For a platform that was effectively leaked in full, extended, minute detail ahead of last WWDC, we've heard shockingly little about anything coming to visionOS 2.0…
It’s sort of interesting to see the financial markets paying attention to #SmallAI. That means that anything that affects “big” AI is important enough to move the markets. Makes sense, if it’s going to affect the $NVDA stock price, it’ll affect the market. https://finimize.com/content/small-talk
@simon
Simon, I'm working with @homeassistant a bit and we just had a fascinating discussion about 'nanoLLMs' that could run locally. They would NOT need the sum-total-of-all-human-knowledge but would really just be there as a smart parser for speech-to-text commands, keeping everything local. This is clearly still not trivial but hopefully one way to reduce the model size.
Do you know of any 'reduced' LLMs that could work in this more limited context?
@scottjenson@simon@homeassistant It's an fascinating area that a lot of people are looking at right now. Moving the LLMs to edge hardware is certainly possible, I'm running LLaMa on my phone locally for instance. But you have to think about architectures. I've seen some interesting architectures built around key framing and feeding LLMs from TinyML models that look potentially pretty powerful.
@artemesia The original claim from Amazon was that they were using AI for their “just walk out” technology. The scan as you go? No. We’ve had that in the U.K. for about 20 years. It’s not exactly new.
@artemesia A lot of AI startups seem to be relying on smoke and mirrors and minimal AI for their MVP. The surprising thing here is that Amazon was doing the same. Although I guess we’re seeing it with self driving cars, so maybe not.
@LewisWorkshop I worked on distributed agent-based systems — back when they were briefly the thing — and pretty much every slide deck I gave around that time had a picture of the Turk in it somewhere. Mostly when I talked about limitations, magic, and what problems "intelligent" systems could and couldn't be expected to solve.
Next week I head to London for #MonkiGras! I have heard such wonderful things about it over the years, and I'm thrilled to attend my first, this one about #ArtificialIntelligence and culture.
Super excited to see Farrah Campbell and Brooke Jamieson speak 🎉
When I prepared a @Raspberry_Pi with the latest #RaspberryPi OS (based on #Debian Bookworm) as a monitoring and observability display, I noticed the "wayvnc" package during the dist-upgrade.
Turns out this is a pre-installed #VNC server package - at least on the Desktop variant.
PiCameras are amazing pieces of hardware with a software that could probably be managed better.
The original PiCamera libraries haven’t seen a commit in 4 years, as they were deprecated with the Bullseye release. I can still run some code that uses that library on some old #RaspberryPi just because I’m keeping them on some ancient version of Raspbian. Otherwise, raspistill and friends break in many possible ways both on both Bullseye and Bookworm, and the package doesn’t even install on Arch ARM because raspberrypi-firmware is now gone.
A couple of years down the line, and its replacement, PiCamera2, is still in beta. It can be installed through a relatively smooth process only on the last two versions of RPi OS via apt, and it’s otherwise very hard to get installed on any other distro - on Arch it depends on packages that officially aren’t even available for ARM (like python-av), when installed via pip it tries to build the world even if some packages (like numpy) are already installed on the system, and I didn’t manage to get it to run on Ubuntu because of permission issues.
It’s really a pity because a vibrant ecosystem of camera apps and scripts had been written using the old version of PiCamera, which could do a lot of things with very low entry barriers. Then a sudden deprecation was announced without a viable alternative, and a couple of years down the line that alternative isn’t quite stable yet. I’ve eventually resorted to leverage the native v4l2 integration over ffmpeg as a cross-platform workaround, but that moves most of the burden to the CPU and I’m not really leveraging this hardware at its best, plus it probably raises the technical bar for a lot of hobbyist makers.
Why was something so important to many users deprecated without any stable alternative on the horizon?
@fabio Because the author of picamera stopped supporting it? Wasn't our decision. The original picamera library wasn't written, or supported, by Raspberry Pi. The new picamera2 library was written in-house, and is extensively documented, https://datasheets.raspberrypi.com/camera/picamera2-manual.pdf. It might not do what you want (yet), but it's being actively maintained and worked on.
@gregoinc@ostueker@Raspberry_Pi I wouldn't want to say definitively unless I had to boards in my hands. But looks to be the packaging update kicking in, you'll be seeing our packaging being refreshed across all our products with the new product specific word marks and the new logo.