, controversies and make some predictions about where this goes next.
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
Advancements in AI-generated art test our understanding of human creativity and laws around derivative art.
While most of us were focused on Ukraine, the midterm elections, or simply returning to normal as best we can, Artificial Intelligence (AI) took a gigantic leap forward in 2022. Seemingly all of a sudden, computers are now eerily capable of human-level creativity. Natural language agents like GPT-3 are able to carry on an intelligent conversation. GitHub CoPilot is able to write major blocks of software code. And new AI-assisted art engines with names like Midjourney, DALL-E and Stable Diffusion delight our eyes, but threaten to disrupt entire creative professions. They raise important questions about artistic ownership, derivative work and compensation.
In this three-part blog series, I’m going to dive in to the brave new world of AI-generated art. How did we get here? How do these engines work? What are some of the ramifications?
[featured image above: “God of Storm Clouds” created by Midjourney AI algorithm]
But first, why should we care? What kind of output are we talking about?
Let’s try one of the big players, the Midjourney algorithm. Midjourney lets you play around in their sandbox for free for about 25 queries. You can register for free at Midjourney.com; they’ll invite you to a Discord chat server. After reading a “Getting Started” agreement and accepting some terms, you can type in a prompt. You might go with: “/imagine portrait of a cute leopard, beautiful happy, Gryffindor outfit, super detailed, hyper realism.”
Wait about 60 seconds, choose one of the four samples generated for you, click the “upscale” button for a bigger image, and voila:
image created by the Midjourney image generation engine, version 4.0. Full prompt used to create it was “portrait of a cute leopard, Beautiful happy, Gryffindor Outfit, white background, biomechanical intricate details, super detailed, hyper realism, heavenly, unreal engine, rtx, magical lighting, HD 8k, 4k”
The Leopard of Gryffindor was created without any human retouching. This is final Midjourney output. The algorithm took the text prompt, and then did all the work.
I look at this image, and I think: Stunning.
Looking at it, I get the kind of “this changes everything” feeling, like the first time I browsed the world-wide web, spoke to Siri or Alexa, rode in an electric vehicle, or did a live video chat with friends across the country for pennies. It’s the kind of revolutionary step-function that causes you to think “this will cause a huge wave of changes and opportunities,” though it’s not even clear what they all are.
Are artists, graphic designers and illustrators doomed? Will these engines ultimately help artists or hurt them? How will the creative ecosystem change when it becomes nearly free to go from idea to visual image?
Once mainly focused at just processing existing images, computers are now extremely capable at generating brand new things. Before diving into a high-level overview of these new generative AI art algorithms, let me emphasize a few things: First, no artist has ever created exactly the above image before, nor will it likely be generated again. That is, Midjourney and its competitors (notably DALL-E and Stable Diffusion) aren’t search engines: they are media creation engines.
In fact, if you typed this same exact prompt into Midjourney again, you’ll get an entirelydifferent image, yet one which also is likely to deliver on the prompt fairly well.
There is an old joke within Computer Science circles that “Artificial Intelligence is what we call things that aren’t working yet.” That’s now sounding quaint. AI is all around us, making better and better recommendations, completing sentences, curating our media feeds, “optimizing” the prices of what we buy, helping us with driving assistance on the road, defending our computer networks and detecting spam.
Part I: The Artists in the Machine, 1950-2015+
How did this revolutionary achievement come about? Two ways, just as bankruptcy came about for Mike Campbell in Hemingway’s The Sun Also Rises: First gradually. Then suddenly.
Computer scientists have spent more than fifty years trying to perfect art generation algorithms. These five decades can be roughly divided into two distinct eras, each with entirely different approaches: “Procedural” and “Deep Learning.” And, as we’ll see in Part II, the Deep-Learning era had three parallel but critical deep learning efforts which all converged to make it the clear winner: Natural Language, Image Classifiers, and Diffusion Models.
But first, let’s rewind the videotape. How did we get here?
Procedural Era: 1970’s-1990’s
If you asked most computer users, the naive approach to generating computer art would be to try to encode various “rules of painting” into software programs, via the very “if this then that” kind of logic that computers excel at. And that’s precisely how it began.
In 1973, American computer scientist Harold Cohen, resident at Stanford University’s Artificial Intelligence Laboratory (SAIL) created AARON, the first computer program dedicated to generating art. Cohen was actually an accomplished, talented artist and a computer scientist. He thought it would be intriguing to try to “teach” a computer how to draw and paint.
His thinking was to encode various “rules about drawing” into software components and then have them work together to compose a complete piece of art. Cohen relied upon his skill as an exceptional artist, and coded his own “style” into his software.
AARON was an artificial intelligence program first written in the C programming language (a low level language compiled for speed), and later LISP (a language designed for symbolic manipulation.) AARON knew about various rules of drawing, such as how to “draw a wavy blue line, intersecting with a black line.” Later, constructs were added to combine these primitives together to “draw an adult human face, smiling.” By 1995, Cohen added rules for painting color within the drawn lines.
Though there were aspects of AARON which were artificially intelligent, by and large computer scientists call his a procedural approach. Do this, then that. Pick up a brush, pick an ink color, and draw from point A to B. Construct an image from its components. Join the lines. And you know what? after a few decades of work, Cohen created some really nice pieces, worthy of hanging on a wall. You can see some of them at the Museum of Computer History in Menlo Park, California.
Just a few months ago, other attempts at AI-generated art were flat-looking and derivative, like this image from 2019:
Twenty seven years after AARON’s first AI-generated color painting, algorithms like Midjourney would be quickly rendering photorealistic images from text prompts. But to accomplish it, the primary method is completely different.
Deep Learning Era (1986-Present)
Algorithms which can create photorealistic images-on-demand are the culmination of multiple parallel academic research threads in learning systems dating back several decades.
We’ll get to the generative models which are key to this new wave of “engines of wow” in the next post, but first, it’s helpful to understand a bit about their central component: neural networks.
Since about 2000, you have probably noticed everyday computer services making massive leaps in predictive capabilities; that’s because of neural networks. Turn on Netflix or YouTube, and these services will serve up ever-better recommendations for you. Or, literally speak to Siri, and she will largely understand what you’re saying. Tap on your iPhone’s keyboard, and it’ll automatically suggest which letters or words might follow.
Each of these systems rely upon trained prediction models built by neural networks. And to envision them, a branch of computer scientists and mathematicians had to radically shift their thinkingfrom the procedural approach. A branch of them did so first in the 1950’s and 60’s, and then again in a machine-learning renaissance which began in earnest in the mid-1980’s.
The key insight: these researchers speculated that instead of procedural coding, perhaps something akin to “intelligence” could be fashioned from general purpose software models, which would algorithmically “learn” patterns from a massive body of well-labeled training data. This is the field of “machine learning,” specifically supervised machine learning, because it’s using accurately pre-labeled data to train a system. That is, rather than “Computer, do this step first, then this step, then that step”, it became “Computer: learn patterns from this well-labeled training dataset; don’t expect me to tell you step-by-step which sequence of operations to do.”
The first big step began in 1958. Frank Rosenblatt, a researcher at Cornell University, created a simplistic precursor to neural networks, the “Perceptron,” basically a one-layer network consisting of visual sensor inputs and software outputs. The Perceptron system was fed a series of punchcards. After 50 trials, the computer “taught” itself to distinguish those cards which were marked on the left from cards marked on the right. The computer which ran this program was a five-ton IBM 704, the size of a room. By today’s standards, it was an extremely simple task, but it worked.
A single-layer perceptron is the basic component of a neural network. A perceptron consists of input values, weights and a bias, a weighted sum and activation function:
Frank Rosenblatt and the Perceptron system, 1958
Rosenblatt described it as the “first machine capable of having an original idea.” But the Perceptron was extremely simplistic; it merely added up the optical signals it detected to “perceive” dark marks on one side of the punchcard versus the other.
In 1969, MIT’s Marvin Minsky, whose father was an eye surgeon, wrote convincingly that neural networks needed multiple layers (like the optical neuron fabric in our eyes) to really do complex things. But his book Perceptrons, though well-respected in hindsight, got little traction at the time. That’s partially because during these intervening decades, the computing power required to “learn” more complex things via multi-layer networks were out of reach computationally. But time marched on, and over the next three decades, computing power, storage, languages and networks all improved dramatically.
From the 1950’s through the early 1980’s, many researchers doubted that computing power would be sufficient for intelligent learning systems via a neural network style approach. Skeptics also wondered if models could ever get to a level of specificity to be worthwhile. Early experiments often “overfit” the training data and simply output the input data. Some would get stuck on local maxima or minima from a training set. There were reasons to doubt this would work.
And then, in 1986, Carnegie Mellon Professor Geoffrey Hinton, whom many consider the “Godfather of Deep Learning” (go Tartans!), demonstrated that “neural networks” could learn to predict shapes and words by statistically “learning” from a large, labeled dataset. Hinton’s revolutionary 1986 breakthrough was the concept of “backpropagation.” This adds both multiple layers to the model (hidden layers), and also iterations through the neural network model using the output of one or more mathematical functions to adjust weights to minimize “loss” or distance from the expected output.
This is rather like the golfer who adjusts each successive golf swing, having observed how far off their last shots were. Eventually, with enough adjustments, they calculate the optimal way to hit the ball to minimize its resting distance from the hole. (This is where terms like “loss function” and “gradient descent” come in.)
In 1986-87, around the time of the 1986 Hinton-Rumelhart-Williams paper on Backpropagation, the whole AI field was in flux between these procedural and learning approaches, and I was earning a Masters in Computer Science at Stanford, concentrating in “Symbolic and Heuristic Computation.” I had classes which dove into the AARON-style type of symbolic, procedural AI, and a few classes touching on neural networks and learning systems. (My masters thesis was in getting a neural network to “learn” how to win the Tower of Hanoi game, which requires apparent backtracking to win.)
In essence, you can think of a neural network as a fabric of software-represented units (neurons) waiting to soak up patterns in data. The methodology to train them is: “here is some input data and the output I expect, learn it. Here’s some more input and its expected output, adjust your weights and assumptions. Got it? Keep updating your priors. OK, let’s keep doing that.” Like a dog learning what “sit” means (do this, get a treat / don’t do this, don’t get a treat), neural networks are able to “learn” over iterations, by adjusting the software model’s weights and thresholds.
Do this enough times, and what you end up with is a trained model that’s able to “recognize” patterns in the input data, outputting predictions, or labels, or anything you’d like classified.
A neural network, and in particular the special type of multi-layered network called a deep learning system, is “trained” on a very large, well-labeled dataset (i.e., with inputs and correct labels.) The training process uses Hinton’s “backpropagation” idea to adjust the weights of the various neuron thresholds in the statistical model, getting closer and closer to “learning” the underlying pattern in the data.
For much more detail on Deep Learning and the mathematics involved, see this excellent overview:
Deep Learning Revolutionizes AI Art
We’ll rely heavily upon this background of neural networks and deep learning for Part II: The Diffusion Revolution. The AI revolution uses deep learning networks to interpret natural language (text to meaning), classify images, and “learns” how to synthetically build an image from random noise.
Gallery
Before leaving, here are a few more images created from text prompts on Midjourney:
God of storm clouds over the horizon, Midjourney
All these images were created via Midjourney engine
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
After much thought, I’ve decided to wind down the video celebration app I created, HipHip.app.
After much thought, I’ve decided to wind down the video celebration app I created, HipHip.app.
All servers will be going offline shortly.
Fun Project, Lots of Learning
I started HipHip as a “give back” project during COVID. I noticed that several people were lamenting online that they were going to miss big milestones in-person: celebrations, graduations, birthdays, anniversaries, and memorials. I had been learning a bunch about user-video upload and creation, and I wanted to put those skills to use.
I built HipHip.app, a celebration video creator. I didn’t actually know at the time that there were such services — and it turns out, it’s a pretty crowded marketplace!
While HipHip delivered hundreds of great videos for people in its roughly two years on the market, it struggled to be anything more than a hobby/lifestyle project. It began under the unique circumstances of lockdown, helping people celebrate. That purpose was well served!
Now that the lockdown/remote phase of COVID is over, the economics of the business showed that it’s unlikely to turn into a self-sustaining business any time soon. There are some category leaders that have really strong search engine presence which is pretty expensive to dislodge.
I want to turn my energies to other projects, and free up time and budget for other things. COVID lockdown is over, and a lot of people want a respite from recording and Zoom-like interactions, including me.
It was a terrific, educational project. It kept me busy, learning, and productive. HipHip delivered hundreds of celebration videos for people around the world.
I’ve learned a ton about programmatic video creation, technology stacks like Next.js, Azure and React, and likely will apply these learnings to new projects, or perhaps share them with others via e-learning courses.
Among the videos people created were graduation videos, videos to celebrate new babies, engagement, birthdays, anniversaries and the attainment of US Citizenship.
In the end, the CPU processing and storage required for online video creation meant that it could not be done for free forever, and after testing a few price-points, there seems only so much willingness to pay in a crowded market.
Thanks everyone for your great feedback and ideas!
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
For less than $100, a little coding and a small circuit, you can remotely monitor a propane tank from anywhere.
One great use-case for the Internet of Things is to remotely monitor gauges and levels. In addition to spot-checks, wiring them up also allows you to auto-log values over time, so you can get a nice graph of usage, economize where you can, and project future usage.
If your property is served by liquid propane gas (LPG), there might be occasions where you want to be able to check the tank’s propane level from your phone. Ideally, you might even want it to notify you when the tank is, say, at 25% or lower, so you can schedule a refill. And wouldn’t it be useful to see which seasons, use-cases or appliances tend to consume the most propane?
There are commercial vendors for remote propane monitoring, but they’re usually tied to some kind of subscription service, some of them come bundled with a mandatory cellphone plan, and often tied to the fuel vendor themselves. I wasn’t interested in paying, say, $5 per month just to be able to check a value.
This made a perfect use-case for a simple microcontroller project.
WARNING: Fiddling with electricity, even very low 1-3V such as what we’re talking about here, around combustible material is a very bad idea. Only do this if you know what you are doing. Be sure all your connections are fully sealed and weatherproofed. Maker assumes all risk.
Ready to accept that risk? OK, here’s how I built it:
First, Some Background: How do Propane Gauges work?
At first, I thought that these gauges somehow have a physical pass-through into the tank, but they’re much more clever than that. They make use of magnets to keep the integrity of the tank fully sealed.
There’s a very good video here from DPS Telecom (a vendor of remote monitoring solutions for businesses and homes) which explains how such fuel gauges work:
So to proceed, you’ll need a gauge for your tank which is “Remote Ready.” This has been standardized in America, and you’ll want what’s called an “R3D compatible” gauge. They can be had for about $9-15.
This simply means that the gauge both reads and conveys a magnetic value. As the needle moves, a magnet also moves, which can be read by a Hall Effect Sensor placed just over the needle (under the black plastic.) Luckily, this standardization means that these gauges will almost certainly work on your existing tank, simply by removing two screws, removing the existing gauge, and screwing the new one in. Or, if your existing gauge says it’s “remote ready,” as many do, you’re already set.
But in looking for a pre-made “reader” cable, not only are they hard to find and rarely sold separately, I was shocked to see several vendors asking $150+ for them! You don’t need a commercial reader cable — you can make your own, with a Hall Effect sensor chip which you can find for about $1.
Building Our Remote Gauge
OK, now that we know how these gauges work, how can we get the Raspberry Pi to “read” it? We need to translate that magnetic signal into some kind of voltage. Once that’s done, since RPi doesn’t have a native analog to digital converter built in (as the Arduino does), you’ll also need some kind of Analog to Digital Converter (ADC) chip, so that RPi can make use of the signal in digital form. I chose the MCP 3008 for this project, which you can find on Amazon for about $9.
Yes, that’s Scotch tape on the prototype gauge. I’ll likely be epoxying it into place after tests are complete.
Experienced makers will no doubt notice that the Raspberry Pi 4, a full-fledged Linux computer, is extreme overkill for this application. But I’ll be using that same Pi to handle several other home automation tasks; this is just one of them.
Among the things to notice is that this MCP3008 chip can convert up to 8 analog voltage inputs. This project uses one of the 8. I’ll likely be using this same setup to remotely monitor things like humidity, temperature, soil moisture, etc. over time.
Wiring Diagram
Figure 1. Wiring Diagram
Steps
Get Raspberry Pi configured with Raspbian
Configure it for headless development via SSH. I used Visual Studio Code and its ssh remote development feature.
Make circuit: Use the MCP3008 to take an analog input and convert it into a digital signal. See wiring diagram above. [Addendum from diagram: I put a 10k Ohm “pull down” resistor connecting the analog input data pin (MCP3008 channel 0) with ground. This is optional, but it tended to steady the values a bit more.]
Solder Hall Effect Sensor 49E chip to 3 wires. Check the 49E datasheet and connect a wire for VCC, Ground and Voltage Out.
Crucial step: Position the Hall Effect Sensor Chip exactly in the small channel, with the smaller side of the trapezoid facing DOWN (i.e., toward the gauge). With my 49E chip, this means the wires, from top to bottom, are VCC, GROUND and SIGNAL.
Using the Raspberry Pi, take measurements at key values. This will vary considerably based upon how long your wires are, and where precisely you have the Hall Effect Sensor placed. My own values are shown in Table 1 below.
Install the apache2 server and pretty NextJS/React app to show the gauge.
Make sure the simple flask application (which is the API that the NextJS front-end calls) starts at RPi boot time. I’ve used gunicorn for this. I’ve set a value in the /etc/rc.local file for it to run at startup.
Table 1. The values I got, once I secured the Hall Effect Sensor (49E) in place
Code
I’m putting some finishing touches on this code and will provide a link to the GitHub repository shortly. There are three main parts:
gauges.py: A library to spot-read a voltage and spit back out the gauge value
server.py: the flask API
frontend: a simple NextJS app which reads the API and presents a nice looking gauge to the user
Learnings
The trickiest part is actually affixing the Hall Sensor to the gauge. Good results are extremely sensitive to precisely where this chip is mounted, and there’s no easy way to lock it in place. I chose a hot glue gun method, but I might revise that to some kind of epoxy.
The 49E Hall Sensor has a very narrow sensitivity range — highest values it ever reported were around 2.08v and lowest value around 1.3v. But that’s enough fidelity to translate it into a percent-full between 100 and 0. Here, it’s good to remember not to let the perfect be the enemy of the good. I don’t actually need single-digit precision; in this use-case, accuracy to within about 5% is more than adequate.
I’m still determining whether or not the RPi is going to live outside. My preference is that it not live outside, and that only the 3-wire strand goes out to the gauge from the RPi. If it does have to live outside, it will be powered by Power Over Ethernet (POE) to keep cables to a minimum. I’ve purchased a POE splitter which will live in the external enclosure if so. This approach will allow just one cable to serve both power and data (ethernet.) But for weather-proofing reasons, I’d rather have the RPi mounted inside (in, say, a garage), and have a long-ish low voltage wire going out to the gauge. I’ve experimented with cable lengths and thus far can confirm that about 15 feet still works very well, even at the low voltages we are talking about. I don’t know if the voltage drop will be too significant beyond that length.
Conclusion
I’ll be adding in a simple email notification system when the value gets below a certain value, and I’ll likely connect the sensor values to something like Azure IoT services to log the levels over time. This will provide nice usage graphs, and even help show if there’s a leak somewhere that needs to be addressed. Hope this technote is a useful starting point for your next RPi project!
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
Its purpose is to let you quickly get the pulse of what Seattleites are writing and talking about. It rolls up headlines and commentary from more than twenty local sources from across the political spectrum.
Seattle isn’t a monoculture. It’s a vibrant mix of many different people, many communities, neighborhoods, coalitions and voices. But there are also a lot of forces nudging people into filtered silos. I wanted to build a site which breaks away from that.
Day to day, I regularly hit a variety of news feeds and listen to a lot of different news sources. I wanted to make that much easier for myself and everyone in the city.
Seattlebrief.com is a grab-a-cup-of-coffee site. It is designed for browsing, very intentionally, not search. Click on the story you’re interested in, and the article will open up in a side window. It focuses on newsfeeds which talk about civic and municipal issues over sports, weather and traffic.
I’ll consider Seattlebrief.com a success if it saves you time catching up on local stories, or introduces you to more voices and perspectives in this great city.
How it works
There are so many interesting and important voices out there, from dedicated news organizations like The Seattle Times to more informal ones like neighborhood blogs. I wanted a quick way to get the pulse of what’s happening. Seattlebrief pulls from the RSS feeds of more than twenty local sites, from all sides of the political spectrum: news sites, neighborhood blogs, municipal government announcements, and activist organizations. The list will no doubt change over time.
Many blog sites and news organizations support Really Simple Syndication (RSS) to publish their latest articles for syndication elsewhere. For instance, you can find Post Alley’s RSS feed here. RSS is used to power Google News and podcast announcements, among other things.
RSS is a bunch of computer code which tells aggregation sites: “here are the recent stories,” usually including a photo thumbnail, author information, and description. Seattlebrief uses this self-declared RSS feed, currently from over 20 sources in and around Seattle. It regularly checks what’s new. Another job then fetches each page and “enriches” these articles with the social sharing metadata that is used to mark up the page for, say, sharing on Facebook or Twitter.
Think of it as a robot that simply goes out to a list of sites and fetches the “social sharing” info for each of them, then puts them in chronological order (by way of publication date) for you. The list of sites Seattlebrief uses will no doubt change over time.
Origin
Over at Post Alley, where I sometimes contribute, there was a writers’ room discussion about the Washington Post’s popular “Morning Mix” series. Morning Mix highlights interesting/viral stories around the web.
Sparked by that idea, I wanted to build a way to let me browse through the week’s Seattle-area headlines and commentary more easily. So I built Seattlebrief.
I’d welcome any key sources I’ve missed. Right now, they must have an RSS feed. And regrettably, some important/thoughtful voices like KUOW have long ago decommissioned their RSS feeds. I’m exploring what might be possible there.
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
Musk decides not to join Twitter’s board, which likely portends poison pill adoption by Twitter, and potentially moves toward greater stake by Musk.
It’s been a busy weekend for Twitter’s CEO and Board. Sunday afternoon, Twitter’s chief executive Parag Agrawal announced that Elon Musk would not be joining its board of directors. That morning, Musk had rejected the board’s offer to join the board in exchange for capping his stake at 14.9% and no doubt other legal restrictions around what he can and cannot say publicly about Twitter.
A battle is shaping up for the Internet’s public square. Here’s a very brief timeline:
March 9, 2020
Silver Lake Partners takes $1 billion stake in Twitter
Private equity fund and activist investor Silver Lake partners scoop up a significant chunk of shares, takes a board seat.
After years of critique / mockery of Twitter policies, Musk posts a poll: “Free speech is essential to a functioning democracy. Do you believe Twitter rigorously adheres to this principle?” He followed this immediately with “The consequences of this poll will be important. Please vote accordingly.” 70% of those responding say “No.”
April 4, 2022
Musk becomes Twitter’s Biggest Shareholder
Over the ensuing days, Musk takes 9.2% stake of Twitter, becoming largest shareholder. The Washington Post notes that he may have run afoul of SEC reporting rules which require public disclosure within 10 days of acquiring a 5%+ threshold. (Seems likely he will pay a fine for this.)
April 5, 2022
Twitter Announces Musk Joins Board
Twitter CEO Parag Agrawal announces that Musk is offered board seat, but the terms of that agreement state that Musk must keep stake capped at 14.9%.
April 8-10 2022
Musk Muses Big Changes
Tweets several new ideas for the social media company, some controversial. One of the most intriguing: Musk suggests that anyone that pays the $3 per month Twitter Blue subscription fee should get a checkmark. A follow-up Tweet clarified that it would be different from the blue badge, but still signify the account is authentic. When challenged in a comment to consider lower prices in other countries, Musk agreed, and suggested a proportional rate tied to the local affordability and currency. (The tweet has since been deleted.)
April 10, 2022, morning
Musk Rejects Board Offer
April 10, 2022, Afternoon
Agrawal tells his team “Don’t Get Distracted”
April 11, 2022
Musk Amends SEC Filing
“From time to time, the Reporting Person [Elon Musk] may engage in discussions with the Board and/or members of the Issuer’s management team concerning, including, without limitation, potential business combinations and strategic alternatives, the business, operations, capital structure, governance, management, strategy of the Issuer and other matters concerning the Issuer.”
Reading the Tea Leaves
What’s really going on?
As Kara Swisher of the New York Times has noted, there’s much to be gleaned from translating some corporate speak:
Elon has decided not to join our board. I sent a brief note to the company, sharing with you all here. pic.twitter.com/lfrXACavvk
fiduciary –> “Elon, don’t mock, or speak ill of the company publicly, you have obligations if you’re on the board which hem you in,”
background check –> “we always reserved the right to reject you based on potential SEC investigation and other things, so this is kind of mutual anyway,”
formal acceptance –> “you must agree in writing not to take 14.9% stake, and be liable if you tweet something defamatory,”
there will be distractions ahead –> “this ain’t over”
What happens next?
Musk is mercurial. He could decide he has better things to do and sell his stake.
But it seems much more likely to me that he will continue to increase his stake. If his goal were to make marginal improvements to Twitter, he would have been inclined to stick to their announced agreement and take the board seat.
He initially filed an SEC form saying he was planning to be a passive investor in the company, but amended it today (April 11th, 2022) to indicate he may be more active, and plans to keep criticizing the platform and demanding change: “From time to time, [Musk] may engage in discussions with the Board and/or members of the Issuer’s management team concerning, including, without limitation, potential business combinations and strategic alternatives, the business, operations, capital structure, governance, management, strategy of the Issuer and other matters concerning the Issuer.”
The billionaire has been vocal about some of the changes he’d like Twitter to make. Over the weekend, he tossed out the idea that users who opt into the premium plan ($3/mo), Twitter Blue, should be given automatic verification and see no ads. This one step, of devaluing “blue checkmarks” would be a sea-change to how Twitter is used today. He noted that Twitter’s top accounts are highly inactive, asking “Is Twitter dying?” He mused about turning Twitter’s San Francisco HQ into a homeless shelter, which invited a retort from Amazon’s Jeff Bezos. As Geekwire reported in May 2020, Amazon has already done so in part, quite successfully, in partnership with Mary’s Place.
Twitter’s board may very well adopt a poison-pill defense. But this isn’t a slam dunk; it needs majority board approval. Take a look at the existing composition of Twitter’s board; it’s no longer founders and insiders. Remember, this board said goodbye to Jack Dorsey, and rumor has it this was in part due to sluggish stock price results and activist shareholder discontent. Twitter’s eleven-member board consists of two “insiders,” both Agrawal and Dorsey, an activist value-driven investor (Silver Lake Partners), and eight relatively independent board members with Silicon Valley and/or finance experience. Poison pill adoption often depresses the value of a stock, and some board members might not be persuaded to do this. Several of Twitter’s board members are from the Silicon Valley braintrust, and are unlikely to want to go head to head against Musk — and some may even very well fully agree with him.
Musk is nothing if not bold. He has risked substantial sums and bet boldly on multiple ventures in the past. He stated that “free speech is essential to a functioning democracy,” and has both an internal incentive and external incentives not to be seen as being bested here.
My guess: he’s unlikely to just sit on the sidelines, as Twitter’s biggest but minority shareholder. He could well make a run for the company, though he may prudently wait for the next recession to do so.
And what happens to Twitter’s employee base, and its policies, during this tumultuous time? It may cause some employees to see the writing on the wall, and depart. Or it might cause some to double-down on a heavy-hand. It could be a very interesting few months indeed.
Does Elon Musk like to play it safe? Or lose? What does his track record suggest?
Steve’s a Seattle-based entrepreneur and software leader, husband and father of three. He’s American-Canadian, and east-coast born and raised. Steve has made the Pacific Northwest his home since 1991, when he moved here to work for Microsoft. He’s started and sold multiple Internet companies. Politically independent, he writes on occasion about city politics and national issues, and created voter-candidate matchmaker Alignvote in the 2019 election cycle. He holds a BS in Applied Math (Computer Science) and Business from Carnegie Mellon University, a Masters in Computer Science from Stanford University in Symbolic and Heuristic Computation, and an MBA from the Harvard Business School, where he graduated a George F. Baker Scholar. Steve volunteers when time allows with Habitat for Humanity, University District Food Bank, Technology Access Foundation (TAF) and other organizations in Seattle.
Wiring diagram for pump motors
