selfhoster11

GLM-5.3, Kimi K3. OpenRouter is good for figuring out which provider you want with the serving precision that you need.

I have at one point faced the “should I buy x86 instead”, and decided low-power Intel CPUs make it viable. Guess I’m gonna pick the x86.

Lol, no.

Tinfoil etc offer TEE inference that works well, directly or via API routers.

Why would you not put a CDN in front? I’d rather do that than CAPTCHA.

Would you mind opening it to the public? We’d be okay making modifications for our use cases, so dumping the code would be fine.

Interoperability clause in the EU might make it legal? Not sure.

Check out this’s recent HN submission: https://gist.github.com/Hellisotherpeople/71ba712f9f899adcb0...

That’s just how some humans write. These words didn’t come from an LLM vacuum.

Difficult texture, but I appreciate the disclosure of AI use and found some of the points interesting enough to write my own essays on.

Moonshot.

Under Linux it is allegedly 110GB, but I’m not sure.

Executive dysfunction mitigation. Voice based interfaces. Heavyweight personal file classification with a few hours of prompt building vs labelling a bespoke classifier’s data set and training a more “lightweight” option in weeks or days. Language translation that isn’t DeepL or Google Translate for random websites. They are not deterministic, but the error rate is a lot better on these tasks vs classical approaches.

That's like saying we shouldn't be letting people access glasses or wheelchairs because we need a way to fix the issue at the root. Nope.

I use both of these and I don't think I've once had an LLM rewrite my comment before posting it on HN.

They are trained on natural language. Not anthropomorphizing them is the worse end of the spectrum.

LLMs are indeed displacing labour. Junior IT roles are drying up in places. Translation and art are also becoming harder to earn from.

They always have a choice, it just doesn't make them as much.

There was Lavabit, though that's an example of just one such event.

Edit: and to some extent Apple, at least in the past.

I am on this very forum as an explicit effort to counterbalance that very view. You have my strong agreement.

I didn't even realize that this is (allegedly) written with AI. If it's AI, then it's the kind of AI writing that's closer to the real deal and that I want to see more widely applied if AI is to be used.

AI slop

No. I reject this framing. It is none of anybody's business how "secure" my device ever is. A smartphone is a piece of electronics, and not a tamper evident identity device.

Which is why that's not what it does. It asks you to input the hostname instead, just like deleting a repo in Github does.

The good rephrasing will not include that voice.

I refuse to engage in "LLMs are evil, period" views. That's like walking out into a battlefield with a samurai sword, while your enemy has Gatling guns. You'll be shredded. The pressure to survive means new tools have to be examined and incorporated as and when needed. The resources needed to run a 24B LLM on a gaming GPU are not costing the earth.

K2.5 and GLM-4.7/-5 were good in my experience, another vote for those.

I could spend 100% doing the work with my own Claude, and 0% reading yours. That's a negative-sum outcome. I do think that the 80%/20% split is better (though anything that is mostly human voice is fine for me).

Red Star OS.

Please add Internet Archive's bot to your auto-allows, at least. Their bot is presumably well behaved, and for public benefit.