anuramat

8 ms·

1. 211.▲
   by anuramat 1y ago

   I suppose things that cause most of the problems don't exist in vi, eg (24 bit) color, image protocols or $TERM dependent features
2. 212.▲
   by anuramat 1y ago

   Well yeah, I'm trying to guess what you're implying; are you just saying "we need better tech"?
3. 213.▲
   by anuramat 1y ago

   "tmux capture-pane" strips escape sequences that break the terminal and sets all the right variables; try using expect with e.g. neovim
4. 214.▲
   by anuramat 1y ago

   So, researchers are insanely lazy/secretly against AI/controlled by the Big Data?
5. 215.▲
   by anuramat 1y ago

   If language is a dead-end application of language models, I don't know what isn't; the tooling is architecture agnostic anyway > after transistors were invented But we don't have "transistors" yet, what's
6. 216.▲
   by anuramat 1y ago

   you can specify the model in the yaml frontmatter, so at least with litellm it should be possible
7. 217.▲
   by anuramat 1y ago

   I also don't see how this requires heuristics, but usecases do exist; eg I set `CLAUDE`, so that a git hook can mark vibe commits -- a prompt would be a waste of tokens and would introduce non-determinism, and MCP is yet another depend
8. 218.▲
   by anuramat 1y ago

   I'd say it's just not a part of the machine-that-is-learning; I don't consider myself to be symbolic, even though I also use python
9. 219.▲
   by anuramat 1y ago

   > stored differently maybe it could be the extreme case of state-dependent memory? < https://en.m.wikipedia.org/wiki/State-dependent_memory >
10. 220.▲
    by anuramat 1y ago

    I thought the entire point of vscode was to be an extensible "lightweight" barebones code editor, as opposed to eg jetbrains stuff; what about vim/emacs then?
11. 221.▲
    by anuramat 1y ago

    you can use your claude code subscription with opencode (not sure how "legal" that is); github copilot straight up provides an api (very "legal")
12. 222.▲
    by anuramat 1y ago

    I can highly recommend bubblewrap for sandboxing -- configuration is just CLI arguments, one per a mounted path; I use a little script that mounts CWD, claude stuff, and XDG data directories in RW, and the rest in RO. To me it seems like th
13. 223.▲
    by anuramat 1y ago

    What's your point, we should all wait until an actual genocide is a possibility, before we even acknowledge that a literal purpose-built nazi AI is concerning ? Even if it had nothing autonomous to it, how can "programmatic antis
14. 224.▲
    by anuramat 1y ago

    I can't imagine interacting with computers without fzf, zoxide and nix anymore; everything else feels relatively inconsequential Wondering if there are any other life-changing tools I'm missing out on...
15. 225.▲
    by anuramat 1y ago

    anthropic provides a custom devcontainer for sandboxing, but I have fallen in love with bubblewrap - it's a single command, and I get to keep all the infrastructure: e.g. it can do nix flakes without duplicating every derivation
16. 226.▲
    by anuramat 2y ago

    > not the defining characteristic Did anyone claim otherwise? Besides, I imagine bridges aren't rebuilt every week -- poor blueprints can only cause a finite amount of pain.
17. 227.▲
    by anuramat 2y ago

    I think you're looking for federated/distributed learning. From what I understand, nobody quite figured it out yet. Relevant 2021 NIPS paper: < https://arxiv.org/abs/2106.10207 > On difference between fed