bellowsgulch

5 ms·

31.
by bellowsgulch 2mo ago
It seems to me that Qwen3.6-35B-A3B is still the local LLM leader, because 27B in either of the latest releases is just too slow to be usable compared to OpenCode Zen free models, or OpenRouter free models. Sad a Qwen3.8-35B-A3B model wasn&
32.
by bellowsgulch 2mo ago
Uh... brew install llama.cpp llama serve -hf unsloth/Qwen3.6-35B-A3B-GGUF:UD-Q4_K_M open http://127.0.0.1:8080
33.
by bellowsgulch 2mo ago
Awesome. Thanks for the heads up.
34.
by bellowsgulch 2mo ago
Are people still using deepseek-v4-flash everywhere? I found after the price increases, mimo-v2.5 seems far more attractive.
35.
by bellowsgulch 2mo ago
I've read a few of these over the years, and none of them seem to be useful. I have three sentences in my custom instructions, and those are basically all useless, too. Even my second one, "Avoid decorative or section-header comme
36.
by bellowsgulch 2mo ago
Reads like: brilliant Carnegie Mellon University computer science grad struggles to find job where he is not replaced by cheap, inferior Indian labor that still gets the job done, even if it takes marginally longer. No shit we're all p
37.
by bellowsgulch 2mo ago
Does anyone remember when you posted on a place like Something Awful or Facepunch getting banned for a day for saying something so fucking stupid you needed to learn a lesson? No? Anyone?
38.
by bellowsgulch 2mo ago
Honestly, I wish other language developers would chime in, too. It's insightful.
39.
by bellowsgulch 2mo ago
Who cares what you think?
40.
by bellowsgulch 2mo ago
I agree. It's actually quite insulting to other countries whose actual atrocities were measured in lives lost. I voted for Kamala but Trump Derangement Syndrome is real. He's just a loser, he's not even remotely brave enough
41.
by bellowsgulch 2mo ago
Growing up with PageRank search felt so, so much different than whatever the hell we get with Google now, which is almost certainly overtuned to some crap metrics that some software engineers are getting paid good for and getting promoted o
42.
by bellowsgulch 2mo ago
GPT‑Realtime‑2 is really weird. Perhaps just because it's bidirectional and now has the failure mode as a possibility, it responds too soon with filler at awkward times, and it's generally overeager. I feel like there was plenty o
43.
by bellowsgulch 2mo ago
poob has it for you
44.
by bellowsgulch 2mo ago
> Pity you had to ruin it by telling us you're exceptionally thoughtful and profound, and that those who disagree, suck! It's just a comment on the Internet, you don't have to take it so seriously.
45.
by bellowsgulch 2mo ago
I've been using OpenCode Zen and OpenCode Go for a while now, and I've been a bit disappointed with how they communicate to customers. There's no status page that I'm aware of, and at least with my private harness, conne
46.
by bellowsgulch 2mo ago
Futurama did it first!
47.
by bellowsgulch 2mo ago
The older I get, the less I want to use new browser features that have already been somehow possible with existing ones that typically have been around for literal decades. Adopting them makes a future with multiple browser engines more lab
48.
by bellowsgulch 2mo ago
They successfully did it, basically, if you read about Xbox's architecture. It's a really interesting look into how we basically wouldn't own our Windows PCs if they used that tech in their mainstream software and hardware.
49.
by bellowsgulch 2mo ago
OpenCode Go is too good to be true, and I suspect it won't last, like GitHub Copilot.
50.
by bellowsgulch 2mo ago
I have no idea why anyone would spend the money Anthropic asks for Claude models these days. There was a brief window in time when it was worth the money to use them a bit through GitHub Copilot, but there's no way in hell I'm spe
51.
by bellowsgulch 2mo ago
I want them to improve the product meaningfully beyond lane keeping, which besides lane changing (partially), it's the only thing it's capable of because they willingly choose not to extend the system to use more visibility.
52.
by bellowsgulch 2mo ago
I thought I would add a comment asking the maintainers to consider not shipping major version zero, since it basically makes semantic versioning worthless and using SemVer and ignoring this advice basically suggests one wasn't reading
53.
by bellowsgulch 2mo ago
This is definitely such a cool feature that I wish cloud providers would expose.
54.
by bellowsgulch 2mo ago
Two things: 1. they have a cheap fuse (cents on the dollar) that’s not automotive-grade so your windshield mounted device fails in hot weather and doesn’t last past the hardware warranty. Known issue, doubt they ever fixed it. 2. 1B paramet
55.
by bellowsgulch 2mo ago
I've tried this, too, and for whatever reason, simple things like having a subagent with vision capabilities hallucinates responses to the blind models. I don't know why, and it's not consistent, but I don't see this hap
56.
by bellowsgulch 2mo ago
The existing ones definitely are, so you know they're bullshitting.
57.
by bellowsgulch 2mo ago
Counterpoint: something has happened in frontier models, and yes they now get discouraged and will sometimes prefer to not continue working on a problem unless you tell them to anyway. I don't know how or why this would be trained on b
58.
by bellowsgulch 2mo ago
Framework, Bevy is definitely not an engine. Does it do one thing? It's probably a library. Is it a collection of libraries that work together focusing on primitives with callbacks? It's probably a framework. Does it load levels?
59.
by bellowsgulch 2mo ago
Fuck Apple.
60.
by bellowsgulch 2mo ago
Yeah, it's really weird watching everyone complain about Claude Code and continue to give Anthropic what amounts, in most cases, to 20 years of my current spend in 1 year, while I'm using a private harness and Pareto frontier mode

More ›