arcanemachiner

6 ms·

1.
by arcanemachiner 5d ago
> The Computer is no longer Personal. I expect my personal computer to only do things that I choose to invoke. The internet and its consequences have been a disaster for the personal computer user.
2.
by arcanemachiner 5d ago
Non-macOS user here. Does macOS not let you run things as root?
3.
by arcanemachiner 5d ago
Dollars to donuts, they are speculating, and not privy to inside information on the topic. However, I believe that runtime model quantization is possible with some publicly-available inference engines (e.g. vLLM), so its not beyond belief t
4.
by arcanemachiner 5d ago
The only revenue model I for this is ads (like AI Stupid Level[0]). Or as a loss leader to get eyeballs to your service (like Margin Lab[1]). EDIT: I forgot (and am shocked) that HN still doesn't seem to support Markdown-style links. [
5.
by arcanemachiner 6d ago
I think the answer is not to debug the code, but, when possible, to debug the outputs. The code may be considered to be a black box much of the time. (This is much more true for my hobby projects than my work projects.)
6.
by arcanemachiner 7d ago
government != population
7.
by arcanemachiner 7d ago
> Android wouldn't be what it is if it wasn't open source. Yes, and now that they've achieved market saturation, it's time to pull up the ladder. Thanks for all the free work, suckers!
8.
by arcanemachiner 7d ago
No thanks, I'm not tuning a bunch of files just for things to break when I switch models or a new one comes out. I'll just use my one-size-fits-all AGENTS.md file and tweak it when the one of the clankers screw up. I don't ha
9.
by arcanemachiner 8d ago
Very solipsistic take. Most people don't live in your bubble.
10.
by arcanemachiner 8d ago
Good guess. I'm usually pretty good with acronyms, but this one's got me stumped.
11.
by arcanemachiner 8d ago
I've been pushing the context window well into the 400-600k+ token range lately (mostly Opus 5). I prefer not to since I'm aware of context collapse, but I've been leaning that way lately. Lately, I've been finding that
12.
by arcanemachiner 8d ago
I assume the data is laundered into a format that qualifies as no longer being the "user's" data, then trained on.
13.
by arcanemachiner 9d ago
DeepSWE is saturated now IMO, and is basically worthless. Lots of new models get around 74%. Shame too, because it was a pretty decent benchmark for a few months there.
14.
by arcanemachiner 9d ago
HN definitely doesn't give a crap about that word.
15.
by arcanemachiner 9d ago
Dude I am in slop fucking hell right now. There is still room for a human touch, without which the agents will lever us harder and faster into a world of incomprehensible garbage.
16.
by arcanemachiner 12d ago
If it does that, you should hope it can also stand in bread lines for you too.
17.
by arcanemachiner 13d ago
You don't need human input. Any coherent input will do the trick. You can, for example, hook it up to a logging system and have it fix errors as they occur on your platform.
18.
by arcanemachiner 14d ago
Banks create money when issuing a loan. This is how fractional reserve banking works. They lend money they don't have (most of). This is institutionalized fraud, and it's been standard operating procedure for centuries.
19.
by arcanemachiner 15d ago
You can chain a lot more commands together with this technique than with a single Edit tool call.
20.
by arcanemachiner 15d ago
They're talking about writing a file with a harness-native Edit tool. They're saying the agents aren't doing that, but are using ad-hoc methods of writing the files. (My agents seem to prefer see these days.)
21.
by arcanemachiner 15d ago
> you’d want to be using GLM 5.3 Flash right now for most things agentic That was yesterday. I think the crown currently belongs to DeepSeek Flash v4.1 for the next few days or weeks.
22.
by arcanemachiner 17d ago
Is "normies" finally a mainstream word? Just wondering how out of the loop I am here.
23.
by arcanemachiner 19d ago
Maybe they should rebrand again. It took a few renames to get the stink off of Blackwater (private military contractor).
24.
by arcanemachiner 20d ago
"Show me the incentive, and I will show you the outcome."
25.
by arcanemachiner 22d ago
Must be system prompts and tool instructions guiding the agents differently.
26.
by arcanemachiner 24d ago
You can always use JSONC or JSON5.
27.
by arcanemachiner 24d ago
You have tried Qwen 3.8 27B before coming to this conclusion, I hope? It's an incremental improvement over 3.6, but I mostly want to make sure you didn't just try running some old junker before coming to this conclusion.
28.
by arcanemachiner 25d ago
How would this scam work? Why would someone pose as a locksmith specifically?
29.
by arcanemachiner 26d ago
So was ChatGPT back in 2021.
30.
by arcanemachiner 26d ago
If it makes you feel any better, I have just upvoted every submission made by your account.

More ›