pxc

34 ms·

481.
by pxc 1y ago
OpenCode isn't a fork of Claude Code. It's descended from an independent open-source project that was called "TermAI": https://news.ycombinator.com/item?id=44483338
482.
by pxc 1y ago
!! This seems really nice for macOS users! Less clunky than the "envfile" option Doom Emacs provides for sure. I also recognize the author's name because I use their direnv integration package all the time! That one is great,
483.
by pxc 1y ago
I am truly, to borrow a phrase from another commenter on this post, a "longtime Emacs dabbler". But I use Evil mode everywhere and sometimes use other Emacs bindings in Insert mode where they don't directly conflict with some
484.
by pxc 1y ago
> I'm curious what emacs users are doing these days. Integrating with a new language ecosystem is a significant amount of work for me because it involves making choices about what packages to configure and how, and which external de
485.
by pxc 1y ago
I found this surprising because that's such an old test that it must certainly be in the training data. I just tried to reproduce and I've been unable to get it (20B model, lowest "reasoning" budget) to fail that test (w
486.
by pxc 1y ago
It could never be a GNU license because it violates freedom 0 as outlined by the FSF: https://www.gnu.org/philosophy/free-sw.html
487.
by pxc 1y ago
> What makes these AIs so amazing is thinking of them as entirely replacing Google with something closer to a god I guess that way of thinking may foster amazement, but it doesn't seem very grounded in how these things work or their
488.
by pxc 1y ago
Is the DeepSeek model you're running a distill, or is it the 671B parameter model?
489.
by pxc 1y ago
> I am actually able to use ChatGPT 4.5 as not just an employee, not even just as a coworker, but at times as a mentor or senior advisor: I can tell it what I am trying to do, and it helps me by applying advanced mathematical insights or
490.
by pxc 1y ago
NVIDIA will probably give us nice, coding-focused fine-tunes of these models at some point, and those might compare more favorably against the smaller Qwen3 Coder.
491.
by pxc 1y ago
Right... knowledge is one of the things (the one thing?) that LLMs are really horrible at, and that goes double for models small enough to run on normal-ish consumer hardware. Shouldn't we prefer to have LLMs just search and summarize
492.
by pxc 1y ago
Qwen3 Coder is 4x its size! Grok 3 is over 22x its size! What does the resource usage look like for GLM 4.5 Air? Is that benchmark in FP16? GPT-OSS-120B will be using between 1/4 and 1/2 the VRAM that GLM-4.5 Air does, right? It s
493.
by pxc 1y ago
Huh? Do the community surveys indicate that or something? JS definitely isn't the language I think of first when I think of Nix users.
494.
by pxc 1y ago
The FSL, like the BuSL that Hashicorp Terraform is now under, is a proprietary software license. "Partially" makes sense for something where some component or subproject is released under a proprietary license and other components
495.
by pxc 1y ago
I like to have a new Emacs frame pop up for Org capture, but when the frame closes, focus is then given to Emacs instead of being restored to whatever I was using before I used the global hotkey to summon the new frame for the Org-capture b
496.
by pxc 1y ago
You might be able to use the extra time to have it do things like run some formatters, linters, run the code in a VM before you inspect it, or modify it for compliance with a style guide that you've written, and continually revise it f
497.
by pxc 1y ago
You can still get it pay-as-you-go on OpenRouter, afaict, and the billing section of the Cerebras Cloud account I just created has a section for Qwen3-Coder-480B as well.
498.
by pxc 1y ago
This feels way less annoying to use than ChatGPT. But I wonder how much the effect is lost when the tool does many of the things that make models like o3 useful (repeated web searches, running code in a sandbox, etc.). For code generation,
499.
by pxc 1y ago
I think it still works fine as a parable, and it doesn't hurt to know a little bit more about how the trees are Oxford are really kept.
500.
by pxc 1y ago
This notably marks a schism with the FSF; it's the first time and context in which "open-source" and "free software" have not been synonymous, coextensive terms. I think it greatly diminishes the value of the concep
501.
by pxc 1y ago
I use gitnr[1] to add fairly comprehensive `.gitignore` files to my repos at creation time. Ultimately, it's the responsibility of each developer to determine whether or not the files they've added to their local clone really belo
502.
by pxc 1y ago
Could this be in part because many of the recent Chinese models (which seem great, tbh) show signs of having been distilled from one or another Claude models? Or is that a silly idea, because distillation is unlikely to be stopped by rate l
503.
by pxc 1y ago
I feel like YouTube's enshittification is already here. The algorithm has long been terrible, they now punish users for disabling watch history, and the ads are more frequent, longer, and more annoying. If not for inertia (lots of vide
504.
by pxc 1y ago
When I can access something in the browser rather than only via native apps, I rely heavily on Dark Reader, Midnight Lizard, and Page Shadow. Love all three to death
505.
by pxc 1y ago
Dark theme is an accessibility issue for people with eye diseases like mine. If you both need high luminance contrast and have photophobia, dark themes are the only ones that are usable. There are workarounds, like inverting all the colors
506.
by pxc 1y ago
Clearly the thing we should want is a healthy, international AI ecosystem characterized both by cooperation and by competition, so that we are free to choose between models developed under different conditions, for compliance with different
507.
by pxc 1y ago
For me, the sense of a greater degree of independence and freedom is also important. Especially when the tech world is out of its mind with AI hype, it's difficult to feel the normal tinkerer's joy when I'm playing with some
508.
by pxc 1y ago
> I feel like I'm the exact opposite here (despite heavily mistrusting these models in general): if I came to the model to ask it a question, and it decides to do a Google search, it pisses me off as I not only could do that, I did
509.
by pxc 1y ago
What languages do you work in? How much code do you keep? Do you end up using it as scaffolding and rewriting it, it leaving most of it as is?
510.
by pxc 1y ago
Don't we also find with natural languages that focusing on training data from only a single language doesn't actually result in better writing in the target language, either?

More ›