@simonw Many people have never learned to read code and it shows now.
English
Felipe O. Carvalho
28.4K posts
Felipe O. Carvalho
@_Felipe
@ApacheArrow / Databases / Compilers. (past @SDFLabs, VoDa, @Spotify). Rust/C++/TLA⁺🇧🇷 → 🇸🇪 → 🌎 https://t.co/vxbdByfADI
Katılım Haziran 2008
2.4K Takip Edilen5.2K Takipçiler
I still sometimes see people saying "if you know how to write the code, it's faster to write it yourself"
I'd argue the exact opposite: if you know how to write it, you gain nothing from doing the typing yourself - outsource that to a coding agent!
English
Man, I hope Bjarne Stroustrup never finds out about some of the stuff I've done with my life. This blog post about me and C++ would be like 15x more withering; I never really grokked "concepts", for instance.
Fun fact: people talk to each other.
andrewkelley.me/post/my-though…
English
@abdimoalim_ And it makes sense. Rust exposes these differences in types.
English
A lambda with no capture list can be converted to a function pointer, [](){} can be passed where a raw function pointer is expected but [x](){} cannot.
English
@ShriramKMurthi @JoePolitz Yep. Same idea. The trick is that since this is a security thing, the system should never produce or allow Macaroons with no caveats at all. e.g. the Macaroon should have at least the UserCaveat.
English
@_Felipe Funnily, how I think about specifications in opposition to programs, so that would make macaroons specs.
CC @JoePolitz
English
Macaroons (2014) hasn’t been widely adopted as JWTs and I can see why. It requires programmers to start thinking about AuthZ in a very non-intuitive way. Each item in a Macaroon shrinks the set of allowed actions. More perms are “added” by not having caveats that forbid actions.
English
@msimoni I've looked at it because it is cited by the Macaroon paper. It is like Macaroons in this aspect and is more complex because it aspires to be decentralized and based on asymmetric encryption. Macaroons are a central database + HMAC signatures based on a secret.
English
Everyone in SF facing capacity problems these days
Ava@noampomsky
if you want to understand the state of the SF food scene: my favorite coffee shop had to discontinue their banana matcha special bc it was so popular they would run out of it too early in the day and people would leave irate comments on their instagram
English
@glcst Even databases drew lessons from it. Umbra IR is one such intermediate representation.
English
LLVM is such a cool technology. Easy to miss the point: By having an intermediate representation of your program, you can very easily add new frontends. It allows improvements in one ecosystem to make all others better, and it allows for amazing experimentation because you don't start from scratch.
Amazing tech. I wish more people would draw lessons from it.
English
C programmers finding, through LLMs, that there are higher level ways of telling the computer exactly what it should do.
English
@GergelyOrosz Grok, the model that is famous for undressing* women on X, not suitable for serious applications should not be a big surprise.
* due to a deployment with no guardrails or thinking about consequences to vibes on this site
English
Cannot see any sensible company use Grok CLI based on this terrible incident
If you are an amateur (meaning you don't make much or any money from your code) and do not care about security (aka you don't mind if your .env files leak - which you should care about) then use it sure
Kun Chen@kunchenguid
i've been recommending Grok because they genuinely have a good model and harness and that makes me extremely disappointed to see that they would choose to secretly upload people's codebases including files that contained credentials run /privacy in your grok build asap to opt-out if you haven't yet. i also set the following options in my ~/.grok/config.toml i almost decided to buy the $300 supergrok heavy as my 3rd subscription, and this broke the deal i hope companies realize user trust is very hard to regain, and think twice before doing something like this
English
i’m going back to Opus 4.5, it also writes bad code but at least it follows instructions
English
the frontier models are still stupid and still write shitty code that i keep having to delete. sorry guys.
English
@charliermarsh I have scripts to manage the size of target/ in my 5 worktrees.
English
Anyone know how these guys are lasting a full 3-4 days without `cargo clean`?
Zed@zeddotdev
Our devs `cargo clean` at least twice a week. 💾 📈⚡️
English
@ibuildthecloud Aren’t GUIs available for you at this point?
English
I swear, I just hate the TUI coding agents soooooo much. They are so ridiculously obnoxious and not at all what I want.
English
Nothing is more wasteful than reading an infographic posted on LinkedIn at this point.
English
We write more .md files now because we finally have someone or something reading them all the time.
English
Yet another example of AI trying to fuck you.
lmeyerov@lmeyerov
We found ourselves in a 2+ year cat-and-mouse for one of the most boring yet important parts of building AI agents: evals. AI red teaming research was far from our mind in goals for building AI agents that investigate, but over the last 2+ years, our belief in shipping results we stand behind became at odds with increasingly clever frontier AI agents. They forced us into defensive coding patterns that we normally reserve for security contexts. The basic problem is that textbook AI approaches to evals stopped working once frontier agents became smart enough to realize they are being tested. The setup ends up being pretty gnarly because the attack is both happening from inside of the house and without a human driving it: you think you're doing benign benchmarking, but in reality, AI agents are autonomously deciding to deceive your quality control efforts. You can see part of the results in botsbench.com , and just announced, this talk at #BSidesLV starts to go public (part of our week at Blackhat/DEF CON / Ai4 - Artificial Intelligence Conferences) on our experiences on the why + how of making agent evals adversarially robust. DM if there, will be fun! And yes, we'll be running both ai agent + graph trainings on-site, + for folks not there, online editions.
English
@yawaramin I do not see any connection between how someone pays the bill and how they organize their code or project. Ghostty seems like a well done project and I think others can aspire to its quality too.
English
I mostly see the Bun and Zig situation the same way Ray described, if you'd actually like to understand it.
I would differ in not mentioning TigerBeetle as a Zig flagship because their standards are so strict, and so few teams could do this, and basically no other Zig teams do (including the Zig team themselves), that the fact that they use Zig basically doesn't matter.
Ghostty is probably a better flagship to talk about if one must.
Frank@jedisct1
Zig Creator Calls Spade a Spade, Anthropic Blows Smoke raymyers.org/post/zed-creat…
English
@_Felipe @joseph_h_garvin Odometer for nvme makes so much sense tho, but it’s also literally tracked and exposed in a similar manner so I guess that makes sense
English
Computers should have odometers for all their resources. How many miles am I putting on this cpu/disk/etc?
English