Immich distributed compute for https://github.com/immich-app/immich/tree/main/machine-learning with a control panel to witness the "boost" (which tbh might be impossible to notice)
- Posts
- 1
- Comments
- 922
- Joined
- 4 yr. ago
- Posts
- 1
- Comments
- 922
- Joined
- 4 yr. ago
Ah... forgot about them for a while, checking : last update from April "#58 - Dev Kits, Compute Packs, & 2025 Timeline" so in theory now they are on board certification. Good occasion to also check https://github.com/SimulaVR/Simula/commits/master/
Out of curiosity, did you try one?
I setup WireGuard only last week so maybe I'm the one who misunderstand something : on your LAN assuming you are NOT using your router (or switch, or a networking device) to be a peer of the VPN, don't you need to add each machine as a peer to the VPN? Also doesn't that leave the most granularity so that the (root) user of each machine can chose to be on/off and more, e.g. split tunneling?
Indeed but by doing so I can connect from the outside World too, e.g. if I'm at the dentist waiting for an appointment, I just connect to the VPN over my 5G connection, no login required.
Because it's low end I'd put :
- headless Debian pre-configured with WiFi and
sshdto then add CopyPartyvia its single.pyfileapt install minidlnato serve media files back to add devices on LAN, e.g. VLC on desktop and mobile devices- mount a large microSD for data
- I'd add a WireGuard VPN configuration file and make both accessible outside the LAN but only on my devices
All that is relatively quick if you have done it before (maybe 30min total) and can run 24/7 for years requiring very little power.
Also FWIW if I wouldn't get an answer within few hours and I knew for a fact that with a fresh install it worked, I'd re-install.
It's perfectly fine to do the process again as it insures your files are safe (either working backup or separate disks, or ideally both) and you know what software is relevant for you, that your configuration files are well known, etc.
Installing a distribution should be a painless and quick process.
Typically my debugging process goes like this :
- error message? Search for it online with the most unique keyword that aren't machine specific
- solutions provided?
- solution understood? try it then loop back, writing notes in own wiki
- solution not understood? bookmark it then try understood solutions first, if not try and loop back
- solutions provided?
- no error message?
- find where the error message is!
- what actually produce the error from the top of the stack? end-user software? service? kernel? hardware? where do they put logs?
- if logs exist and verbosity is not sufficient, increase verbosity and reproduce the problem
- what actually produce the error from the top of the stack? end-user software? service? kernel? hardware? where do they put logs?
- if no verbose enough error message can be obtained, repeat the situation in various conditions
- does any condition make it work?
- search on the difference between the working and non-working condition
- backtrack one layer up the stack, e.g. if end-user software does not change, try service, etc
- does this one provide logs?
- does any condition make it work?
- find where the error message is!
So... it's basically always the same, namely try the lazy way (error log search) and if that's not enough, try further down the stack or more unknown BUT always get information out the try.
TL;DR: I have no idea but if another new machine (e.g. phone) can connect then DHCP works. FWIW NetworkManager logs are in
journalctl -u NetworkManagerand you can manually add/remove Ethernet connections. I'd physically unplug then plug back the cable with WiFi disabled.I'd suggest 0 change at first : boot on a live USB then connect with her Web accounts (e.g. Firefox Profile) then get an easy win. Sure not 100% will work but she'll be 80% there in minutes. If she hates it, logout, reboot, remove key and that's it.
Nobody gives a shit about NIST if they lose the 1 thing that make them useful : their credibility.
If some credible doubt is shed on them ... then NIST is just an acronym with no power.
That being said IMHO a pragmatic heuristic is spotting "Do what I say, not what I do" and thus if NSA relies on PQ, or hybrid, or something well you can deduce from that they assume whatever solution they do NOT use if then not safe in a useful lifespan (which might be totally different from your threat model).
Edit : did tinker with https://openquantumsafe.org/about/ in particular https://github.com/open-quantum-safe so if you have an opinion on that I'd be curious.
Since you do not seem to list self-hosting options, e.g. WireGuard or OpenVPN, then IMHO it'd be good to at least have a line on each about what's the actual backend, e.g. does service X runs on WireGuard, OpenVPN, something else, something proprietary that has been audited by 3rd party if so whom and when.
Edit: suggested self-hosting (but not at home) WireGuard in the previous thread https://lemmy.ml/post/37270537/21536054
Panicking and paranoia is counter productive.
If you do "lot of risky and dangerous shit" then it's even more important that you do so mindfully. If you get careless because you are tired you increase the risk.
Personally my "trick" is to learn from others, e.g. in few weeks in Paris there will be https://splintercon.net/paris/ where tools and processes will be explained. I can learn from them.
Also my way to stay calm isn't just to be mindful or learn... but do stuff, no matter how small. If you learn about a new thread, address it today. It doesn't mean fix the problem entirely (it'd nice if you could) but rather do something, ANYTHING, about it. If it's not solved, write notes about it and resume tomorrow or whenever you can. Every small effort does add up over time.
Finally I find that sport helps a lot to "evacuate" stress. If I feel some pressure from work or the overall situation, I go outside and sweat it out. It doesn't magically make the World better but it insures I'm a bit more in shape to try to tackle whatever is thrown at me.
I head about Audiobookshelf and I already use Escapepod on mobile, LMS for music (self hosted), so wondered few times about installing it too.
So can Audiobookshelf be used for clips and annotation?
I find that more and more often I listen to a podcast, it gives me an idea and I want to :
- pause
- record my idea, by speaking it aloud for e.g. 30s
- unpause to keep on listening
- review the idea later on, e.g. on my desktop, while maintaining provenance, e.g. this idea "monetization of own content creation for creators on my on-going VR project" (ideally as text at this stage, so using STT) was sparking by listening to podcast "Voices of VR episode 1226 on VR Chat and monetization" around 25min in.
Edit : just checked on their demo server and there is a
bookmarkoption. It's just text for me, it doesn't clip part of the audio, but it does associate some typed text to a moment in time. It might be enough for me.Might want to look into split tunneling.
As they say in Italian "If my Grandmother had wheels she would have been a bike"
Sure... that's an "interesting" premise but we live in our World, where a lot of American companies are structured the way they are by design. US companies get money from venture capital. That capital is solely designed to dominate. VC money is NOT a loan you get from a bank where you provide a collateral. No VC money is targeting 1 thing : market dominance and 10x returns. Your mom&pop shop will never get VC money because they never say they'll corner the croissant market, rather they say might sell some baked goods to some people in your limited neighborhood. VC money will NEVER accept such a deal because it might eventually get 2x, at best, and the tiny shop does not even need a lot of capital, just enough for the oven, few people, etc. Peanuts in terms of investment money.
So... American companies are not "evil" because they want to or because a lack of luck. No, rather they become so because of the very structure money is made in the US. The Silicon Valley isn't special because of Stanford or Berkeley and so many smart grad students. No it's special because it pulls people from the entire World who dream of dominating markets. It then either select them or transform then select... and in the end you get the same kind of companies with the same kind of strategy with the same kind of money doing the same thing : domination by lowering price, cornering a marketing, raising price, enshittifying. Why? Because it works. It's a proven business model. Right now it works on ads, and thus privacy... but if another model comes, it'll use that.
TL;DR: it's not perfect but it sure beats most if not all of BigTech depending of course on your needs.
You would be well served by staying current on the technology as it develops.
What did I miss? https://fabien.benetou.fr/Content/SelfHostingArtificialIntelligence#UsedLocally
Edited : direct link to tried things.
Yep, that'd be me. That said if I were to buy a new GPU today (well, tomorrow, waiting on Valve announcement for its next HMD) I might still get an NVIDIA because even though I'm convinced 99% of LLM/GenAI is pure hype, if 1% might be useful, might be built ethically and might run on my hardware, I'd be annoyed if it wouldn't because ROCm is just a tech demo but is too far performance wise. That'd say the percentage is so ridiculously low I'd probably pick the card which treats the open ecosystem best.
Nice, leaving Google is a lot of work but worth it, kudos!
FWIW Im torrenting on my server 24/7 for years. I'm only torrenting Linux ISO though, using
transmissionin a container.Started to write a long paragraph to explain the difference between privacy and anonymity but I now believe this new user is (no idea why) collecting engagement via rage bait. I won't participate in their posts anymore.
It might even come from a good place, namely trying to always do "better" and be "more private" but in practice it's just lead to confusion.
Well the EU and California are. Just yesterday I was reading https://www.lesnumeriques.com/informatique/une-premiere-victoire-polytechnique-claque-la-porte-a-microsoft-365-et-choisit-le-logiciel-libre-la-revolte-gagne-les-universites-n243905.html which is one example among many, here one of the most famous French engineering school switches away from Microsoft due to lack of GDPR clarity. Numerous administrations in the EU do the same. Yes it slow, yes it's not enough BUT it's something. I can't say if the momentum will grow and transform to a mass exodus, or if it's not a real trend and my bias highlights what I hope to see. Still it is happening.