selfhoster11

5 ms·

31.
by selfhoster11 8mo ago
It doesn't take much hardware. I have run larger models.
32.
by selfhoster11 8mo ago
I have seen such drives. ATAboy promises to be a nice option if manufacturing and populating a PCB is not too expensive.
33.
by selfhoster11 8mo ago
It's far better than everyone has nukes than just a few people who are highly interested in ruining your mind and/or finances. Governments, crime syndicates can pay for HHH-less AI.
34.
by selfhoster11 8mo ago
Counterpoint: B corporations. It's clearly possible for companies to self-impose safeguards: ESG/DEI, Bcorp, choosing to open source, and so on. If investors squeal, find better investors or tell them to put up with it. You can ma
35.
by selfhoster11 8mo ago
I don't think the parent comment was saying it can be solved, only that the LLM paradigm is better at dealing with complexity. I agree that they are not great at it yet, but I've seen vast improvements in the past 3 months alone.
36.
by selfhoster11 8mo ago
GeoIP this nonsense. Legal liability is solved as a "good-faith effort" and those living in jurisdictions where this doesn't apply (or use a VPN) don't need to be stripped of privacy.
37.
by selfhoster11 8mo ago
Murder is immoral. That much is certain. ICE murders people.
38.
by selfhoster11 8mo ago
FWIW, you aren't alone man. Stay strong.
39.
by selfhoster11 8mo ago
Except the parent comment said you can stream the weights from an SSD. The full weights, uncompressed. It takes a little longer (a lot longer), but the model at least works without lossy pre-processing.
40.
by selfhoster11 9mo ago
Try doing this as a normies without technical guidance. Technically correct, this time, is not the benchmark.
41.
by selfhoster11 9mo ago
OpenAI does KYC. I refuse to deal with that.
42.
by selfhoster11 9mo ago
I disagree. Music needs to be archived just as much as books and other publications. Yes, there is heat, but lots of people support them.
43.
by selfhoster11 9mo ago
Good thing that there are some people to counterbalance that, then.
44.
by selfhoster11 9mo ago
Not the point. The point is that nobody who doesn't know him knows whether he can be trusted, and the ISO itself might be prepared on a compromised machine unintentionally. This is very much a supply chain issue.
45.
by selfhoster11 9mo ago
The users.
46.
by selfhoster11 9mo ago
They already are paying off. The nature of LLMs means that they will require expensive, fast hardware that's a large capex.
47.
by selfhoster11 9mo ago
So only those with money should be allowed to play with tools? If you have no income, are unbanked, or are in an area with low earnings, that compiler is out of reach. And then you're back to piracy.
48.
by selfhoster11 9mo ago
You can do that yourself with opencode and a $10-15 subscription to a good model. I used to be a skeptic, but this month the pairing of tooling and models I discovered is finally good enough to build a static Go app with a TypeScript front-
49.
by selfhoster11 9mo ago
Heads up that this is "more true" for non-reasoning LLMs. Reasoning gives an LLM a lot more runway to respond to out of distribution inputs by devoting more compute on understanding the code it's changing, and to play with id
50.
by selfhoster11 9mo ago
1. You can run decent local AI now - see /r/LocalLlama. You pay the electricity cost and hardware capex (which isn't that expensive for smaller models). 2. Chinese APIs like Moonshot and DeepSeek have extremely cheap pricing,
51.
by selfhoster11 9mo ago
Gemma 3 27B, some smaller models in the 8-16B size range, and up to 32B can be run on hardware that fits in the "consumer" bracket. RAM is more expensive now, but most people can afford a machine with 32GB and maybe a small graphi
52.
by selfhoster11 9mo ago
Far less than you'd think for local LLMs.
53.
by selfhoster11 9mo ago
Kimi K2 Thinking. Tell it in the system prompt to push back.
54.
by selfhoster11 9mo ago
It's not that hard. Stationery shops do exist, even online if you don't have one locally.
55.
by selfhoster11 9mo ago
Better this than 300th React.JS bloatware of the year.
56.
by selfhoster11 10mo ago
You may want to use the new "derestricted" variants of gpt-oss. While the ostensible goal of these variants is to de-censor them, it ends up removing the models' obsession with policy and wasting thinking tokens that could be
57.
by selfhoster11 10mo ago
I mean, yes. Very much so. People should be upset about a relatively affordable hobby getting to this point.
58.
by selfhoster11 10mo ago
In the US, yes.
59.
by selfhoster11 10mo ago
Ideally, yes.
60.
by selfhoster11 10mo ago
Historical preservation, retro computing, period-correct hardware setups.

More ›