azrulite

@somewheresy So many approaches in theory should be better with LLM's. The trouble is that they weren't trained on those approaches, so they get nerfed hard.

is anyone else always lol at the fact that we spent YEARS trying to figure out hybrid search, document embeddings, "RAG" and graph databases, just for the models to improve enough to wield tools against a filesystem, and the SOTA achieved by Doing Literally None Of That

@ShriramKMurthi I feel the author has a good point, but for some reason shoehorned this nonsense analogy in. It is poor writing, but the essence of their message is good.

What a dumb premise. It’s fine to think it, provided you can also refute it for yourself in 10 seconds or fewer.

@ayushagarwal027 Most the references are GitHub repositories to dependencies of the Rust implementation, or some STM32 documentation sites. This should be a 1000 word blogpost with a few pics, not a paper with the title "Rust or C for embedded". I would be mortified if my name was on this paper.

@ayushagarwal027 This is a total nonsense case study. Garbage methodology. It is about as worthless as making two AI's write something and then analyzing and comparing their results. May be even worse, than that.

🦀 Rust vs C for embedded firmware : researchers finally put it to the test. Two teams. Same functionality. One wrote it in C. One in Rust. They ran side by side for months on real industrial IoT hardware. The result? 👇 No strong reason to prefer C over Rust, not on memory footprint, not on execution speed. And Rust's runtime (Ariel OS) actually came out smaller than the traditional bare-metal C stack. The "C is faster for embedded" argument just got a lot harder to make. ⚡ 🔗 arxiv.org/abs/2604.25679 #Rust #RustLang #EmbeddedSystems #IoT #SystemsProgramming #C

@IceSolst But I see what you are saying. If you have no prior understanding of a codebase, you would start from a neutral position when reviewing changes.

@IceSolst Thank god I don't have the job of reviewing random PR's over a bunch of random github projects, that would sucks.

I distrust human and LLM generated code equally I similarly distrust human and LLM conducted code reviews equally And our tools to verify program correctness need a lot of improvement

@cloneofsimo I feel like I'm starting to see less of both. More and more people are starting to consolidate into some middle.

Ai discourse used to be either 1. LLM is fucking nothingburger bubble, its fucking useless 2. Ai will kill all jobs and we will starve in hyper inequality And uh... now I see less of 1. Infact, i cant remember last time people took the former opinion seriously (apart from Ga*y ma*kus)

@zeeg "lets ignore all the horrible UI bloat of the last 20 years and build some simple shit that runs anywhere" I don't know what purpose the quote has there other than restating the opinion.

TUIs are not good sorry yall a CLI is a utility, and situational. this should not be confused with stuffing a full interactive GUI into a low capability platform. "lets ignore all the great UI technology of the last 20 years and build some caveman shit"

@cmuratori Seems like a clever way of calling the person he is replying to a braindead vegetable.

How did we get here?

@pistondonda @mil000 The problem is, these models while they can benchmaxx coding and some other stuff, they are incredibly stupid. Just look at how poorly the flaghsip open weight models do on arc-agi 2.

@mil000 at some point behind will mean less and less if the pricing doesn't chart the same way. ~80% as good for 10% of the price is a good business model. especially if you outsource most of the overhead for free.

@ThePrimeagen Stop learning how to read and write. Start learning to think. Machines write and read now. They'll never think like you.

wut

@_winter_wonders company has a bunch of webdevs >oh shit guys the competitor has this thing we need to make it asap this is the result

can somebody anybody tell me what in the fuck the point of a TUI written in JavaScript is

@scheminglunatic no jop for haskeller i dont write the law

I AM COMPUTE POOR! AND REGULAR POOR! GIVE ME A JOP!

@vxunderground @LundukeJournal @gf_256 He's implying there's a conflict of interest. A trans person cannot audit a Rust codebase and greenlight it, becuse they would be lenient towards that codebase, because it aligns with their activism. Not saying I agree with it, but it's pretty clear what he's trying to say.

I don't understand. You're critical of Gf256's lifestyle (trans?). I thought your point was that because they're trans, they're incapable of performing the role assigned to them, or something. Otherwise, if that is not the case, then I don't understand the point of this post. I'm confused

Remember the security firm that Ubuntu hired to audit the (ill-advised, highly buggy) Rust-rewrites of all of the GNU Coreutils? Turns out that security firm is run by @gf_256, who: - Appears to be a man who thinks he's a woman ("trans"). - Uses an anime cartoon of a girl as his avatar. - Appears to have an OnlyFans page. I repeat: Ubuntu hired a "Trans" man, with an anime girl avatar and an OnlyFans page... to audit Rust code. It's hard to get more on-the-nose than that.

@zuess05 1) they're not 2) they are, but some are having regressions back to junior level due to laziness

I am actually curious. If junior devs are using Claude to write code at a senior level... And senior devs are using Claude to write code at a senior level... What exactly is the difference between the two roles right now, other than a $100k gap in salary?

@OpenAI What happened to this company?

@OpenAI Hmm.. Revenue..

One week since the launch of GPT-5.5, and it’s already our strongest model launch yet. API revenue is growing more than 2x faster than any prior release, while Codex doubled revenue in under seven days as enterprise demand for agentic coding tools keeps climbing.

@FateOfMuffins @AcerFur How many actions? Did you help provide them some context?

@AcerFur I don't really have faith in the official leaderboard when I've had GPT 5.4 solve some levels that they claimed 5.4 was running around blind like a headless chicken... by pasting screenshots into ChatGPT instead of the json.

@dwarkesh_sp @reinerpope

.@reinerpope works out from first principles how much frontier models are overtrained relative to Chinchilla optimal. One of the cleverest deductions from the lecture, and the one I enjoyed the most.