HN user
Could be a good tradeoff for the flash model though. 3.5 -> 3.6 is a tiny bit cheaper and maybe faster?
artificialanalysis.ai has it going from 165 tps -> 304 tps. openrouter.ai needs more data but it has it going from ~100 tps -> ~150 tps, though at peak 3.5 has reached 156tps.
So.... What models have actually failed?
Where does the less than a third the price come from? From the provided benchmark suite, it came slightly under half.
I haven't notice any significant changes.
My Linux experience gets more polished every year. Last serious issue I remember is a AMD GPU kernel crash around 2022(?).
I feel like Android has always had its glitches. I can live with the current set, but each update has me nervous about the new set of bugs that will appear.
Seems like it https://www.phoronix.com/review/amd-ryzen-ai-halo/9
This is not surprising and perhaps one of the things I respect about it. Zig is pretty minimalist and I would prefer denying my pet features over becoming design by committee.
I'm a bit curious, what features could you not do without?
I think of Emacs as more of a lisp machine than text editor. I'm also not sure what an IDE has over Emacs. I have autocomplete, compiler checks, and run tests all within Emacs.
For general automation, my blog is built with Emacs Lisp.
If you don't pass an allocator param in Zig, then the function basically can't allocate.
I haven't tried Z.ai, but both Ollama ($20) and OpencodeGo ($10) seem to give me more generous limits than the Claude $20
I'm surprised this is the default. How good are the snapshot management tools?
I thought parentheses were fairly intuitive. They are not for grouping, more like representing an AST.
For other languages, similarly, x is not the same as x()
I made a rusty version of this. No linked lists, but that's not important in the grand Scheme of things.
haha, I definitely wouldn't call my setup fast. But it did get a lot better after Native Compilation and a lot of the major stuff has started async to feel more responsive at least.
No, but I use GLM5.1 instead of Claude/GPT.
I just started Compiling with Continuations by Andrew W. Appel.
I hope that by the end of the book I don't find out that CPS compiled code produces bad stack traces.
Which specific models were you using?
In March I switched to Opencode + Kimi K2.5 and found it was a step behind. I switched to GLM 5.1 and has felt like a step above. Its probably some combination of me forgetting the baseline, model improvements, and OpenCode improvements.
$20 a month has been good enough for my coding use cases. I wouldn't call myself a vibe coder. Stuff I do is create graphs/visualizations, review, polish code, generate toy examples for learning.
I've been fine with Linux/KDE. Even the fractional scaling support is decent.
Edit: Guess it depends on the app
Trump cannot instruct Google to put picture of dogs on their homepage.
I'm sorry, but that was a horrible example. Corporations have no obligation to donate money to the ballroom yet Google has donated millions.
Well, I wouldn't use this since I have my own box. In case its useful:
- I run hermes on the box and it has some scheduled cron jobs.
- I gave it an account on a custom Git forge. It cannot commit without my direct permission, though it can blow the setup up in other ways lol.
- I interact by assigning it issues and talking through Discord.
I started writing a Scheme interpreter about a year ago and got pretty far. I dropped it a few months back since I got a new job.
I'm thinking of getting back and am wondering if the niche (and difficult for me to implement) features are worth it. I might be better off skipping dynamic-unwind, maybe even ripping out call/cc, in favor working on the debugability, ecosystem, performance, and package management story.
Ironically, the top comment (this) is now about AI.
Zig gives you more control than Rust, which should theoretically lead to a higher performance ceiling.
There's not much magic in Zig. Keep hitting goto-definition and you can eventually see the OS switch statements and syscalls.
Try it if you want full control over every memory and IO operation and "drop". If you hit "goto-definition", you eventually get to see the OS switch statements and syscalls. There's not much magic.
Do not try it if you are scared of memory management and memory leaks.
We're talking about ~1 trillion $$$ valuations here tho
When I joined in 2016, it CitC would make it look (and still does) like you had the entire monorepo on your local filesystem on your machine.
Git5 would copy some directories but builds would still fallback to files from the monorepo if you didn't track them. It was convenient for me since I could just grep and do fuzzy matching from my editor. Now I have to do some extra work to avoid grepping the entire monorepo. LLMs sometimes still try to grep the entire repo lol.
Now, you could use a perforace, mercurial, or jj interface and it works fine.
From the user interface perspective though, it does essentially look like you've pulled all of google3 into your disk.
Luckily, they still support the text editor + CLI tools workflow so I can still use Emacs effectively.
We've moved on from Git5. Although it was a pain, I kind of liked that Git5 made the monorepo less monolithic to my editor.
Yeah, under my desk. I rarely remote which is a good excuse for me to disconnect from work anyways.
I use a workstation specifically to improve latency. Needed to get approval at some point to get a refresh though.