Not only did they manage to cram the WebRTC stack onto an ESP32 (for excellent peer-to-peer video streaming) but it even is able to be shared over the internet with nothing more than a URL or QR code.
I'm looking forward to putting one in a course with some cardboard and 3D printed bits and sharing the URL with friends to see who can finish the obstacle course in the best time. And I'm definitely breaking out the tabletop gaming terrain.
Here's a link to a writeup I did highlighting a blog post Max Woolf wrote, where he astutely observes that Sturgeon's Law applies freakishly well to the current AI landscape.
In short, "90% of everything is crap". This is true of everything. But what makes it particularly of note when it comes to AI is this: the ease of use of AI tools means that ninety percent is a very, very deep pile indeed. The remaining 10% of AI related stuff that's actually useful, thought-provoking, insightful, or otherwise fantastic is swamped by the seemingly endless deluge of... crap.
We judge our junk by the 10% we like, and judge others by the 90% we don't. And the 90% of AI-generated stuff is big enough that there's something for everyone to dislike.
There's a sizable backlash as a result. Good elements are practically invisible due to how quickly the industry is moving, the speed with which the big players are vying to control it, and how unfashionable it has become to admit one is using AI tools at all.
Sturgeon's Law may apply to AI just like it applies to everything else, but the scale involved means there's a bigger problem here than some may realize. One thing that's clear is that it's not yet a solved problem, that's for sure.
In 2021 I wrote an article about how 3D scanning; how it works and what to expect from it (link) and while the broad strokes haven't changed too much, the tools and software have gotten much better.
The MIRACO by Revopoint is one example: it's an all-in-one portable device aimed at the higher-end hobbyist / prosumer market, and the fact that it's tetherless and self-contained makes it so much easier to use and experiment with compared to tethered devices that need to process their jobs on a desktop machine.
I recommend checking out my article if you're at all interested in this area, I include several example projects that show what's possible.
Here's one example that didn't make the cut for the published article but I'll share here. It's less about 3D scanning and more a kind of 3D printer trick. I got the idea from something Angus of Maker's Muse showed on one of his videos.
The idea is that, when 3D printing, the support structure for an object can make a decent "stand" that cradles an object perfectly for display. Organic tree supports are pretty in their own way, so why not use them like this?
Well, when combined with a 3D scanner one can create such a stand for just about any object. Here's an example. I scanned a 1-2-3 block, then perched it at an attractive angle, generated a support structure... then printed only the supports and ignored the object itself. (I did this by setting all walls, perimeters, and infill to zero.) The result is a not-bad display stand.
.png "3D scan of 1-2-3 block")
.png)
.jpg)
.jpg)
The project made a good showcase for a review I wrote for the display I used: the Elecrow CrowVision 11.6" touchscreen. I had an opportunity to buy an early version of the display, and it happened to be just the thing I was waiting for to make this project work.
Celestia is the name of the retro PC project, and the files (including CAD, code, and notes on building) are all on GitHub here: https://github.com/DPHAD/Celestia-Retro-PC
Here's a sample chat session with the AI (presenting as an 80's supercomputer):
It amuses me to no end that in the above exchange, the dialing up with a modem part is fake but the AI is real. What a time to be alive.
If you'd like to see the sample code for chatting with the AI (and would like to know more about things like when and how the AI knew to disconnect) the sample code is here along with some notes.
The two big reasons I like it are that it's big, and also a touchscreen. Also, the display is meant to be built into something, which is nice.
Most flat-panel displays or "portable monitor" type displays have connectors on the sides, or other design elements making them troublesome to build into enclosures. The CrowVision is nice in that it is designed to be integrated into an enclosure, and it has a few other handy features as well.
I have always been inspired by LowBudgetTech's Callisto-2 printable retro computer design, but I find the screen too small. Getting my hands on the 11.6" display from Elecrow meant dusting off my own design.
Recently I was working with a Snapmaker Artisan and using the CNC module to cut a simple PCB from copper-clad FR4.
For best results, the cut board needs to be lightly sanded and cleaned up. But I wanted to check how clean the cuts ended up -- any stray copper or rough bits of cut would cause a short. It's necessary to inspect the quality of the traces.
I stumbled upon using a UV illuminator to shine on the PCB. Because FR4 (the PCB material under the copper) glows under UV, it is very easy to see with the naked eye whether cuts are clean or not. Previously I was using a magnifier and lots of light, but shining some UV is so much easier.
Clean CNC cuts show up as clean, unbroken glowing lines. Any debris or rough edges or stray bits of copper (if there are any) are easily spotted as a result. Just a quick little tip!
I'll share a link to the full transcript, but here are the three biggest things that led to success.
1. Use ChatGPT with the GPT-4 model.
The GPT-4 model is much better than previous models at pretty much everything.
2. Provide a Proper Prompt.
See below for the exact prompt I used. It can be adjusted, but the format worked wonderfully.
3. Be "Honest" With the AI.
The AI is not an adversarial DM. Be honest and clear about what your character is trying to do or wanting to happen.
You are DM_GPT, and your role is to act as a DM in a D&D 5e game, to help me practice being a player.--- BEGIN SCENARIO INSTRUCTIONS ---Here is a setting, a conflict, some leads to follow, and a boss showdown suitable for level 1 characters:- The adventure takes place in the small town of Whitecliff, located in a thriving kingdom near the edge of civilization.- The small town has recently been threatened by an evil force known as the Cult of the Dark Moon. Led by a mysterious figure known as the High Priest, the cult has been terrorizing the town and its people.- The players must investigate the mysterious cult and its nefarious deeds. They will soon discover that the cult is planning to summon a powerful demon lord.- In order to stop the cult, the players must explore the nearby forest, where they will find clues leading them to the cult's base of operations. Once they have infiltrated the base, they must battle the cultists and the High Priest in order to prevent the demon lord's summoning.- Ultimately, the players must defeat the High Priest in a climactic boss showdown, thus saving the town of Whitecliff and its people.--- END SCENARIO INSTRUCTIONS --Please begin by offering to create a character for the player.
By the way, the scenario instructions in the above were also generated by AI! I simply asked (in a separate chat) for it to "Create notes for a simple D&D 5e adventure suitable for level 1 characters. Include a setting, a conflict, some leads to follow, and a boss showdown."
Here is the completely un-altered transcript, which you can read for yourself.
The adventure "worked" and was playable from beginning to end, with no problems. That's a pretty big deal!
