zenoprax

7 ms·

1.
by zenoprax 2mo ago
Yeah, I just use the "netinstaller" ISOs since it's much smaller and never needs to be updated. If I had a need for air-gapped/offline installs I'd either download a larger ISO or just manually install packages from
2.
by zenoprax 2mo ago
Yes, models are a good use for P2P especially if everyone agrees to share the same torrent and someone (or a cohort) commit to seeding for the long haul.
3.
by zenoprax 2mo ago
Every change to the source is effectively a new torrent. This creates a ton of fragmentation as data is reorganized, remixed, reencoded, and so on. You can see this with many Linux distros: there is no single Debian torrent that people seed
4.
by zenoprax 2mo ago
You only need one person/organization to commit to seeding. The majority of people do not want to seed at all without some sort of incentive. If this site represents a coordinated datahoarding effort then there will be at least a few p
5.
by zenoprax 2mo ago
I have three identical Lenovo SFF PCs with a U.2 SSD in each. I'm currently running them in a Ceph cluster but I'll be tearing that down soon. I could run some benchmarks with three in one box and report back? Would be a one-time
6.
by zenoprax 2mo ago
Wow, I had no idea. This is probably going to do the trick nicely. I really appreciate the suggestion and link!
7.
by zenoprax 2mo ago
I've been trying to find a solution for this too! I was considering using Rclone but too many things are using SQLite for me to trust rsync. I was also going to go with CoreOS but I'm leaning towards Fedora Cloud now in case I nee
8.
by zenoprax 2mo ago
Thanks for the second link - looks interesting. I have an old and loud 16-bay server that boots every 7 days or so if no one is home, `syncoid there here`, and shuts off. I'll get pinged by uptimerobot once in a while if it's over
9.
by zenoprax 2mo ago
Have you tuned your Retro Encabulator recently?
10.
by zenoprax 2mo ago
The samples are striking in their simplicity compared to the over-controlled system prompts we have now. For example, the intended format is as follows: "GPT‑2 generates synthetic text samples in response to the model being primed with
11.
by zenoprax 2mo ago
> Both labs even explicitly promise the customer owns the outputs. > to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output If the argument is that the model itself is under cop
12.
by zenoprax 2mo ago
Can they even see the domain? Assuming you're not using the ISP's DNS of course.
13.
by zenoprax 2mo ago
> I'm just hoping he uses it as a tool and not a crutch. That's not enough given how much adults are struggling with this.
14.
by zenoprax 2mo ago
> The format does not make it irrevocable, it makes the revocation impossible to enforce. Fair point and I should have used less ambiguous terms. I'm also thinking about the "Permitted for Home Use Only" nature of DVDs. Pl
15.
by zenoprax 2mo ago
> pirating a copy of the content is fine because others can/do still "buy" or "own" (rent/lease) it. Piracy is not theft. It is an infringement on another's right to distribute intellectual property. Th
16.
by zenoprax 2mo ago
9070XT operator here: I'm using llama.cpp with the same model and quant and I'm getting 87,000 for my context limit. I tried the Unsloth models but they lowered it to around 30-40K so I went back to upstream. I'm on Linux and
17.
by zenoprax 2mo ago
Beware of interactions with git and syncthing. It's fine for straightforward repos where all you do is commit but as soon as you start doing more complicated branching and re-basing you will start to generate lots of `sync-conflict` fi
18.
by zenoprax 2mo ago
I don't think you had bad intentions. You may know the testing is more involved but there is a common perception that people can just do a single online test and get a diagnosis. People should be informed that it's just one indi
19.
by zenoprax 2mo ago
> it's a serious dysfunction and it's sad to see that it's being romanticized lately in social media or plainly discredited by some people/media The author of the article is contributing to this by using their medical
20.
by zenoprax 2mo ago
I haven't tried to join any in the last 2-3 years but I can see changes within some of the more progressive private trackers I can access. Hopefully the trend continues.
21.
by zenoprax 2mo ago
Unlike public trackers, private trackers are largely filled with seeders with very few leechers. If 50 people have a file and one person goes to download you are only getting a tiny slice of that at best and for small files there may not ev
22.
by zenoprax 2mo ago
This is the second time I've seen this project mentioned on HN. Their white paper is very interesting and aligns closely with my interests but I don't see much on the main site about how to get involved. Are they using their Mat
23.
by zenoprax 3mo ago
> There's only $50b in datacenter buildout nationally (Source: Gemini, 2026). Your reference lacks authority, veracity, and reproducibility.
24.
by zenoprax 3mo ago
Great link. The most relevant part for me is the last couple minutes (22:00): it's only against the law to destroy evidence if it can be established that such evidence exists. This feels like a more elaborate version of accidentally lo
25.
by zenoprax 3mo ago
Agreed. I think there is a much better story hiding under the surface but I wouldn't read a follow up because it's like Claude is holding them hostage for content.
26.
by zenoprax 3mo ago
I'll try it out, thanks. (Using an AMD 9070 XT)
27.
by zenoprax 3mo ago
Great, thanks! I'll try it out.
28.
by zenoprax 3mo ago
For some reason the unsloth models leave hardly any room for context. I've switched to the regular (non-unsloth) and get about 25 t/s and get about 80,000 more context tokens for the same quant.
29.
by zenoprax 3mo ago
I've seen the same thing. I tried the "superpowers" meta-harness and gave it a simple web app task and it spent 4 hours to make a basic timer app. I might try restricting the amount of thinking it is allowed to do to 500-1000
30.
by zenoprax 3mo ago
Sure, but this is true for all lossy compression (audio, images, etc.) Given 16GB of VRAM, what will give me the best experience in OpenCode? Currently using Qwen3.8_Q_3

More ›