regexorcist

4 ms·

61.
by regexorcist 5mo ago
> they sure look a lot more commercialized Same here. Just looking into going to Coachella or Burning Man seems like so much work. I'll take a summer festival in Spain, Germany or Netherlands instead.
62.
by regexorcist 5mo ago
I use oh-my-openagent. It does an incredible job at planning and executing by orchestrating subagents with different roles.
63.
by regexorcist 5mo ago
Sure, I have a 64G MBP with an M1 Ultra. The best model for me by far has been the 35B A3B, in particular the 8Q_KL unslouth variant. The dense model works but it's much slower, and I don't really see a difference in quality with
64.
by regexorcist 5mo ago
I've been saying this for a couple months now since I got decent hardware and started using my local Qwen 3.6 exclusively. I have no doubt the future for individuals and medium-sized companies is local private AI.
65.
by regexorcist 5mo ago
This has been pathetic to watch in general and first hand at the companies I've been at lately. Management scrambling to find anything to throw AI to, resulting without exception in embarrassing demos. I'm excited about AI but not
66.
by regexorcist 5mo ago
I became the school computer genius by teaching everyone to cheat in exams with Winpopup. My earliest memories though are also in DOS with an Olivetti computer.
67.
by regexorcist 5mo ago
No, fuck LaLiga. UCL always delivers for us fans and so does the world cup. It'll be extremely popular as always, for good reason.
68.
by regexorcist 5mo ago
Curious if you tested llama.cpp and still found oMLX faster? I haven't tried the latter myself, might give it a go.
69.
by regexorcist 5mo ago
Yeah it's crazy how they're burning developer goodwill. I've personally cancelled and resent them for not being able to delete my claude code session (that button was misteriously the only one in the UI to throw an error, I t
70.
by regexorcist 5mo ago
I simply can't go back to cloud AI. Privacy and full control are more important to me than speed and SOTA models.
71.
by regexorcist 5mo ago
You have a point but it's definitely not TBs for 1M. Should be more like 100G.
72.
by regexorcist 5mo ago
https://github.com/anthropics/knowledge-work-plugins/tree/ma...
73.
by regexorcist 5mo ago
No need to be. I'm rocking Asahi on a M1 and it's the absolute peak Linux desktop experience.
74.
by regexorcist 5mo ago
It's mostly marketing and hype. This "product" is a collection of vibecoded skills.
75.
by regexorcist 5mo ago
You don't mention runtime, hardware and harness which are critical. The 35B A3B model should be pretty fast, you do need a decent setup but nothing too fancy. I'm using Q8_XL from unslouth with llama.cpp and opencode and it's
76.
by regexorcist 5mo ago
> So many expect an M5 Max Mac Studio with 1TB/s+ bandwidth and specs up to 256GB or 512GB, probably for ~$10k later this year. This is what I'm really waiting for. It will enable models comparable to current SOTA at the enthus
77.
by regexorcist 5mo ago
There are many recent examples of market manipulation, but this isn't one of them. Digital sovereignty is being pushed in the EU too, and it's a good thing.
78.
by regexorcist 5mo ago
Try oh-my-openagent plan mode.
79.
by regexorcist 5mo ago
Yeah this is key, a lot of people are still just looking at the number of params and thinking these models are toys. What Qwen 3.6 has shown is that reasoning and tool calling are just as important if not more.
80.
by regexorcist 5mo ago
I'm very much pro-AI but I'd quit your tech team on the spot if I were asked to review those.
81.
by regexorcist 5mo ago
Sounds like a game changer if I see that kind of speed up on my hardware. So far I've prefered Qwen 3.6 because of its better tool handling, even though Gemma 4 is faster, but I saw they've updated the model template and that&#x27
82.
by regexorcist 5mo ago
How so? I find it much the same to other templating engines like Jinja, though I'm definitely not a fan of the syntax. But that hardly matters anymore with LLMs.
83.
by regexorcist 5mo ago
Projects like opencode are making the distinction between GUI and TUI almost meaningless. And that "only" downside is a massive, deal-breaking one. At this point I only have a browser besides the terminal, and I can see that going
84.
by regexorcist 5mo ago
No it can never be the same. The terminal is about not having to switch from the keyboard. My entire workflow is tmux panes with different TUIs and terminals. Not to mention performance, with a neovim IDE you may have tens of them open in d
85.
by regexorcist 5mo ago
That's a wild assumption and most certainly wrong. Open models will continue to evolve with or without Chinese labs.
86.
by regexorcist 6mo ago
Indirectly introduced me to programming as my dad was obsessed with it growing up. He was a hobbyist with zero formal training, working a completely unrelated job, and still made a ton of money selling tailor made admin software to all kind
87.
by regexorcist 6mo ago
I work on Go and Rust mostly. The experience can be wildly different based on model, quantization, and harness. The fact that it didn't work for you doesn't mean everyone is trying to hype local models, people are getting real wor
88.
by regexorcist 6mo ago
Opencode has been the best one for me so far.
89.
by regexorcist 6mo ago
I suggest to have opencode drive the model. I also use neovim and these days I mostly just have a tmux pane side by side. But opencode does support ACP mode which you can use with codecompanion and the like.
90.
by regexorcist 6mo ago
I think you'd be surprised, I find that the harness is what makes the real difference. I also prefer to be on the loop, actively guide and review. Local models are definitely much less autonomous as of today so if you need to be churni

More ›