First I basically described the project. Since it's quite small, it's easy to give a small summary. Then I started out describing my goals, that I wanted to fix some user issues. Then mostly gave the issue description, some extra information (my idea of what might be wrong). 4/5 times, it made the correct decision and I didn't have to promt any extra. I did have to review the code and polish some minor parts though.
I tried to focus on staying on the subject. For example when I wanted to embed the themes to create a theme switcher ad-hoc and that was done, it could easily generate some new themes for me without any issue at all.
After every edit, I did a `git diff`, modified if needed (using vim), then ran the program and tested it out.
My experience is that it's always easier to fix things fast/correct if you already know the codebase and can give hints to the agent.
Probably not many cares, but as a developer for many years I've been following the LLM race with excitement. And instead of being afraid of loosing job etc, using these tools for grunt work that previously was both booring and perhaps not too satisfying, I would say it's amazing. It's not perfect, but if you know your domain and know your goal, then it's just something that makes it both fun and productive at a whole new level.
And for my old side-projects that didn't get enough love, it's a perfect match.
I've been running Frigate for many years, using a PN50 NUC and a Coral USB dongle, the Coral is a must, at least in my case. I had a full blown Ubiquiti/Unifi setup with cameras + their software. Way to many false alarms compared to Frigate. Now I run 10+ cameras with 24/7 recording and alarms with images pushed to Telegram. The identification is instant as well as the telegram message.
Running a mix of Ubiquti/TP-Link VIGI+TAPO/Reolink. I'm running everything in containers and everything works perfect!
Polling HN: is there any upgrade to Coral? It's 5 years old at this point, and with the explosion of AI apps & HW acceleration, I'm surprised there doesn't seem to be anything to update Coral's niche, of an IO-attached NPU.
OpenVINO might be a good alternative, as many Intel-based mini pc’s support it. Or a decent desktop with an Intel CPU. Or maybe something with an Arc GPU (integrated or dedicated).
Disclaimer: I didn’t try it yet but the last rabbit hole regarding OpenVINO comparisons looked too good to be true and it seems Frigate supports it too. Win-win.
TIL openmv.io, looks really neat for small project. Especially cool with the thermal vision, that would be a very nice addition to improve false positives for <living-things> detection.
But for surveillence, it's usually the sensor/camera quality that is the most important. I've struggled hard to find an affordable IP camera that can actually handle both shutter speed + quality in order to for example read license plates.
are you using with only one Coral USB dongle at the same time (plugged in the PN50 NUC) and get successful object or person identification with frigate?
And why telegram? Is it connected to frigate only for notifications resulting from the identifications?
a) 8x PoE cameras
b) 2x WiFi cameras + sometimes some esp32cam etc.
Yes, only one Coral dongle and it's handles all cameras perfectly. With some masks I rarely get any false positives and it is like 99% correct hit-rate.
Telegram is just a way to get a fast glance of an detection, so it sends me an image with what type of detection it was and the frame it found it in with detection frame around the object. This is handled via Home Assistant and some automation I've written. The results comes via mqtt to hass.
It aligns quite well with their Linux client that hogs so much memory that I need to run it in a cgroups with a memory limit to avoid eating too much memory :)
Indeed, quite fun! I tried to lure some agents about a secret mission, but they didn't want to go on a goose hunt and shot me. But one dude actually wanted free beer and hot dogs (as long as they weren't vegan and no mayo).
Nice approach! I added a very basic keyword filter in my rss reader (https://github.com/lallassu/gorss) to do some sort of "cleaning". But having a section in the reader that would filter out the articles more intelligent would be very nice, and maybe bundled them into clusters.
I've made 2 projects that I use everyday for several years now. Not sure if I'm proud really, but they are such useful tools in my daily life so I guess I should be!
One is a RSS feed reader (GORSS) for the terminal that I use to always be up to date with stuff that interests me. The other is a simple todo-list that I use for work, shopping etc (DoIT).
I tried to focus on staying on the subject. For example when I wanted to embed the themes to create a theme switcher ad-hoc and that was done, it could easily generate some new themes for me without any issue at all.
After every edit, I did a `git diff`, modified if needed (using vim), then ran the program and tested it out.
My experience is that it's always easier to fix things fast/correct if you already know the codebase and can give hints to the agent.
reply