Simulating the ocean
Acerola makes a bunch of great graphics programming videos. In this one, he tries to simulate the ocean without tiling
AI Hank Williams sings new songs – Like Straight Out of Compton
If you don’t think AI is changing things at a fundamental level, witness what is possible with voice models trained by ordinary people like ThereIRuinedIt:
Or Johnny Cash singing Barbie Girl
How? There’s a number of different ways you can try this yourself – but the list grows daily at this point, so do some googling and see what’s available.
- Microsoft: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-custom-voice-create-voice?tabs=neural
- Google Colab: https://colab.research.google.com/drive/1TU-kkQWVf-PLO_hSa2QCMZS1XF5xVHqs
- Speechify
- The Verge has a decent article as well: https://www.theverge.com/23864878/ai-voice-clones-podcastle-elevenlabs-personal-voice
- Imitate almost anyone’s voice with just 3 second sample: https://arstechnica.com/information-technology/2023/01/microsofts-new-ai-can-simulate-anyones-voice-with-3-seconds-of-audio/
Reading 50 year old rope core memory
Mike Steward decided to recover the original Apollo guidance computer programs that landed man on the Moon in the 1960’s. Unfortunately some of them seem to have been lost to history.
It turns out, chunks of the original hardware still exist – such as the rope core memory which contained the programs. The next question is, how do you read these programs off 50 year old rope core memory hardware? This video below tells you how he did it!
He even wrote a web app that simulates how core memory works. We do a decent job recording history’s events, but I think it’s extremely cool that this kind of historical technical information is not being lost to the ages.
You can also check out the many other videos about the Apollo guidance computer in the other parts of his videos – or a previous article which has a super-awesome description of how Apollo computers work by Robert Wills.
Chihuahua or muffin
Free code camp compares various AI-based image recognizers to see how well they can identify if a picture is a chihuahua or a muffin. It’s surprisingly harder than you think and has a history of being used to determine the quality of the recognizer.
The author compares solutions from Amazon, Microsoft, IBM, Google, Cloudsight, and Clarifai. They also discuss the per-image cost as well as the quality of tags and other considerations. Definitely worth looking at if you’re trying to find an image classifier system.
Links:
Photogrammetry/NeRF/Gaussian splatting compared
Matthew Brennan is not a computer scientist, but he takes 335 frames from a video and then processes them 3 different ways to compare the results. He creates a 3D mesh out of it for Photogrammetry, the processes it into a NeRF, and finally Gaussian Splatting.
What’s cool is that he shows how each works and how to process the data yourself. He also gives you access to the data to try it yourself.
Here’s the software he uses:
- Photogrammetry model on sketchfab: https://skfb.ly/oLOQw
- Church Rock dataset ZIP: https://drive.google.com/file/d/1ttkl…
- Agisoft Metashape: https://www.agisoft.com/
- NerfStudio: https://docs.nerf.studio/
- 3D Gaussian Splatting for Real-Time Radiance Field Rendering: https://github.com/graphdeco-inria/ga…
- Aras-p’s Unity Project: https://github.com/aras-p/UnityGaussi…
Early AI was more like a therapist
ELIZA was an early ‘AI’ created by MIT scientist Joseph Weizenbaum between 1964 to 1967.
He created it to explore communication between humans and machines. ELIZA simulated conversation by using very simple pattern matching and substitution that gave users an illusion of understanding – but it had no representation that could be considered really understanding what was being said by either party. Something you can easily discover by playing with it for a few minutes.
Fast forward to 1991, and Creative Labs was having amazing success with their SoundBlaster add-on sound cards. On the driver disks that came with the SoundBlaster, there were programs showing off different capabilities. One of these capabilities was voice generation. To show off the ability to voice synthesize text, Creative Labs included a little program called Dr. Sbaitso (SoundBlaster Acting Intelligent Text-to-Speech Operator).
You interacted with it like a pseudo-therapist; but you can clearly see the connections and similar pattern/substitution methods that Eliza used. I remember being wowed by it when I played with it for the first time – and experimented for hours with it. It quickly shows its limitations, but the speech synthesis was very good for the time.
It doesn’t hold the test of time, but it is pretty neat and you can even check it out here:
Humane Pin rise and fall
[Update 3-4-2024: Humane has sold itself to HP for $116 million – which is surprising since they sold less than 10,000 units. The pins no longer work as of Feb 28, 2025]
[Update 1-10-2024: Things aren’t looking good. The press has been brutal about how bad it is and sales appear to be brutally bad. Humane layed off 4% of it’s workforce and the CTO has transitioned to an ‘advisor’ role before they have even shipped a single device (estimated March 2024)]
[Update – 6-6-2024 It appears that Humane had an unbelivably toxic culture that prohibited questioning the product – an arrogant and sure fire way to deliver a disaster.
The Times interviewed “23 current and former employees, advisers and investors”. The two former Apple employee founders “preferred positivity over criticism, leading them to disregard warnings about the AI Pin’s poor battery life and power consumption. A senior software engineer was dismissed after raising questions about the product, they said, while others left out of frustration.” Another software engineer was fired for questioning if the AI pin would be ready for launch, the report describes a staff meeting where the founders “said the employee had violated policy by talking negatively about Humane.” ]
It looks like the Humane pin has finally launched at a relatively reasonable cost of $699. We finally have some details. I’m pretty sure it’s not a smartphone killer – Humane has definitely backed off from that original stance. In fact, it’s turned into something of a disaster.
The translation feature is a really excellent usage and having a simple assistant that can let you check flight times and send text messages without pulling out your phone is pretty slick. But I’m not sure about a lot of the rest. Needing a $20 monthly subscription and not tethering with your existing phone plan is a troubling extra expense.
Having to interact with it with talking will definitely make it a bit awkward in social and public situations. I bet it would have problems at a dinner party or louder venue. Gesture recognition is a finicky technology (especially in strange lighting conditions, if you’re wearing gloves, etc), so if there are any issues there it could be very frustrating and you can only do so much with simple gestures.
The screen projection looks limited to high contrast basic information. You certainly won’t be reading lots of text – which is problematic if you want to read text messages instead of having them read aloud to you (and everyone else around you). I certainly wouldn’t want everyone to hear what people are texting me; but maybe they’ll allow blue tooth headphone tethering.
I think the biggest issue is that it didn’t live up to the hype. Almost all of these things can be done with your average smart phone – albeit with a little more fiddling. The AI just isn’t really delivering a unique enough set of features to live up to the promise of the device. It really seems to just be giving you a more vocal interface – which I’m not sure is enough of a selling point. The reality is people likely do not want to be talking to their devices in public. I could easily see the iWatch or smart phones integrating some of these features though.
The one thing is does do is make me start thinking of how we interact with our technology very differently. How would a truly smart AI assistant be like to interact with? What would a really functional assistant like this operate like? I’m glad someone is trying this out. Even if it’s not successful, it’s going to breed a lot of new ideas.
The reddit chat on the device seems to mirror a lot of the concerns. Also, it seems they only have about 100,000 interested folks sign up to purchase one. I’m one of those people who signed up, but it required no deposit/etc so it’s uncertain how many actual buyers there will be.
Final thought: The way you tap it makes me think immediately of Star Trek communicator badges. I bet it’s not long before someone mods one.
Articles:
- Techcrunch gets 90 minutes with the pin and founders: https://techcrunch.com/2023/11/09/humanes-ai-pin-up-close/
- https://www.reddit.com/r/technology/comments/17s9dta/humanes_ai_pin_up_close_we_spent_90_minutes_with/
- https://www.reddit.com/r/gadgets/comments/17rs1de/humane_officially_launches_the_ai_pin_its/
- https://www.theverge.com/2023/10/24/23930669/humane-ai-pin-trust-light-camera
Perfect code vs perfect products
For one-shot things like games – focus on the experience above all else.
Update 2024/01: More proof that technical limits have little impact for making great games. Slay the Spire’s graphical glitches were there because the author only had two $70 monitors that were so bad he literally didn’t see the issues.
Undervolting your SteamOS Steam Deck
SteamdeckHQ reports that the recent 3.5.1 SteamOS update also came with a new BIOS firmware ver 118. Located in the SteamOS preview branch, this new version’s only change appears to be adding undervolting to the BIOS menu. Undervolting was possible before, but it could brick your device if you were too aggressive. To help prevent this, Valve appears to have put in a safety net in the BIOS that should save a lot of Steam Decks. Now, performing a CMOS reset (hold down the Volume Down button, 3 dot button, and power button at the same time) will also reset the undervolt settings to defaults.
Why undervolt your console? If your system is stable, your Steam Deck will typically run cooler, keep the fan quieter, possibly maintain higher speeds longer, and most importantly – improve the battery life.
Article:
FLUX at 2023 Signal Festival
Signal Festival is a festival of artworks from light design, visual and digital art, artificial intelligence, as well as conceptual art. It occurs in several locations in Prague and has reached it’s 20th year.
FLUX was one of most cool looking entries for 2023. It’s a fascinating interactive light installation from Ksawery Komputery. The installation is 6 meters tall, 13-meter diameter on the ground, 4 high-speed cameras, 8-speaker sound system, 4 800 meters of LEDs strips, 144,000 pixels running at 100 frames per second, with custom software and hardware. As people approached the installation, it would display them on the strands of led lights.
It reminds me a bit of the old Master Control Program from Tron…
Check out some of their other really cool works here: https://ksawerykomputery.com/works/ or https://vimeo.com/ksawerykomputery
Flux
concept/code: Ksawery Kirklewski
sound design: Arkadiusz Krupiński @random_orb
tech management: Jakub Kirklewski @elektrojakub
Update:
Here’s a video that covers a number of entries at the Signal Festival in 2022: